doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pntd.0002994 | Characterization of Aedes aegypti Innate-Immune Pathways that Limit Chikungunya Virus Replication | Replication of arboviruses in their arthropod vectors is controlled by innate immune responses. The RNA sequence-specific break down mechanism, RNA interference (RNAi), has been shown to be an important innate antiviral response in mosquitoes. In addition, immune signaling pathways have been reported to mediate arbovirus infections in mosquitoes; namely the JAK/STAT, immune deficiency (IMD) and Toll pathways. Very little is known about these pathways in response to chikungunya virus (CHIKV) infection, a mosquito-borne alphavirus (Togaviridae) transmitted by aedine species to humans resulting in a febrile and arthralgic disease. In this study, the contribution of several innate immune responses to control CHIKV replication was investigated. In vitro experiments identified the RNAi pathway as a key antiviral pathway. CHIKV was shown to repress the activity of the Toll signaling pathway in vitro but neither JAK/STAT, IMD nor Toll pathways were found to mediate antiviral activities. In vivo data further confirmed our in vitro identification of the vital role of RNAi in antiviral defence. Taken together these results indicate a complex interaction between CHIKV replication and mosquito innate immune responses and demonstrate similarities as well as differences in the control of alphaviruses and other arboviruses by mosquito immune pathways.
| Chikungunya virus (CHIKV) is a mosquito-borne human-pathogenic arbovirus of the Togaviridae family, genus Alphavirus. Arbovirus replication in vectors, such as mosquitoes, is not passively tolerated but leads to immune responses, that control virus infection. These responses therefore represent interesting targets for novel intervention strategies. Mosquito antiviral immune responses, such as small RNA pathways or immune signaling pathways, are increasingly well studied but it is not known which one mediate antiviral effects against CHIKV in particular. Here we screened four key immune responses in vitro for their antiviral potential against CHIKV and only the exogenous RNA interference was found to mediate antiviral activity. This was confirmed in vivo in Aedes aegypti mosquitoes. Immune signaling pathways were not found to mediate antiviral activity but were inhibited by CHIKV. This shows interesting differences and similarities to other mosquito-borne alphaviruses that increase our understanding of alphavirus-mosquito interactions.
| Arthropod-borne viruses (arboviruses) replicate in both their vertebrate host and arthropod vector. This poses a unique problem for arboviruses as they are exposed to both the vertebrate and invertebrate immune systems. Arthropod vectors of arboviruses, such as mosquitoes, do not have the combination of innate and adaptive immune systems of vertebrates and must therefore rely on the innate immune system to control arbovirus replication. When the mosquito vector ingests a blood-meal from a viraemic vertebrate host, arboviruses infect the midgut epithelial cells. Successful replication and passage through the midgut allows the virus to disseminate into the hemocoel and infect other tissues such as the trachea, fat body, and salivary glands. Once the virus is detected in saliva, the mosquito becomes competent for transmission to the next vertebrate host [1], [2]. Therefore, interactions between the replicating virus and the mosquito immune defence system produce an outcome that can influence subsequent viral transmission, as shown for the flavivirus dengue (DENV) [3]. This emphasizes the importance of further understanding innate immunity in arbovirus vectors.
Several innate immune responses have been reported to have an antiviral effect in mosquitoes, and these include RNA interference (RNAi), as well as responses mediated by Toll, Immune Deficiency (IMD) and JAK/STAT signaling pathways [4]–[6]. The exogenous (antiviral) RNAi pathway is induced by the presence of long double-stranded RNA (dsRNA) molecules, which in the case of RNA viruses may arise from secondary RNA structures in the viral genome, the viral genome itself (in case of dsRNA viruses) and/or viral replication intermediates. Much of our understanding of arthropod RNAi is based on research on Drosophila melanogaster, which has also proved very useful in providing insights into antiviral responses in mosquitoes [7]. In the exogenous RNAi pathway, these dsRNAs are recognized by the RNAse III enzyme Dicer 2 (Dcr-2) [8] and cleaved into 21 nt small interfering RNA (siRNA) also known as virus induced small interfering RNAs (viRNAs) [9]–[15], a hallmark of RNAi induction. The siRNAs/viRNAs are loaded into the multi-protein RNA Induced Silencing Complex (RISC), of which the core catalytic component is the endonuclease Argonaute-2 (Ago-2) [16]. Ago-2 unwinds the siRNAs/viRNAs and retains one strand (guide strand) to recognize single-stranded (viral) complementary sequence such as genomic RNA or mRNA, which triggers the endonucleic cleavage of the complementary RNA by Ago-2 [17]. A key role for mosquito Ago-2 and Dcr-2 in antiviral responses has also been demonstrated experimentally [18], [19]. Sequence specific degradation of RNA results in repression of virus replication and virus production. Similarly, Toll, IMD and JAK/STAT signaling pathways have been described in Drosophila with pathway homologues also identified in mosquitoes. In the mosquito Aedes aegypti, the Toll and JAK/STAT signaling pathways have been shown to induce antiviral activities targeting DENV [20], [21]. In culicine mosquitoes, West Nile virus (WNV) (Flaviviridae) was also shown to be inhibited by the JAK/STAT pathway and that this response is thought to be controlled by the cytokine Vago [22]. In the case of alphaviruses of the Togaviridae family, such as Sindbis virus (SINV) or Semliki Forest virus (SFV), the data are less clear. At least in mosquito cells, antiviral activity against SFV was shown to be mediated by either IMD or JAK/STAT pathways following bacterial stimulation [23]. Recent research in Drosophila further implies those pathways in control of SINV infection [24], [25] although no upregulation of target genes was shown in Ae. aegypti-derived Aag2 cells [26]. However, in the case of Anopheles gambiae infection by the alphavirus o'nyong-nyong (ONNV), the contribution of the IMD pathway is only minor [27]. These pathways may therefore act in a virus-specific manner. In contrast to vertebrate cells where viral inhibition of innate immunity is an area of intense research, similar processes in arbovirus/vector interactions are poorly understood. Several arboviruses have previously been shown to inhibit or evade the antiviral action of these host responses in mosquitoes. An RNAi evasion strategy has been suggested for SFV, while flavivirus subgenomic RNA acts as an RNAi inhibitor [11], [28]. Inhibition of immune signaling pathways such as JAK/STAT, Toll and IMD has been described for SFV [23], while DENV has been shown to interfere with Toll and IMD signaling [29]. This suggests further complexity in arbovirus/vector interactions and the relevance of this is only just emerging, and needs to be assessed for each virus family and even each virus individually.
Chikungunya virus (CHIKV) belongs to the genus Alphavirus of the family Togaviridae. It has become an increasingly important arbovirus in tropical and subtropical regions, resulting in febrile and arthralgic disease in humans [30]. After the large outbreak in the Indian Ocean in 2004, CHIKV infections have spread throughout the Indian Ocean and Africa into Southern Europe [31]. Moreover, the virus has emerged for the first time in the Americas (St. Martin/Martinique, Caribbean) in December 2013, thus potentially threatening other parts of the New World (see WHO Website for information). CHIKV is spread by aedine mosquitoes with Ae. aegypti being the most important vector and more recently Ae. albopictus following changes in the E1 envelope protein [32], [33]. The viral genome is a single stranded positive sense RNA containing two open reading frames: one expressing the non-structural polyprotein which will be cleaved into non-structural viral proteins (nsP1-4), and one structural polyprotein which will be cleaved into the structural proteins. The mRNA encoding the structural polyprotein is transcribed from the subgenomic promoter during virus replication. Little is known about innate immune responses induced by CHIKV during infection of mosquitoes. Identification of small RNAs derived from CHIKV in aedine mosquitoes and their derived cell lines proves the ability of the RNAi pathway to target CHIKV. Moreover, 21 nt viRNAs as well as another class of virus-derived small RNAs with characteristics of Piwi-interacting RNAs (piRNAs) were discovered [12]. However, nothing is known about the functional ability of this RNAi response to limit CHIKV replication in mosquito cells or mosquitoes, or the involvement of other innate immune responses.
In this study, we investigated the major antiviral Ae. aegypti immune pathways and assayed their ability to control CHIKV replication in cell culture, where experimental conditions are easily controlled and manipulated. The exogenous RNAi pathway was identified as a key control mechanism of CHIKV replication and this was further confirmed in mosquitoes. Moreover, CHIKV repressed Toll, IMD and JAK/STAT pathway stimulation in vitro by induction of host cell shut off and none of the pathways were able to mediate antiviral responses against CHIKV replication. These data indicate the importance of RNAi as a mosquito antiviral response also targeting CHIKV replication as well a complex interplay with other host responses.
CHIKV E1-226V strain (provided by the French National Reference Center for Arboviruses, Institut Pasteur) was used for mosquito infections. The strain was isolated from a patient on La Reunion Island in November 2005 as part of a survey during an outbreak, as described in [34].
Vero cells were cultured in DMEM (Dulbecco's Modified Eagle Medium) supplemented with 10% fetal bovine serum (FBS), 1000 units/ml penicillin, 1 mg/ml streptomycin, and maintained at 37°C and 5% CO2. Ae. aegypti Aag2 and Ae. albopictus C6/36 cells (for virus propagation) were grown in Leibovitz L-15 insect medium with 10% FBS, 10% tryptose phosphate broth, 1000 units/ml penicillin, and 1 mg/ml streptomycin at 28°C.
For CHIKV titer determination, Vero cells were grown to confluent monolayers in 6-well plates, infected with 10-fold serial dilutions of virus for 1 h, and then overlaid with an agarose-nutrient mixture. After 3 days incubation at 37°C, cells were stained with a solution of crystal violet (0.2% in 10% formaldehyde and 20% ethanol). The total number of plaques was counted and the titer was calculated. Titers are indicated as plaque forming units (PFU)/ml.
Use of plasmids pAct-Renilla (coding sequence of Renilla luciferase (RLuc) under control of the Drosophila actin 5C promoter), p6×2DRAF-Luc (coding sequence of firefly luciferase (FFLuc) under control of a multimerised Drosophila STAT-responsive element), pJL195 (coding sequence of FFLuc under control of the Drosophila attacin A promoter) and pJM648 (coding sequence of FFLuc under control of the Drosophila drosomycin promoter) in signaling pathway experiments have been previously described [23]. Luciferase SP6 Control DNA plasmid (Promega) expressing FFLuc under control of SP6 promoter was used as a template for in vitro transcription of a control RNA.
Nonstructural open reading frame of CHIKVRep(3F)RLuc-SG-FFLuc replicon contains RLuc encoding sequence inserted between codons for amino acid residues 383 and 384 of CHIKV nsP3. In place of the structural open reading frame, mRNA encoding for FFLuc is transcribed following replication by use of the subgenomic promoter. Replicon CHIKVRep-SG-eGFP contains the nonstructural open reading frame without any reporter genes and mRNA of eGFP is transcribed following replication by use of the subgenomic promoter.
Replicon DNA was linearized by digestion with NotI, purified using a PCR purification kit (Roche) and 1 µg of the linearized DNA was in vitro transcribed using MEGAscript SP6 polymerase kit (Ambion) in the presence of Cap analog (Ambion). Control Luciferase SP6 DNA was linearized by digestion with XmnI, purified by gel extraction and in vitro transcribed as before. For all subsequent experiments, 2 µl of in vitro transcription was transfected.
Total RNA was extracted from adult Ae. aegypti using the RNA II Nucleospin kit (Macherey-Nagel GmbH & Co.) according to the manufacturer's instructions. cDNAs were generated from 60 ng of total RNA by reverse transcription using SuperScript II Reverse Transcriptase (Invitrogen) and oligo dT. To synthesize dsRNA, cDNA was amplified with gene specific primers (table 1 designated as being for use in vivo) incorporating the T7 RNA polymerase promoter sequences (in bold) at the 5′ ends. Primers were designed to amplify a unique ∼500 bp region in the PIWI domain for Ago-2. PCR was carried out using the KOD Hot Start DNA Polymerase (Novagen). PCR products were purified with the QIAquick Gel Extraction kit (Qiagen) and dsRNA was produced using the MEGAscript RNAi kit (Ambion) according to the manufacturer's instructions.
RNA was extracted from Aag2 cells using TRIzol and reverse transcribed using Superscript III Reverse Transcriptase (Invitrogen) following the manufacturer's instructions. PCR products were generated with T7 promoter sequences at either end of the fragment using the primers listed in table 1 and designated as for use in vitro. The PCR product was blunt end cloned into a pJet1.2 vector (Thermo Scientific) following the manufacturer's instructions. Cloned PCR fragments were verified by sequencing. PCR was performed on the cloned fragments and the products purified using the PCR purification kit (Roche). peGFP-C1 (Clontech) was used as a template for the amplification of control dsRNA, targeting eGFP. dsRNA was synthesized using the Megascript RNAi kit (Ambion) following manufacturer's instructions.
24 h prior to transfection, 1.7×105 Aag2 cells were seeded in 24-well plates to reach 70% confluence the following day. For knockdown experiments, 500 ng of dsRNAs were transfected into Aag2 cells using Opti-MEM and Lipofectamine2000 (Invitrogen) according to manufacturer's instructions. At 24 h post dsRNA transfection, cells were transfected with capped in vitro transcribed replicon RNA derived from CHIKVRep(3F)RLuc-SG-FFLuc using Lipofectamine2000. Cells were lysed in 1× passive lysis buffer (Promega) 24 h post replicon transfection and luciferase activities determined.
For the signaling pathway stimulation experiments, bacterial stocks were prepared by inoculation of 1 µl E. coli JM109 in 5 ml L-Broth and incubation at 37°C for 16 h or 5 µl B. subtilis inoculated in 5 ml L-Broth incubated at 37°C for 8 h. E. coli is used to stimulate the JAK/STAT and IMD pathways and B. subtilis is used to stimulate the Toll pathway. Cultures were centrifuged at 1174 g at 4°C for 10 minutes. The bacterial pellet was then resuspended in 500 µl PBS and heat inactivated at 80°C for 10 minutes. At 24 h post seeding, Aag2 cells were transfected with 12.5 ng pAct-Renilla, 25 ng p6×2DRAF-Luc, 25 ng pJL195 or 500 ng pJM648 and capped in vitro transcribed CHIKV replicon RNA (derived from CHIKVRep-SG-eGFP) at 24 h post seeding, using Lipofectamine2000 according to the manufacturer's protocol. Cells were stimulated at 24 h post transfection with heat inactivated E. coli JM109 or B. subtilis for 1 hour at 28°C. 12 h post stimulation, cells were lysed in 1× passive lysis buffer and luciferase activities determined.
Ae. aegypti Liverpool strain (provided by D. Severson, University of Notre Dame, IN, USA) were maintained on 10% sugar solution at 28°C with a photophase of 16 h and 80% relative humidity according to the standard rearing procedures.
RNAi-based gene-silencing assays were conducted by injecting 500 ng of dsRNA (dsAgo-2 or dsFFLuc) in water into the thorax of cold-anesthetized 4 day-old females using a Nanoject II microinjector (Drummond Scientific Company). Blood feeding was carried out 48 h post dsRNA injection. Gene silencing validations were performed at day 1, 2, 3, 4 and 7 after ingestion of the infectious blood-meal. As controls, mosquitoes injected with PBS and non-injected unfed mosquitoes were used.
Adult female mosquitoes were deprived of sugar source 24 h before infection and allowed to feed on artificial blood-meals consisting of a virus suspension (1/3 vol/vol), washed rabbit erythrocytes (2/3 vol/vol), and 5 mM ATP. The artificial blood-meal was provided in glass feeders covered with a chicken skin membrane and maintained at 37°C. Females placed in plastic boxes were allowed to feed for 15 minutes. Engorged females were selected, transferred into cardboard containers, provided with sugar solution and maintained in BSL-3 insectaries until analysis. To determine the titer of the infectious blood-meal which would be used for subsequent RNAi-based gene-silencing assays, three different virus titers: 106, 107, 108 PFU/ml were tested. At indicated time points, 10 females were tested for the presence of CHIKV by immunofluorescence assay (IFA) on head squashes using mouse ascitic fluid raised against the virus [35]. Disseminated infection rates (DIR) corresponding to the number of females with disseminated infection among tested females, were determined. At various days post infection, mosquitoes were analyzed as follows: midgut and head were isolated from each individual and ground in 150 µl DMEM before being homogenised and filtered. 50 µl of the filtrate was titrated by plaque assay on Vero cells to estimate the number of infectious viral particles. The remaining 100 µl was used for quantification of Ago-2 and ribosomal S7 RNA by real time quantitative RT-PCR.
Dissected mosquito organs (midgut and head) were homogenized separately and used for RNA extraction using the NucleoSpin 96 RNA kit (Macherey-Nagel GmbH & Co). An equal amount of RNA extracted from each organ was examined for each time point. Quantification was carried out by real time quantitative RT-PCR using the Power SYBR Green RNA-to-CT one step kit (Applied Biosystem).
Total RNA was extracted from cultured cells using TRIzol (Invitrogen) according to the manufacturer's instructions. Polyadenylated RNAs were reverse transcribed using the Superscript III kit (Invitrogen) and oligo dT primers (Promega) according to the manufacturer's recommendation. Quantification was carried out by real time quantitative PCR using the Fast SYBR green master mix (Invitrogen).
All PCR reactions were done in triplicate. Specificity of the PCR reactions was assessed by analysis of melting curves for each data point. Values were normalized against the Aedes aegypti ribosomal protein S7 gene. Following real-time quantitative PCR assays, analysis of relative expression of Ago-1, Ago-2 and S7 was performed according to the 2−ΔΔCt method [36].
Luciferase expression was determined using the Dual Luciferase kit (Promega). Luciferase activities were determined on a Glomax Multi+ Microplate Multimode reader (Promega).
Virus titer means were compared using the Kruskall-Wallis test from the STATA software (StataCorp LP). Statistical significance for replicon replication after knockdown was determined by a paired Student's t-test and statistical significance in the stimulation experiments was determined using a general linear mixed model.
Aag2 cells have proven to be a reliable model for the study of aedine immune responses including those against viruses given the presence of all major immune signaling pathways and small RNA pathways in the cell line [10], [11], [14], [26], [29], [37]. Not only have mosquito cell lines been shown to have intact immune pathways but they also provide a robust system where experiments can be performed in a controlled manner and have been used extensively in the mosquito immunity field [10]–[12], [23], [29], [38]–[40]. Viral replicons are useful tools to investigate innate immune responses in a tightly controlled manner in cell culture experiments. We first engineered two CHIKV reporter replicons on the basis of the CHIKV E1-226V strain. The first replicon, named CHIKVRep(3F)RLuc-SG-FFLuc, encodes a non-structural polyprotein with RLuc inserted into the C-terminal region of nsP3 and mRNA of FFLuc expressed from the viral subgenomic promoter instead of the mRNA of structural proteins. A second replicon is expressing mRNA of eGFP from the viral subgenomic promoter: CHIKVRep-SG-eGFP (Fig. 1A). This replicon was designed with a fluorescent protein as a reporter rather than luciferase to allow a multitude of different experiments to be performed.
To determine if these replicons are functional in mosquito cells and to characterize their kinetics with regards to the expression of the reporter proteins, Aag2 cells were transfected with in vitro transcribed capped CHIKVRep(3F)RLuc-SG-FFLuc, lysed at different time points post transfection (24, 48 and 72 hours post transfection; hpt) and luciferase expression determined. Luciferase expression (both FFLuc [data not shown] and RLuc) can be measured by 24 hpt with a peak at 48 hpt and a slight decrease at 72 hpt (Fig. 1B). These time points were chosen to allow sufficient time for replicon replication but also taking into consideration the transient nature of transfections. These data suggest that the engineered replicons are replicating in Aag2 cells.
The RNAi pathway has been identified as being the major antiviral pathway in control of replication of a number of arboviruses in mosquitoes [4], [5]. The generally accepted method of confirming antiviral activity in mosquito cell culture and mosquitoes is RNAi mediated knockdown of components of individual small RNA pathways [3], [18], [19], [37]. Therefore, in order to determine whether the exogenous RNAi pathway also limits CHIKV replication in Aag2 cells, unique dsRNAs were designed and validated against the exogenous RNAi pathway component Ago-2 and the miRNA pathway component Ago-1 [4], [5]. Knockdown was determined by quantitative RT-PCR (Fig. 2A). Having shown efficient knockdown of both Ago-1 and Ago-2 (42% or 25% respectively), their contribution in control of CHIKV replication was assessed. Aag2 cells were transfected with dsRNA (Ago-1 and Ago-2 specific or eGFP-specific as a control), followed by transfection of in vitro transcribed capped CHIKVRep(3F)RLuc-SG-FFLuc and lysed after 24 hpt of replicon. A significant (Student's t-test; p = 0.045) 9-fold increase in RLuc expression was observed in cells treated with Ago-2 specific dsRNA compared to cells with control dsRNA. In contrast, no significant increase in RLuc expression was observed for cells transfected with Ago-1 specific dsRNA (Fig. 2B). These data suggest that the miRNA pathway is not involved in the inhibition of CHIKV replication in Ae. aegypti-derived cells, in contrast to the exogenous RNAi response that is able to inhibit CHIKV replication in vitro.
Having shown an inhibitory effect of the exogenous RNAi pathway on CHIKV replication in mosquito cells, similar experiments were performed in mosquitoes to determine if Ae. aegypti RNAi components are required for defence against CHIKV in vivo.
First, CHIKV infection kinetics was determined in Ae. aegypti mosquitoes, infected with bloodmeals containing different virus titers, by subsequent determination of dissemination rates by IFA on head squashes at several time points (Fig. 3). Maximum dissemination was observed at day 7 post infection (pi) with 52% for a blood-meal titer of 106 PFU/ml, at day 6 pi (100%) for 107 PFU/ml, and at day 3 pi (100%) for 108 PFU/ml. Based on these patterns, the intermediate titer of 107 PFU/ml was chosen for further assays on RNAi-based gene-silencing. Gene silencing efficiency was then tested. 500 ng dsRNA (Ago-2 specific or FFLuc specific as a control) was injected into mosquitoes 48 h prior to infectious blood-meal (CHIKV at 107 PFU/ml). Ago-2 expression (relative to S7 expression) in midguts or heads of injected mosquitoes was determined by real-time quantitative PCR at time points indicated (1, 2, 3, 4 and 7 days pi). The S7 gene was chosen as a control housekeeping gene due to its stability in infected and non-infected conditions [data not shown]. Mosquitoes were analyzed at days 0, 1, 2, 3, 4 and 7 pi. Injection of Ago-2 specific dsRNA strongly decreased the expression of Ago-2 in midguts (Fig. 4A: ranging from 80% silencing at day 2 to 70% silencing at day 7) and heads (Fig. 4C: ranging from 68% at day 2 to 50% at day 7) compared to the controls (FFLuc specific dsRNA dsFFLuc, PBS or non-injected mosquitoes). Silencing was effective from day 2 up to 7 pi.
To test whether silencing of Ago-2 expression would increase CHIKV replication in Ae. aegypti following an infectious blood-meal, midguts and heads of the previous experiments were examined for CHIKV production. The number of infectious viral particles was determined by plaque assay of midguts and heads of 10 females at day 1, 2, 3, 4, and 7 pi (Fig. 4 B and D). A non-significant increase of virus in midguts was observed between mosquitoes injected with Ago-2 specific dsRNA and controls (dsFFLuc, PBS and non-injected) at day 1, 2 and 3 pi (Kruskall-Wallis Test: p>0.05). However, at day 4 and 7 pi a significant increase in the number of virus particles in midguts was detected following Ago-2 knockdown (Kruskall-Wallis test: p<0.05) (Fig. 4B). To determine the effect of gene silencing on viral dissemination, the number of viral particles in heads was also determined. At day 4 pi, heads of mosquitoes injected with Ago-2-specific dsRNA contained significantly more infectious viral particles (103.1±102.3) (Fig. 4D) than controls (Kruskall-Wallis test: p<0.05). This effect was transitory as at day 7 pi, viral loads remained similar when compared to other treatments. Taken together the in vitro and in vivo data imply a control of virus replication by the exogenous RNAi pathway with the virus being unable or not efficiently able to avoid this antiviral response.
Similar to other arboviruses, CHIKV may have evolved mechanisms which allow the virus to evade or suppress the induction of innate immune signaling pathways [6]. To investigate this possibility, signaling assays were performed in Aag2 cells similar to those described previously [23]. First, inhibition of innate immune pathways by CHIKV RNA was determined. Gene expression studies have shown the presence of JAK/STAT, IMD and Toll in Aag2 cells, which make them suitable models for the subsequent experiments [26], [29]. Preliminary experiments indicated the suitability of reporter gene expression studies and bacterial stimulation to study these pathways also in our Aag2 cells [not shown]. Therefore, Aag2 cells were co-transfected with in vitro transcribed capped CHIKVRep-SG-eGFP (not expressing luciferase) and plasmids encoding FFLuc under control of promoters that are activated in response to immune signaling of the JAK/STAT (p6×2DRAF-Luc), IMD (pJL195) and Toll (pJM648) pathways and an internal transfection control plasmid pAct-Renilla. This was followed by stimulation of the different pathways by either heat inactivated E. coli or B. subtilis and expression of FFLuc and RLuc was detected. Stimulated cells without CHIKV replicon showed a significant increase in FFLuc expression in case of E. coli or B. subtilis respectively for both JAK/STAT, IMD and Toll stimulations (Fig. 5 A–C control versus E. coli or B. subtilis [p = <0.001]). Cells co-transfected with the CHIKV RNA and pathway specific constructs, showed a much reduced level of stimulation compared to cells lacking the CHIKV replicon after IMD stimulation and a highly significant reduction in the level of stimulation after JAK/STAT and Toll stimulation (2.5 fold less JAK/STAT stimulation, 3.4 fold less IMD stimulation and 4.9 fold less Toll pathway stimulation in the presence of CHIKV) (Fig. 5 A–C control versus CHIKV+E. coli or CHIKV+B. subtilis [p = <0.001]). However, presence of the CHIKV replicon RNA also significantly reduced RLuc expression, regardless of which pathway was tested, compared to cells lacking the CHIKV replicon (Fig. 5 A–C control vs CHIKV [p = <0.001]). Moreover, inclusion of the CHIKV replicon RNA reduced the RLuc expression levels without stimulation of other pathways (Fig. 5 A–C control versus CHIKV [p = <0.001]). The reduction in RLuc expression is specifically due to the presence of the CHIKV replicon as co-transfection of pACT-Renilla plasmid and a control luciferase RNA expressing FFLuc did not result in a reduction of RLuc expression (Fig. 5D). These data suggest that, similar to findings for the related alphavirus SFV [23], CHIKV can shut down, albeit incompletely, the host cell transcription/translation systems and this general targeting mechanism also interferes with immune pathway stimulation, especially with regards to the Toll pathway which cannot be stimulated in the presence of CHIKV. In order to determine on which cellular process CHIKV is exerting its effect, CHIKV replicon RNA was co-transfected with control FFLuc expressing RNA and the translation of FFLuc assessed by luciferase assay. While there is a small reduction in FFLuc expression upon co-transfection of CHIKVRep-SG-eGFP replicon, there is not the same significant effect on the control RNA in the presence of CHIKV (Fig. 5E) as there is during the co-transfection with the pACT-Renilla plasmid. Therefore, similar to the related SINV and SFV, it is likely that CHIKV host cell shut-off occurs at a transcriptional level [41]–[43]. It should be noted that this incomplete or weak shut-off does not appear to have the detrimental effects often observed in alphavirus-infected vertebrate cells, but nonetheless affects host cell signaling.
As CHIKV was shown to inhibit immune signaling via host cell shut off, we hypothesized these pathways could induce antiviral activities. In order to investigate these possibilities, Aag2 cells were stimulated with either heat inactivated E. coli (to induce JAK/STAT and IMD signaling) or B. subtilis (to induce Toll signaling) prior to transfection of in vitro transcribed and capped CHIKVRep(3F)RLuc-SG-FFLuc. Replication was determined by measuring by FFLuc expression (Fig. 6). Stimulation of cells prior to replicon transfection had no effect on CHIKV replication regardless of whether cells were stimulated with gram negative or gram positive bacteria. This suggests the Toll, IMD and JAK/STAT pathways cannot affect CHIKV replicon replication.
Innate immune responses are important for regulating arboviral replication in mosquitoes. The JAK/STAT, IMD and Toll immune signaling pathways have been individually implicated in the control of arbovirus replication in mosquitoes in a virus/pathway-dependent manner [20], [21], [23], [25]. In addition, the exogenous RNAi pathway is a key mosquito antiviral pathway [4], [5]. The re-emerging CHIKV is known to induce an RNAi response in infected mosquitoes [12]. CHIKV infection of mosquitoes and their derived cell lines results in the production of small RNAs indicating that small RNA pathways are activate against CHIKV and that the presence of an RNAi inhibitor resulted in increased virus replication and virus production [12]. This study, however, did not identify which small RNA pathways have antiviral activity. Expression of the RNAi inhibitor B2 by CHIKV affected both exogenous RNAi and piRNA pathways thus making it unclear which of these pathways affected viral replication. Indeed, our study identifies one such pathway specifically. Here we have demonstrated that Ago-2 plays an important role in the antiviral RNAi response to CHIKV both in vitro and in vivo, and that the induced exogenous RNAi response mediates effective antiviral activities. The exogenous RNAi pathway has previously been shown to mediate antiviral activities against three other alphaviruses, SINV, SFV and ONNV [18], [19], [37]. In the case of CHIKV infection, production of small RNAs produced by the exogenous RNAi and piRNA pathways has also been demonstrated [12] and our data demonstrate that the exogenous RNAi does indeed mediate antiviral activity. In the case of SINV infection of Ae. aegypti mosquitoes, knockdown of Ago-2 resulted in a transient increase in virus replication and titer early in infection; however, similar to our findings, the effect was lost by day 7 [18]. Additionally, engineering the Flock House Virus RNAi inhibitor B2 protein into SINV increased the mortality rate, indicating that RNAi is vital in controlling virus replication to levels that are non-pathogenic to the arthropod vector [44]. A similar effect was reported for exogenous RNAi control of DENV replication in Ae. aegypti [3]. Our results extend the importance of the exogenous RNAi pathway to another virus of medical importance, CHIKV. Knockdown results for Ago-2 directly show that CHIKV replication is inhibited by the RNAi pathway and that the small RNAs produced in the exogenous RNAi pathway do mediate antiviral activity. It has been postulated that arboviruses may in fact subject themselves to RNAi control to ensure vector survival, which may explain why no RNA silencing suppressor (RSS) proteins have been identified to date in arboviruses. Recently, an RSS was found in the flaviviruses DENV and WNV. The RSS was not a viral protein but a highly structured region in the 3′ UTR of the flavivirus genome called subgenomic flavivirus RNA (sfRNA) which is proposed to act as an RNAi inhibitor [28]. It is yet to be seen if a similar mechanism is employed by alphaviruses to escape replication control by RNAi although an evasion mechanism has been suggested for SFV [11].
Antiviral responses other than RNAi in mosquitoes are beginning to emerge, and these innate immune pathways play a vital role in antiviral defence in a virus dependent manner. We have identified that in particular the Toll immune signaling pathway is strongly inhibited by CHIKV in vitro, most likely by a mechanism involving host shut-off. The JAK/STAT and IMD pathways have previously been reported to be involved in anti-viral defense against SINV in Drosophila [24], [25]. On the other hand, gene array studies suggested inhibition of Toll signaling late in infection and possibly activation of IMD signaling in SINV-infected Ae. aegypti mosquitoes [45]. However, in Aag2 cells infected with SINV gene expression studies revealed no activation of these pathways (with the exception of weak upregulation of STAT itself) [26]. The design of these experiments with SINV makes it difficult to come to any conclusion with regards to viral inhibition, but recent work has shown that SFV as well as DENV have been able to inhibit innate immune signaling in mosquito cells [23], [29]. Despite this, IMD and/or JAK/STAT but not Toll can mediate antiviral activity against SFV, while Toll and JAK/STAT induce antiviral responses against DENV [20], [21], [23]. We found no effect of any of these pathways on CHIKV replicon replication either because they do not act antivirally or because viral inhibition of any antiviral effects is strong enough to mask these. This points to virus specific differences or highly efficient inhibition of the immune pathways by CHIKV. Interestingly, only a minor activity by the IMD pathway was shown against ONNV [27], which is closely related to CHIKV therefore these viruses may respond in a more similar manner. Overall, this interplay between antiviral signaling and viral inhibition is strongly reminiscent of viral interactions with the innate immune responses, such as the interferon system of vertebrate cells [46]–[48]. At least for SFV, and our observations with CHIKV, shut off of host gene expression appears to mediate signaling inhibition in mosquito cells, as it does in vertebrate cells [23]. The mechanism by which shut-off occurs is not clear, although our data (Fig. 5E) suggest an effect prior to translation, possibly at the transcriptional level as occurs in vertebrate cells [41]–. We cannot rule out competition between host and viral mRNAs, differences in mRNA stability or other processes that would lead to differences in translation, however, as reporter mRNAs are not affected by CHIKV replicon this appears unlikely. Other inhibitory strategies employed by arboviruses may also resemble those used in vertebrate cells. In the presence of RNAi, virus replication may still be controlled to a non-harmful level. Recent work has indicated candidates for antimicrobial peptides, attC and dptB, that inhibit SINV replication in Drosophila [24]. Similar mechanisms in mosquito cells remain to be determined. Clearly, there is no direct effect of these classical immune signaling pathways on CHIKV replication under our experimental conditions.
In conclusion, we demonstrated that the exogenous RNAi pathway plays a vital role in limiting CHIKV replication in cell culture and in mosquitoes. Knockdown of Ago-2 resulted in a significant increase in RNA replication and virus titers. Additionally, we have shown that CHIKV significantly represses Toll pathway stimulation, and stimulation of major insect immune pathways did not limit CHIKV replication. This indicates that antiviral immunity is a complex process which needs more research to tease out the complexities of the virus/host interactions and differences exist even between closely related alphaviruses. Exogenous RNAi is also active against CHIKV while analysis of immune signaling pathways indicates differences to other arboviruses including other alphaviruses. Taken together this and other studies suggest that the RNAi response is the most generally active antiviral pathway in mosquitoes.
|
10.1371/journal.ppat.1004256 | Novel Drosophila Viruses Encode Host-Specific Suppressors of RNAi | The ongoing conflict between viruses and their hosts can drive the co-evolution between host immune genes and viral suppressors of immunity. It has been suggested that an evolutionary ‘arms race’ may occur between rapidly evolving components of the antiviral RNAi pathway of Drosophila and viral genes that antagonize it. We have recently shown that viral protein 1 (VP1) of Drosophila melanogaster Nora virus (DmelNV) suppresses Argonaute-2 (AGO2)-mediated target RNA cleavage (slicer activity) to antagonize antiviral RNAi. Here we show that viral AGO2 antagonists of divergent Nora-like viruses can have host specific activities. We have identified novel Nora-like viruses in wild-caught populations of D. immigrans (DimmNV) and D. subobscura (DsubNV) that are 36% and 26% divergent from DmelNV at the amino acid level. We show that DimmNV and DsubNV VP1 are unable to suppress RNAi in D. melanogaster S2 cells, whereas DmelNV VP1 potently suppresses RNAi in this host species. Moreover, we show that the RNAi suppressor activity of DimmNV VP1 is restricted to its natural host species, D. immigrans. Specifically, we find that DimmNV VP1 interacts with D. immigrans AGO2, but not with D. melanogaster AGO2, and that it suppresses slicer activity in embryo lysates from D. immigrans, but not in lysates from D. melanogaster. This species-specific interaction is reflected in the ability of DimmNV VP1 to enhance RNA production by a recombinant Sindbis virus in a host-specific manner. Our results emphasize the importance of analyzing viral RNAi suppressor activity in the relevant host species. We suggest that rapid co-evolution between RNA viruses and their hosts may result in host species-specific activities of RNAi suppressor proteins, and therefore that viral RNAi suppressors could be host-specificity factors.
| Viruses and their hosts can engage in an evolutionary arms race. Viruses may select for hosts with more effective immune responses, whereas the immune response of the host may select for viruses that evade the immune system. These viral counter-defenses may in turn drive adaptations in host immune genes. A potential outcome of this perpetual cycle is that the interaction between virus and host becomes more specific. In insects, the host antiviral RNAi machinery exerts strong evolutionary pressure that has led to the evolution of viral proteins that can antagonize the RNAi response. We have identified novel viruses that infect different fruit fly species and we show that the RNAi suppressor proteins of these viruses can be specific to their host. Furthermore, we show that these proteins can enhance virus replication in a host-specific manner. These results are in line with the hypothesis that virus-host co-evolution shapes the genomes of both virus and host. Moreover, our results suggest that RNAi suppressor proteins have the potential to determine host specificity of viruses.
| As obligate intracellular parasites, viruses modulate and exploit the host cellular environment for their replication. The host antiviral defense system restricts virus infections, and in turn, viruses dedicate a significant fraction of their coding capacity to produce factors that antagonize the antiviral immune response [1], [2]. Co-evolution of virus and host may therefore lead to a host-specific adaptation of viral counter-defense to the host antiviral defense system, which can contribute to host specificity of the virus [3].
The RNA interference (RNAi) pathway is a major antiviral defense system in plants, arthropods, nematodes and fungi [4]–[7] and has recently been suggested to control virus infection in mammals [8], [9]. Double stranded RNA (dsRNA), which is typically produced during virus infection but absent from non-infected cells [10], triggers the RNAi pathway. In insects, cleavage of viral dsRNA by the ribonuclease Dicer-2 (Dcr-2) generates viral small interfering RNAs (vsiRNAs) [11]–[23]. Dcr-2 and its binding partner R2D2 bind these vsiRNAs and load the small RNA duplexes into an Argonaute-2 (AGO2) containing RNA induced silencing complex (RISC) [24]. One strand of the vsiRNA is retained and guides the recognition and cleavage of complementary viral RNAs by AGO2 [11], [25]–[28]. In response, insect and plant viruses encode suppressors of RNAi (VSRs) to counteract the antiviral RNAi pathway [29]. Different mechanisms for RNAi suppression have been identified; for example, some VSRs bind long dsRNA and/or siRNAs to shield them from Dicer cleavage or prevent their loading into Argonaute [11], [30]–[38]. Other suppressors interact with Argonaute proteins to inhibit their activity or induce their degradation [14], [39]–[45].
The ongoing arms race with viruses can impose a strong selective pressure on immune genes of the host [46]. Consistent with this, Dcr-2, R2D2, and AGO2 belong to the 3% fastest evolving genes in D. melanogaster and D. simulans and show very high rates of adaptive amino acid substitution with evidence for recent selective sweeps in multiple Drosophila species [47]–[49]. It has been hypothesized that this rapid adaptive evolution may be driven by antagonistic co-evolution with viral suppressors of RNAi [50], as the RNAi pathway continues to evolve new ways to escape viral antagonists, leading to counter-adaptations by viruses that require further adaptations in the RNAi pathway of the host. A potential outcome of this antagonistic co-evolution is that viral RNAi suppressors become specialized to suppress RNAi in their host species, while losing this activity in non-host species. This may be unlikely for viral antagonists that bind dsRNA, which often efficiently suppress RNAi in both host and non-host species, and in some cases even across kingdoms [51]–[55]. However, when viruses antagonize protein components of the RNAi pathway, there is ample opportunity for co-evolution and the evolution of host-specificity.
Nora virus of Drosophila melanogaster (DmelNV) is a recently identified natural fruit fly pathogen, which contains a single-stranded positive-sense RNA genome and appears to fall within the order of Picornavirales [56]. In contrast to other picorna-like viruses, DmelNV encodes four open reading frames: ORF 2 encodes replication proteins with clear homology to other Picornavirales members, ORF 4 encodes capsid proteins [57] (Figure 1A). No homology exists between the protein products of ORF1 or ORF3 and proteins of other viruses.
DmelNV causes persistent infections in laboratory stocks as well as in wild caught flies. Persistent infections are thought to reflect a dynamic equilibrium between host defense responses and viral counter-defense mechanisms [58]. The widespread abundance and persistent nature of DmelNV infections may suggest an equilibrium between antiviral RNAi and viral counter-defense, in which replication is restrained, but the infection is not cleared. Consistent with this, we recently showed that DmelNV is a target and a suppressor of the antiviral RNAi pathway [14]. We identified viral protein 1 (VP1), the product of open reading frame 1, as an RNAi suppressor that counteracts AGO2 mediated target RNA cleavage (slicer activity).
Here we present two novel Nora-like viruses identified by metagenomic sequencing of wild populations of D. immigrans (DimmNV) and D. subobscura (DsubNV), and we use these viral genomes to study RNAi antagonism from an evolutionary perspective. We find that the RNAi suppressor activity of DimmNV VP1 appears to be restricted to its natural host species, whereas DmelNV VP1 does not display any evidence of host specificity. We conclude that co-evolution between Nora viruses and their Drosophila hosts can result in host species-specific antagonism of AGO2, and therefore that viral suppressors of RNAi are candidate host specificity determinants.
RNAi genes evolve rapidly and adaptively in multiple species of Drosophila [47], [48]. We therefore hypothesized that the interaction between RNAi proteins and viral suppressors of RNAi may also evolve rapidly when viruses adapt to different hosts. In particular, optimization of such interactions in a specific host species may come at the cost of losing the interaction in non-host species. To test these hypotheses, we set out to identify novel Nora-like viruses from divergent Drosophila species.
During an exploratory RT-PCR survey of Nora virus prevalence in wild Drosophila, we identified two novel Nora-like viruses in wild populations of D. immigrans (DimmNV) and D. subobscura (DsubNV). Following this observation, we took a metagenomic RNA-sequencing approach to recover near-complete viral genomes for both viruses from population samples of D. immigrans and D. subobscura collected in the United Kingdom. The viral sequences were 12,265 nt and 12,276 nt, respectively (compare to 12,333 nt for DmelNV) and include all protein coding regions, a conserved CCTGGGSGGGGGTTA motif in their 5′ untranslated region, and a 3′ poly-A tract (Figure S1A). These novel viruses are more closely related to the Nora virus originally identified in D. melanogaster (DmelNV) [56] than they are to the Nora-like virus recently described in the horn fly Haematobia irritans [59], two Nora-like viruses identifiable in the transcriptomes of the lacewing Chrysopa pallens and the moth Spodoptera exigua, or the more distantly related Nora-like virus described in the wasp Nasonia vitripennis [60] (Figure 1B).
Overall, DmelNV is more divergent from DimmNV than it is from DsubNV (65% vs. 71% overall nucleotide identity, respectively), but phylogenetic analysis based on the coat protein (VP4) suggests that DmelNV and DimmNV may be each other's closest relatives. The low genome-wide nonsynonymous to synonymous substitution ratio (dN/dS = 0.076, SE = 0.003) estimated by PAML [61] indicates that evolution of the protein sequence is highly constrained. However, divergence between the three viruses is too high to reliably estimate dS [62], [63] and the estimated dN/dS may represent an upper limit.
Amino-acid divergence between the viruses varies substantially between genes (Figure 1C). For example, amino-acid identity between DimmNV and DmelNV varies from 82% for VP4 (capsid) to only 43% for VP3 (unknown function), with VP1 showing an intermediate level of conservation (51% amino acid identity). A sliding-window analysis of nonsynonymous divergence shows that DimmNV is much more divergent from DmelNV and DsubNV in VP1 and VP2, but that the three viruses are equidistant from each other in VP3 and VP4. This may be a result of host-mediated selection, perhaps reflecting the closer relationship between D. melanogaster and D. subobscura, or it may be a result of recombination in the history of these three viruses.
To test whether the interaction between antiviral RNAi components and viral RNAi antagonists is host specific, we first analyzed whether the DimmNV and DsubNV VP1 proteins are able to suppress RNAi in the S2 cell line from D. melanogaster. To this end, we cloned the full-length (FL) VP1 sequences and N- and C-terminal deletion mutants thereof (ΔN and ΔC) as N-terminal fusions to the V5 epitope in an insect expression plasmid (Figure S1B). We verified expression of the DimmNV VP1 constructs by western blot after transfection in Drosophila S2 cells (Figure 2A). With the exception of the DimmNV VP1ΔN362, all DimmNV VP1 constructs were expressed at least at the level of DmelNV VP1FL that efficiently suppresses RNAi in reporter assays in S2 cells [14]. We then analyzed the ability of the DimmNV VP1 constructs to suppress RNAi in reporter assays. We transfected S2 cells with firefly and Renilla luciferase (Fluc and Rluc) reporter plasmids along with VP1 expression plasmids, and induced silencing of the Fluc reporter by soaking the cells in Fluc specific dsRNA. As reported earlier [14], all DmelNV VP1 constructs, except DmelNV VP1ΔC74, suppressed RNAi-mediated silencing of the Fluc reporter. In contrast, none of the DimmNV VP1 constructs efficiently suppressed silencing of the reporter (Figure 2B). To confirm these results, we used an RNAi sensor assay that is independent of dsRNA uptake by S2 cells. In this sensor assay, the Rluc reporter is silenced by expression of an inverted repeat that folds into an Rluc-specific RNA hairpin. In line with the previous RNAi sensor assay, DimmNV VP1 did not suppress hairpin-induced silencing of the Rluc reporter in D. melanogaster S2 cells, whereas DmelNV VP1 efficiently suppressed RNAi (Figure 2C). In addition, we tested if the VP1 constructs can suppress RNAi in a sensor assay in which silencing is induced by co-transfection of siRNAs. Also in this assay, DimmNV VP1 was unable to suppress silencing of the Fluc reporter, whereas DmelNV VP1 efficiently suppressed RNAi-based silencing (Figure S2). Similarly, the DsubNV VP1 constructs were unable to suppress long dsRNA or siRNA induced RNAi in D. melanogaster derived S2 cells (Figure S2A–C). Moreover, recombinant DmelNV VP1 efficiently suppressed AGO2 slicer activity in embryo lysates of D. melanogaster, whereas DsubNV VP1 was unable to do so (Figure S2D). Together, these results indicate that VP1 of DimmNV and DsubNV do not suppress RNAi in D. melanogaster.
The inability of DimmNV VP1 and DsubNV VP1 to suppress RNAi in Drosophila S2 cells may be explained in two ways. First, viral RNAi suppressors may have a species-specific activity, following the prediction that prolonged virus-host coevolution may result in efficient RNAi suppressive activity in host species but not in non-host species. Second, some Nora-like viruses may either be unable to suppress RNAi, or they may encode RNAi suppressor activity in different regions of the viral genome, as has been observed for members of a single plant virus family [64]–[66]. To address the first possibility, we tested the ability of DimmNV VP1 and DmelNV VP1 to suppress RNAi in both host species using in vitro RNA cleavage (slicer) assays [67] in lysates of embryos from D. melanogaster and D. immigrans. Unfortunately, we were not successful in producing slicer competent lysates for D. subobscura. Moreover, members of the Drosophila obscura group encode multiple AGO2-like proteins of unknown function [68]. These proteins may be functionally redundant, and may not all be targeted by a VSR. We therefore chose not to include D. subobscura and DsubNV in subsequent analyses.
In slicer assays, RNAi dependent cleavage of a 32P cap-labelled target RNA is induced by the addition of a target specific siRNA. Since the target RNA is radio-labelled at its 5′ cap, the 5′ cleavage product can be visualized by autoradiography after polyacrylamide gel electrophoresis. As expected, in both D. melanogaster and D. immigrans embryo lysates a specific cleavage product was observed after incubation with a target specific siRNA (Figure 3A, lanes 2 and 7). In line with our earlier report [14], recombinant DmelNV VP1 protein potently inhibited cleavage of the target RNA in D. melanogaster embryo lysate, whereas the control, Maltose Binding Protein (MBP), was unable to do so (Figure 3A, compare lanes 3 and 4). In contrast, recombinant DimmNV VP1 protein did not inhibit slicer activity in D. melanogaster embryo lysate (Figure 3A, lane 5), which is in line with our observation that DimmNV VP1 did not suppress RNAi in cell-based reporter assays in D. melanogaster cells (Figure 2). Surprisingly, in the D. immigrans embryo lysate both the DmelNV VP1 and the DimmNV VP1 protein substantially inhibited target RNA cleavage (Figure 3A, lanes 9 and 10). Again, as expected, the MBP control protein did not inhibit slicer activity (Figure 3A, lane 8). Quantification of independent experiments indicates that both DmelNV and DimmNV VP1 proteins suppressed slicer activity to a similar extent in the D. immigrans embryo lysate (Figure 3B). These results, together with those from the cell-based reporter assays, indicate that DimmNV VP1 inhibits slicer activity in its natural host D. immigrans, but is unable to suppress RNAi in a heterologous D. melanogaster background. In contrast, DmelNV VP1 inhibits slicer activity in both a D. melanogaster and a D. immigrans background.
We recently showed that DmelNV VP1 inhibits RNA cleavage (slicer) activity of a pre-assembled RISC in D. melanogaster [14], suggesting that VP1 interacts with AGO2 to suppress its catalytic activity. To investigate a physical interaction between VP1 and AGO2, we analyzed DmelNV VP1 immunoprecipitations (IPs) for the presence of AGO2. To this end, we transfected S2 cells with a functional V5 epitope-tagged VP1 construct (V5-VP1) that encodes the C-terminal 124 amino acids of VP1 along with a FLAG-tagged AGO2 cDNA construct. Immunoprecipitation of V5-VP1 resulted in specific co-precipitation of the FLAG-AGO2 protein (Figure 4A). In contrast, the vector control failed to co-purify FLAG-AGO2. To confirm the interaction between VP1 and AGO2, we performed the reverse experiment. IP of FLAG-AGO2 protein co-precipitated V5-VP1, while a FLAG-control vector was unable to do so (Figure 4B). Although the interaction between VP1 and AGO2 is evident, only a minor fraction of VP1 was immunoprecipitated along with AGO2. This observation is in agreement with our microscopic analyses, in which only a small fraction of FLAG-AGO2 protein co-localizes with VP1-EGFP (data not shown). To confirm these results, we immunoprecipitated V5-VP1 protein and probed for endogenous AGO2 in the IP fraction. As expected, we observed a strong enrichment of endogenous AGO2 protein after VP1 IP, whereas IP of cells transfected with control plasmid did not co-precipitate AGO2 protein (Figure 4C). These results indicate that DmelNV VP1 interacts with Dmel AGO2 in Drosophila S2 cells.
These data and the results from our previous report [14] indicate that DmelNV VP1 interacts with Dmel AGO2 to antagonize the antiviral RNAi response. Similarly, given the observation that DimmNV VP1 suppresses slicer activity in D. immigrans lysates, it is likely that DimmNV VP1 interacts with Dimm AGO2. We hypothesized that the inability of DimmNV VP1 to suppress RNAi in D. melanogaster may then be due to an inefficient interaction with Dmel AGO2. To test these hypotheses, we analyzed VP1 interactions with host and non-host AGO2 proteins by co-IPs. First, we co-expressed V5 epitope-tagged DmelNV VP1 or DimmNV VP1 with Dmel FLAG-AGO2 in S2 cells and immunopurified the VP1 proteins using V5 affinity beads. As controls, we analyzed IPs of cells transfected with empty vector. As observed above (Figure 4), IP of DmelNV VP1 co-precipitated Dmel FLAG-AGO2 protein. In contrast, IP of DimmNV VP1 did not enrich Dmel FLAG-AGO2 in the IP fraction, compared to IP of the vector control (Figure 5A). To confirm these results, we analyzed the interaction between VP1 proteins and endogenous D. melanogaster AGO2. While DmelNV VP1, but not the control vector, co-precipitated endogenous Dmel AGO2 (Figure 4C, Figure 5B), DimmNV VP1 failed to co-IP endogenous Dmel AGO2, which mirrors our observation with epitope-tagged Dmel AGO2. These observations imply that the inability of DimmNV VP1 to suppress RNAi in D. melanogaster is due to its inability to efficiently interact with Dmel AGO2.
We next set out to analyze the interaction of DimmNV VP1 with Dimm AGO2. To this end, we cloned the D. immigrans AGO2 cDNA sequence downstream of the FLAG epitope (Dimm FLAG-AGO2). As expected, the predicted protein domains of Dimm FLAG-AGO2 are similar to those of Dmel AGO2, suggesting that the overall protein structure of Dimm and Dmel AGO2 are alike. Overall amino acid identity is 56% (63% when excluding the poly-glutamine repeats), with a higher level of conservation in the PIWI domain (77% identity) than in the PAZ domain (45% identity). We thus analyzed the interaction of DmelNV VP1 or the DimmNV VP1 with Dimm FLAG-AGO2 in co-IP. Both DmelNV VP1 and DimmNV VP1 efficiently co-purified the Dimm AGO2 protein (Figure 5C). These results show that AGO2-VP1 interactions correlate with RNAi suppressor activity: DmelNV VP1 interacts with both Dmel and Dimm AGO2 and suppresses slicer activity of these hosts; DimmNV VP1 interacts with Dimm AGO2, but not Dmel AGO2, and suppresses slicer activity in D. immigrans, but not in D. melanogaster.
The species-specific interaction of DimmNV VP1 with Dimm AGO2 suggests that this interaction is the major determinant for the observed species specificity in slicer activity. To test this hypothesis, we set out to reconstitute Dimm AGO2-based silencing in D. melanogaster S2 cells and to analyze whether DimmNV VP1 could suppress this reconstituted pathway. To this end, we reduced endogenous AGO2 expression in D. melanogaster S2 cells using RNAi, and rescued its activity with either a Dmel AGO2 or Dimm AGO2 cDNA construct.
First, we assessed the efficacy of knockdown of AGO2 expression in S2 cells using dsRNA targeting the coding sequence (CDS) or the 3′ untranslated region (3′ UTR) of the endogenous Dmel AGO2 transcript. To monitor AGO2 activity in these S2 cells we induced RNAi with the Rluc-specific RNA hairpin (described in Figure 2C). Compared to a non-specific dsRNA control, dsRNA against the CDS or the 3′UTR of AGO2 efficiently reduced hairpin-induced silencing of the Rluc reporter (Figure 6A). This experiment thus creates the opportunity to knock down endogenous AGO2 expression with UTR-targeting dsRNA and rescue silencing defects with Dmel AGO2 or Dimm AGO2 cDNA constructs that lack the AGO2 3′UTR sequence and are therefore not targeted by this RNAi approach. Strikingly, both Dmel AGO2 and Dimm AGO2 rescued silencing activity in D. melanogaster cells, whereas Dmel AGO1 only slightly increased silencing activity relative to the vector control (Figure 6B). These results indicate that Dimm AGO2 is fully functional in a D. melanogaster background and that the limited sequence identity to Dmel AGO2 does not impede its ability to interact with Dmel Dcr-2, R2D2 and other components of the D. melanogaster RISC complex.
Using this AGO2 rescue assay, we investigated whether DimmNV VP1 suppressed Dmel and Dimm AGO2-mediated silencing. DimmNV VP1 expression did not impede Dmel Ago2-mediated RNAi (Figure 6B), which is in line with our observations that DimmNV VP1 did not inhibit RNAi in D. melanogaster S2 cells (Figure 2A). In contrast, we observed that Dimm AGO2-mediated silencing was efficiently suppressed by DimmNV VP1 (Figure 6B). We were unable to analyze DmelNV VP1 in this assay, as its potent RNAi suppressive activity would impede silencing of endogenous Dmel AGO2, which is required for this assay.
Together, these results indicate that the interaction of VP1 with AGO2 is the major determinant for its RNAi suppressive activity. Moreover, these data imply that the VP1-AGO2 interaction is a major determinant for the species-specific effects of VP1.
Together, our data suggest that the interaction between viral RNAi suppressors and its cellular protein targets can be host specific. Thus, DimmNV VP1 suppresses AGO2-mediated silencing of its D. immigrans host, but not in non-host D. melanogaster; in contrast, DmelNV VP1 seems to be more promiscuous and inhibits AGO2-mediated RNAi in both D. melanogaster and D. immigrans. An exciting hypothesis is therefore that the species-specific interaction between VP1 and AGO2 can mediate host specificity of Drosophila Nora viruses. To test this hypothesis, we generated replication-competent Sindbis virus (SINV) recombinants expressing either DimmNV VP1, DmelNV VP1, or, as a control, GFP from a second subgenomic promoter (Figure 7A). As SINV is restricted by antiviral RNAi in Drosophila [14], [69], suppression of RNAi by expression of an exogenous viral RNAi suppressor is expected to yield higher viral RNA levels. Indeed, we previously showed that a DmelNV VP1 transgene renders SINV more pathogenic in D. melanogaster in an RNAi-dependent manner [14]. Our hypothesis thus predicts that the DimmNV VP1-expressing SINV recombinant reaches higher viral RNA levels than Sindbis-GFP in D. immigrans, but not in D. melanogaster, whereas Sindbis-DmelNV VP1 is expected to produce more viral RNA than SINV-GFP in both D. immigrans and D. melanogaster. We first verified stable expression of the VP1 transgenes by SINV recombinants by western blot (Figure 7B). Next, we analyzed whether SINV recombinants are equally replication competent in the C6/36 cell line that does not express functional Dicer-2. In this background, the presence of the VP1 transgene should not provide a replicative advantage over the GFP transgene. Indeed, VP1-expressing Sindbis virus recombinants replicated to slightly lower viral RNA levels than SINV-GFP in C6/36 cells (Figure 7C), indicating that none of the recombinant viruses suffer from major replication defects.
We next analyzed replication of SINV recombinants in D. melanogaster and D. immigrans hosts. As expected [14], in D. melanogaster the DmelNV VP1 transgene strongly increased viral RNA levels compared to SINV-GFP infection at 7 days post-infection (dpi) (Figure 7D, left panel). In general, D. immigrans only supported low levels of SINV replication. Nevertheless, in this host DmelNV VP1 increased SINV RNA levels, which is in line with our observation that this protein has RNAi suppressive activity in both hosts. The effects of DimmNV on viral RNA production also mirrored host specificity of its biochemical activity. Viral RNA levels of SINV-DimmNV VP1 were similar to SINV-GFP RNA levels in D. melanogaster (Figure 7D, left panel). In D. immigrans however, a strong increase in viral RNA levels was observed. Thus, DimmNV VP1 enhances viral RNA levels of recombinant Sindbis virus in a host species-specific manner, suggesting that the interaction of viral RNAi suppressors with AGO2 may be a determinant of host-specific pathogenicity.
Viruses and their hosts engage in an ongoing arms race in which viral counter-defense mechanisms drive the adaptive evolution of host immune genes, which in turn requires ongoing counter-adaptations in viral immune antagonists [3], [46]. This cycle of adaptation and counter-adaptation may result in species-specific interactions between virus and host [46], [70].
The antiviral RNAi genes R2D2, Dcr-2 and AGO2 belong to the 3% fastest evolving genes of Drosophila melanogaster and show evidence of positive selection in multiple species [47]. Strikingly, rapid evolution is observed in the antiviral RNAi pathway, whereas the microRNA pathway does not show evidence for rapid evolution. It is therefore possible that antagonistic host-parasite interactions – either through prolonged coevolution or through invasion by novel pathogens – are responsible for the observed rapid adaptive evolution in RNAi genes. Similarly, reciprocal antagonism between microbial pathogens and their hosts has been suggested to be the cause of positive selection observed in other insect immune genes, such as Relish and α-2-Macroglobulin [71]–[73].
Nora virus is a positive-sense RNA virus that was recently identified in laboratory stocks of Drosophila melanogaster [56]. Its unique genome organization and capsid structure suggests that Nora virus is the founding member of a novel virus family [57]. We report here that divergent Nora-like virus sequences are found in wild-caught D. immigrans and D. subobscura flies. Together with the recent isolation of Nora-like virus sequences from the horn fly Haematobia irritans and the parasitoid wasps Nasonia vitripennis and N. giraulti [59], [60] and the presence of Nora-like sequences in the transcriptomes of the lacewing Chrysopa pallens and the moth Spodoptera exigua (this report), our observations suggest that Nora virus is a member of a large family of widespread pathogens that infects multiple insect species.
Although little is known regarding the natural host range of Nora viruses, it is worth noting that neither of our population samples of D. immigrans or D. subobscura contained sequences derived from the other Nora lineages (i.e. DmelNV was not identified in D. immigrans or D. subobscura, and similarly for the other Nora-like viruses [DJO, unpublished data]), despite being initially collected as mixed samples of multiple Drosophila species. It is therefore possible that, as is the case for the purely vertically transmitted Sigma viruses, Nora viruses rarely move between hosts [74].
Plant and insect viruses can suppress the antiviral RNAi pathway of their hosts via a variety of mechanisms [11], [14], [29], [30], [35], [38], [43], [75]. We recently showed that Nora virus VP1 suppresses RNAi by inhibiting AGO2 slicer activity of a pre-assembled RISC [14]. Here we show that the RNAi suppressor activity of VP1 from Nora-like viruses can be host specific and that its RNAi suppressive activity correlates with its ability to interact with AGO2. DimmNV VP1 efficiently interacts with Dimm AGO2 and suppresses AGO2-mediated slicer activity in D. immigrans embryo lysates. In contrast, DimmNV VP1 was unable to suppress RNAi in D. melanogaster cells, did not interact with Dmel AGO2, and did not inhibit slicer activity in D. melanogaster embryo lysates. These results are consistent with a model in which adaption and co-evolution of DimmNV with its host resulted in a species-specific AGO2-VP1 interaction.
Our findings have important practical implications. Experimentally amenable model systems, such as Drosophila melanogaster or Arabidopsis thaliana, are often used to identify and characterize viral suppressors of RNAi, including those of viruses that naturally do not infect these hosts. Our observation that RNAi suppressor proteins may have species-specific activity suggests that it is important to take into account the correct evolutionary context in experiments aimed at the identification of viral suppressors of RNAi. For example, we note that we would not have detected RNAi suppressive activity in DimmNV, if we had solely relied on experiments in D. melanogaster.
In striking contrast to DimmNV, DmelNV VP1 did not show species-specific activity. It can engage in an interaction with both Dimm and Dmel AGO2 and, accordingly, it inhibited slicer activity in both D. immigrans and D. melanogaster embryo lysates. We suggest that there are two potential explanations for this. First, it may be that these viruses differ in natural host range; the broader-spectrum functionality of DmelNV VP1 across divergent hosts could be maintained by selection if DmelNV has a wider host range than DimmNV. In support of this hypothesis, although none of these three viruses was identified from the other host species, DmelNV (but not DimmNV) has been identified in wild Drosophila simulans (DJO, unpublished data). Second, if there is not a substantial trade-off associated with host-specialization and if DmelNV has colonized D. melanogaster quite recently, it could just be a matter of time until DmelNV loses its broad-spectrum VSR.
We successfully reconstituted Dimm AGO2-based silencing in D. melanogaster cells. This result suggests that the limited amino acid identity with Dmel AGO2 (∼63%) does not impede its ability to interact with Dmel Dicer-2 and R2D2 or other components of RISC and RISC-loading complexes. Thus, even though RNAi genes are rapidly evolving and show high rates of adaptive substitution, these results imply that this diversification has not impeded cross-species interactions of RNAi genes, even over the tens of millions of years that separate D. melanogaster and D. immigrans. This conservation of function may imply that the need for interaction between Dicer-2, R2D2, AGO2, and other RNAi pathway genes imposes a constraint on the evolution of these genes, and thus their opportunity to evolve in response to virus-mediated selection.
Together, our results suggest that rapid co-evolution between RNA viruses and their hosts may result in host species-specific activities of RNAi suppressor proteins. Moreover, our observation that DimmNV VP1 enhances viral RNA levels in a host-specific manner, suggest that viral RNAi suppressors are putative host-specificity factors.
Wild Drosophila populations were surveyed for the prevalence of Dmel Nora virus using RT-PCR (unpublished data; PCR primers: forward 5′-GACCATTGGCACAAATCACCATTTG-3′, reverse 5′-TCTTAGGCCGGTTGTCTTCACCC-3′), which resulted in the identification of Nora virus-like PCR products from D. immigrans and from members of the obscura group (sampled in Edinburgh, UK; longitude 55.928N, latitude 3.170W). A metagenomic approach was then used to obtain near-complete viral genomes. Flies were collected from elsewhere in the UK and samples were pooled by species for RNA extraction and Illumina double-stranded nuclease normalized RNA-sequencing. For D. subobscura, only male flies were used as females are difficult to distinguish morphologically from close relatives. RNA was extracted from each collection using a standard Trizol (Invitrogen) procedure, according to the manufacturer's instructions, and pooled in proportion to the number of contributing flies. In total, the two pools comprised 338 male D. subobscura (60 flies collected July 2011 Edinburgh 55.928N, 3.170W; 60 flies October 2011 Edinburgh 55.928N, 3.170W; 38 flies July 2011 Sussex 51.100N, 0.164E; 180 flies August 2011 Perthshire 56.316N, 3.790W) and 498 D. immigrans (63 flies, July 2011 Edinburgh 55.928N, 3.170W; 285 flies July 2011 Edinburgh N55.921, W3.193; 150 flies July 2011 Sussex 51.100N, 0.164E). Total RNA was provided to the Beijing Genomics Institute (Hong Kong) for normalization and 90-nt paired-end Illumina sequencing. Paired-end reads were quality trimmed using ConDeTri version 2 [76] and assembled de novo using the Trinity transcriptome assembler with default settings (r2011-08-20, ref. [77]). We used tBlastn with a DmelNV protein query to identify two partially overlapping Nora-like contigs from D. immigrans, and a single contig from D. subobscura. Quality-trimmed paired-end reads were mapped back to these contigs using Stampy (version 1.0.21, ref. [78]) to obtain a consensus sequence, based on majority calls at each position. In total, 286,242 reads mapped to DimmNV (0.45% of all reads derived from D. immigrans, median read depth 1200-fold) and 68,914 reads mapped to DsubNV (0.13% of all reads derived from D. subobscura, median read depth 133-fold). Consensus sequences have been submitted to GenBank under accession numbers KF242510 (DsubNV) and KF242511 (DimmNV).
The relationship between DmelNV (GenBank NC_007919.3; [57]), DsubNV, DimmNV and other Nora-like sequences was inferred from VP4 (capsid protein), which is the most conserved gene and the one with the most coverage in the non-Drosophila sequences. The other Nora-like sequences included Nasonia vitripennis Nora-like virus (GenBank FJ790488; [60]), Haematobia irritans Nora-like virus (GenBank HO004689, HO000459, and HO000794; [59]), and two Nora-like sequences newly identified here in the transcriptomes of Spodoptera exigua (GenBank GAOR01000957; [79]) and Chrysopa pallens (GenBank GAGF01018485; [80]). We excluded sequences virtually identical to DmelNV that appear in the transcriptomes of Leptopilina boulardi and Leptopilina heterotoma (GenBank GAJA01006738, GAJC01010128 and GAJA01017939; [81]), as these species are widely cultured on D. melanogaster in the laboratory. For protein alignment, see text S1.
For the N. vitripennis Nora-like virus we selected the longest sequence (FJ790488) for analysis. Two approaches to phylogenetic inference were used. First, MrBayes (v3.2.1, ref. [82]) with discrete gamma-distributed rate variation and model-jumping between amino acid substitution models. Two parallel runs of four heated chains were used, and convergence was assessed by examination of the potential scale reduction factor (PSRF) and the variance in split-frequencies between runs (PSRF ∼1 for all parameters; variance in split-frequencies <0.001). Second, a maximum-likelihood analysis was run using PhyML [83] under a WAG amino-acid substitution model [84] with discrete gamma-distributed rate variation. Data were bootstrapped 1000 times to infer bootstrap node-support. The nonsynonymous divergence along each of the branches leading to DmelNV, DsubNV, and DimmNV was inferred using the method of Li [85], relative to an ancestral sequence inferred by maximum likelihood using PAML [61]. Sliding windows of 50 codons wide were placed every 30 codons. Nominal genome-wide ‘significance’ thresholds for peaks were derived by repeating the sliding-window analysis on 1000 randomizations of codon-position order.
The following constructs were described previously: all DmelNV VP1 constructs [14], pAFW-AGO1 and pAFW-AGO2 [86], pAFW (Drosophila Genomics Resource Center, https://dgrc.cgb.indiana.edu), pMT-Luc [38], pMT-Rluc [38], pRmHa-Renilla-hairpin [87], pAc5-V5-His-A (Invitrogen), and pAc5-V5-His-Ntag [14].
cDNA of D. immigrans and D. subobscura was made using Promega MMLV-RT in the presence of Promega RNasin Plus according to manufacturer's instructions. Subsequently, DimmNV VP1 and DsubNV VP1 sequences were PCR amplified from D. immigrans and D. subobscura cDNA and cloned as full-length and deletion constructs downstream of the V5-His tag in pAc5-V5-His-Ntag (details available upon request).
The D. immigrans AGO2 cDNA sequence (GenBank KF362118), including partial 5′ and 3′ UTRs, was PCR amplified using the primer pair 5′-TGCAGCAAAAATTAGAAGCAAA-3′ and 5′-AGCCGTACCTAGAACCAGCA-3′. The resulting PCR product was used as a template in a nested PCR using primer pair 5′-AGTTCTAGACCGCGGGAATGGGTAAAAAGAACAAGTTCAAACCA-3′ and 5′-AGTTCTAGACCGCGGGAAGCGCTGTGGCACAGCTTCCGC-3′. The nested PCR product was subsequently cloned into the pAFW vector using the SacII and SalI restriction sites. To fuse the DimmNV VP1ΔN295 protein to the C-terminus of the maltose binding protein (MBP), we PCR amplified the VP1 coding sequence from pAc5.1-Ntag-DimmNV VP1FL with primer pair 5′-AGTGGATCCCCAAAACTTCCAAGTGTACCTTCAAAG -3′ and 5′-GGTGTCGACTTAGTTTTGTTTATTTTTGTACCAATCGTTGG -3′. The DsubNV VP1ΔN281 sequence was amplified from pAc5.1-Ntag-DsubNV VP1FL with primer pair 5′-TGACGGATCCCCAAACAAACCTCTAAAACC -3′ and 5′-ACTGGTCGACTCATTGTTGCTGAGTTGATTTG -3′. The resulting PCR products were cloned into the pMal-C2X vector (New England Biolabs) using BamHI and SalI restriction sites.
Double-stranded RNA was generated by in vitro transcription using T7 promoter-flanked PCR fragments as a template, as described previously [88]. For production of AGO2 dsRNA, a fragment of the coding sequence or the 3′ untranslated region of Dmel AGO2 was PCR amplified using primer combination 5′-TAATACGACTCACTATAGGGAGATACTATGGTGAAGAACGGGTCG-3′ and 5′-TAATACGACTCACTATAGGGAGAGAACATGTCCTCAATCTCCTCC-3′, or primer combination 5′-TAATACGACTCACTATAGGGAGAGCAACGTATTGAATCTTATT-3′ and 5′-TAATACGACTCACTATAGGGAGAAGAACAATATTTGGCGGACC-3′, respectively.
miRNA and RNAi sensor assays in Drosophila S2 cells were performed as described [14], [88]. For hairpin-induced silencing of the Rluc reporter, 5×104 S2 cells were seeded per well in a 96-well plate. The seeded cells were co-transfected with 10 ng pMT-Fluc, 10 ng pMT-Rluc, 50 ng pRmHa-Renilla-hairpin, and 50 ng of expression plasmids encoding VP1 and/or AGO per well using Effectene transfection reagent (Qiagen). The pAc5-Ntag-DmelNV VP1Δ284 and pAc5-Ntag-DimmNV VP1ΔN295 plasmids were used for VP1 expression. For knockdown of endogenous AGO2, 5 ng of AGO2 dsRNA or control dsRNA was co-transfected along with reporter plasmids. Two days after transfection, the expression of the luciferase reporters and the Rluc hairpin was induced by the addition of 0.5 mM CuSO4 per well. The next day, cells were lysed and Fluc and Rluc activity was measured with the Dual luciferase reporter assay system (Promega) according to manufacturer's protocol.
For immunoprecipitations, S2 cells were seeded in 6-well plates at a density of 2×106 cells per well. The next day, cells were transfected with AGO2 and/or VP1 expression plasmids using Effectene transfection reagent (Qiagen). Expression plasmids encoding DmelNV VP1ΔN351, DmelNV VP1ΔN284, or DimmNV VP1ΔN295 were used for co-immunoprecipitation experiments, as indicated in the figure legends. Three days after transfection, cells were washed twice with PBS and subsequently resuspended in lysis buffer (30 mM HEPES-KOH, 150 mM NaCl, 2 mM Mg(OAc), 0.1% NP-40, 5 mM DTT) supplemented with protease inhibitor cocktail (Roche). After incubation on ice for 10 minutes, the samples were passed forty times through a 25-gauge needle, followed by incubation on ice for 10 minutes. Subsequently, cell lysates were centrifuged at 13,000 rpm for 30 minutes and a sample of the supernatant was taken to analyze the input for IP. To remove proteins that non-specifically bind to the IP beads, the remaining supernatant was incubated with Pierce protein G agarose at 4°C for 5 hours while mixing end-over-end. Next, the protein G agarose was separated from the supernatant by centrifugation, after which the supernatant was incubated overnight with anti-V5 agarose affinity gel (Invitrogen) at 4°C while mixing end-over-end. The next day, the anti-V5 agarose was separated from the supernatant by centrifugation, and a sample was taken from the supernatant. After the remaining supernatant was removed, the V5-agarose was washed three times with lysis buffer, and three times with either wash buffer 150 (25 mM Tris-Cl, 150 mM NaCl) or wash buffer 200 (25 mM Tris-Cl, 200 mM NaCl). All wash steps were done with 40 to 60 times beads-volume of wash buffer. Subsequently, the beads were boiled in SDS sample buffer at 95°C for 10 minutes, followed by a brief centrifugation step to collect the beads at the bottom of the tube. The proteins in the supernatant were then separated on a SDS-PAGE gel, after which they were transferred onto a nitrocellulose membrane by western blot. Primary antibodies used for western blot detection were anti-FLAG-M2 (1∶1000 dilution; Sigma), anti-V5 (1∶5000 dilution; Invitrogen), anti-AGO2 (1∶500 dilution; generously provided by the Siomi lab), and anti-tubulin-alpha (1∶1000 dilution, Sanbio); secondary antibodies were goat anti-mouse-IRdye680 (1∶15,000 dilution; LI-COR), and goat anti-rabbit-IRdye800 (1∶15,000 dilution; LI-COR). All western blots were scanned using an Odyssey infrared imager (LI-COR biosciences).
To purify recombinant VP1 as MBP fusion proteins, the pMal-C2X-DimmNV VP1ΔN295 and the pMal-C2X-DsubNV VP1ΔN281 plasmids were transformed into the Escherichia coli BL21 (DE3) strain. Subsequently, expression of recombinant protein was induced by addition of 0.2 mM IPTG. Protein expression was allowed to proceed overnight at 18°C. The next day, recombinant MBP-DimmNV VP1 and MBP-DsubNV VP1 were purified using amylose resin (New England Biolabs) according to the manufacturer's protocol. Purified protein was subsequently transferred to a dialysis membrane (molecular weight cut-off 12–14 kDa) and incubated overnight in dialysis buffer (20 mM Tris-Cl, 0.5 mM EDTA, 5 mM MgCl2, 1 mM DTT, 140 mM NaCl, 2.7 mM KCl) at 4°C, followed by a second dialysis step for 5 hours at 4°C. The dialyzed protein solution was stored at −80°C in dialysis buffer containing 30% glycerol. Purification of MBP-DmelNV VP1ΔN284 has been described previously [14].
A new D. immigrans isofemale line was established from flies collected in June 2012 in Edinburgh (Coordinates 55.921N, 3.193W). D. immigrans was cultured similarly as D. melanogaster on standard media. Embryo lysates were generated from D. immigrans and from an RNAi-competent D. melanogaster laboratory control strain (w1118). In vitro target RNA cleavage assays in D. melanogaster embryo lysates were performed as described [14]. Minor changes were incorporated for the slicer assay in D. immigrans embryo lysate: the reaction contained 0.9 mM MgCl2 and was allowed to proceed for 5 hours at 25°C before RNA extraction. Suppressor activities of MBP-DmelNV VP1ΔN284, DsubNV VP1ΔN281, and MBP-DimmNV VP1ΔN295 proteins were analyzed in slicer assays.
To produce recombinant Sindbis viruses, N-terminal V5 tagged DmelNV VP1ΔN284 and DimmNV VP1ΔN295 were PCR amplified from the respective insect expression vectors using primers V5 Fw: AGTTCTAGAAACATGGGTAAGCCTATCC; Dmel VP1 Rv: GGTTCTAGATTAACATTGTTGTTTCTGCGAG; and Dimm VP1 Rv: TGACTCTAGATTAGTTTTGTTTATTTTTGTACC. PCR products were cloned into the XbaI site following the second subgenomic promoter of the pTE3'2J vector [89]. The resulting plasmids were linearized with XhoI, and in vitro transcribed using the mMESSAGE mMACHINE SP6 High Yield Capped RNA Transcription kit (Ambion). Transcribed RNA was then purified using the RNeasy kit (Qiagen) and transfected into BHK-21 cells to produce infectious virus. Supernatant was harvested and titered by plaque assay on BHK-21 cells. Sindbis-GFP was described previously [69].
The replicative capacity of recombinant viruses was analyzed on Dicer-2 deficient C6/36 cells. The cells were cultures as described previously [90] and inoculated at an multiplicity of infection of 0.01. Cells were harvested directly after inoculation (t = 0) and at 24 h thereafter and total RNA was isolated using isol-RNA lysis reagent (5 Prime). The RNA was treated with DNaseI and used as template for cDNA synthesis using Taqman reverse transcription reagents (Roche). Viral RNA levels were determined by qPCR using the GoTaq qPCR Master Mix (Promega) and primers for either Sindbis (SINV NS4 Fw: AACTCTGCCACAGATCAGCC; SINV NS4 Rv: GGGGCAGAAGGTTGCAGTAT) and Aedes Albopictus RpL5 for normalization (Aalb RpL5 Fw TCGCTTACGCCCGCATTGAGGGTGAT; Aalb RpL5 Rv: TCGCCGGTCACATCGGTACAGCCA).
Flies (Drosophila melanogaster w1118 and Drosophila immigrans) were grown on standard yeast/agar medium at 25°C on a 12-h light/dark cycle. Flies were cured of Wolbachia sp. by tetracycline treatments as described in [91]. Five to seven-day-old female flies were CO2-anesthetized and intrathoracical single injections of 50.6 nL, corresponding to 5,000 plaque forming units for each virus, were performed using a nanoinjector Nanoject II (Drummond Scientific Company) as described in [92].
For each time point, total RNA from three independent pools of three flies was isolated using TRIzol Reagent (Life Technologies). RNase-free DNase I treatment (Roche) was performed according to manufacturer's instructions, followed by acid-phenol/chloroform (Life Technologies) inactivation. Total RNA was quantified using a ND-1000 NanoDrop spectrophotometer (Thermo Fisher Scientific). Reverse transcription was performed using SuperScript II Reverse Transcriptase with random hexamers as primers (Life Technologies) on 2 µg of total RNA. Quantitative PCR was performed with three technical replicates for each cDNA sample using FastStart SYBR Green Master (Rox) (Roche) on a ViiA7 Real-Time PCR instrument (Life Technologies). As negative controls, cDNA reactions without reverse transcriptase and PCR amplification without cDNA template were included. Oligonucleotide primers were as follows (F, forward; R, reverse) Sindbis virus: SINV-NSP3_F, AAAACGCCTACCATGCAGTG; SINV-NSP3_R, TTTTCCGGCTGCGTAAATGC, and for normalization Dimm-AGO2_F, TTTTGTGCTGGGCGACAAAC; Dimm-AGO2_R, ATTCACCGCTTCGCAAATCG and Dmel-RpL32_F, CGGATCGATATGCTAAGCTGT; Dmel-RpL32_R, GCGCTTGTTCGATCCGTA. Relative viral RNA levels were calculated using the 2−ΔΔCT method [93] relative to input viral RNA, determined in flies that were harvested immediately after inoculation. Following log-transformation to homogenize variances, a T-test was used to compare relative RNA levels in SINV-VP1 recombinants to those in SINV-GFP.
D. immigrans AGO2 cDNA sequence: KF362118; DsubNV consensus sequence: KF242510; DimmNV consensus sequence: KF242511; DmelNV: NC_007919.3; Nasonia vitripennis Nora-like virus: FJ790488; Haematobia irritans Nora-like virus: HO004689, HO000459, and HO000794; Transcriptome of Spodoptera exigua: GAOR01000957; Transcriptome of Chrysopa pallens: GAGF01018485.
|
10.1371/journal.ppat.1006708 | Selection for avian leukosis virus integration sites determines the clonal progression of B-cell lymphomas | Avian leukosis virus (ALV) is a simple retrovirus that causes a wide range of tumors in chickens, the most common of which are B-cell lymphomas. The viral genome integrates into the host genome and uses its strong promoter and enhancer sequences to alter the expression of nearby genes, frequently inducing tumors. In this study, we compare the preferences for ALV integration sites in cultured cells and in tumors, by analysis of over 87,000 unique integration sites. In tissue culture we observed integration was relatively random with slight preferences for genes, transcription start sites and CpG islands. We also observed a preference for integrations in or near expressed and spliced genes. The integration pattern in cultured cells changed over the course of selection for oncogenic characteristics in tumors. In comparison to tissue culture, ALV integrations are more highly selected for proximity to transcription start sites in tumors. There is also a significant selection of ALV integrations away from CpG islands in the highly clonally expanded cells in tumors. Additionally, we utilized a high throughput method to quantify the magnitude of clonality in different stages of tumorigenesis. An ALV-induced tumor carries between 700 and 3000 unique integrations, with an average of 2.3 to 4 copies of proviral DNA per infected cell. We observed increasing tumor clonality during progression of B-cell lymphomas and identified gene players (especially TERT and MYB) and biological processes involved in tumor progression.
| The Avian Leukosis Virus (ALV) is a simple retrovirus that causes cancer in chickens. The virus integrates its genome into the host genome and induces changes in expression of nearby genes. Here, we determine the sites of viral integrations and their role in the progression of tumors. We report pathways and novel gene players that might cooperate and play a role in the progression of B-cell lymphomas. Our study provides new insights into the changes during lymphoma initiation, progression, and metastasis, as a result of selection for specific ALV integration sites.
| Avian leukosis virus (ALV) is a simple retrovirus that causes cancer, primarily B-cell lymphomas in chickens [1–3]. The ALV genome does not contain a viral oncogene and induces aberrant host gene expression via use of strong viral enhancer and promoter elements. Relative to other well studied retroviruses like HIV-1 and MLV, ALV was shown to integrate relatively randomly into the host genomic DNA, with little bias for genomic features [4–7]. The FACT (facilitates chromatin transcription) complex, a chromatin remodeler, was recently reported to promote ALV integration [8].
ALV-induced lymphomas develop in a multistage process, appearing initially as neoplastic follicles in the bursa, some of which develop into primary bursal tumors. Primary tumors can then metastasize to distant organs and form secondary tumors [9]. We are studying rapid-onset lymphomas, which develop in less than 3 months after infection of chicken embryos with ALV [10,11].
Cellular transformation occurs through multiple genetic changes in oncogenes and tumor suppressor genes, as well as noncoding RNAs [12]. These oncogenic changes can occur via different genetic mechanisms; insertional mutagenesis by retroviruses is one such mechanism. Viral integration into the host genome can alter host gene expression and induce cancer development [1–3]. In turn, common viral integration targets observed in multiple tumors can help identify oncogenes [2,13–19]. Consequently, retroviral-mediated lymphomagenesis in chickens provides an excellent experimental model system for analysis of neoplastic change in tumors of B-cell lineage [20]. We have previously identified common proviral integrations in ALV-induced B-cell lymphomas, notably in the TERT promoter region, and in hemangiomas [15–17].
Since selection in tumorigenesis alters the pattern of viral integration sites, we also analyze integrations in cultured cells, to identify preferences of ALV integrations in an unbiased way. We investigate how ALV integrations in tissue culture correlate with previously unreported genomic features. We observe an enrichment of ALV integrations in the 5’ end of gene bodies, proximal to CpG islands and transcription start sites, as well as a preference for expressed and highly spliced genes. No association was observed with levels of alternative splicing. In order to determine the effects of selection for oncogenic characteristics, we compare integration sites in cultured cells with those in ALV-induced B-cell lymphomas. We observe that ALV integrations are more strongly selected for proximity to transcription start sites in tumors. In addition, there are fewer integrations near CpG islands in tumors.
ALV tumors are thought to be clonal, as determined by previous work [9,15]. However, the clonality of these neoplasms has not been empirically defined. We analyze ALV infection in tumors by quantifying the clonal abundance and distribution of integrations during progression of tumors. Using the statistical Gini index, we calculate the empirical degree of oligoclonality and extent of clonal expansion in different stages of tumorigenesis [21,22]. Quantifying the clonality index and average number of integrations per cell within individual tumors helps determine the clonal architecture and hierarchies of lymphomagenesis. Furthermore, the gene ontology analysis of host genes most proximal to proviral sites provides insight into underlying gene players and their contribution to oncogenic transformation. Thus, our work helps unravel how the integration sites of ALV are selected for in oncogenesis and play a role in the clonal progression of tumors. This is the most in depth analysis for ALV integration sites to date and is novel in terms of being able to follow the patterns of integration from early infection (in tissue culture) through to early and late tumor development.
Via deep sequencing, we analyzed approximately 87,000 unique ALV integration sites (UISs) in tissue culture cells and in tumors, as summarized in S1 Table. Randomly generated integration sites were used as a control for all subsequent analysis. ALV integrations were analyzed for different ALV subgroups (A, C and J) in different infected cell types, including primary chicken embryo fibroblasts (CEF), DT-40, a chicken B-cell lymphoma cell line, and the human HeLa tumor cell line. After mapping the UISs to the host genome, we used the HOMER bioinformatics tool [23] to associate integrations with specific annotated genomic sites (Table 1).
Upon analysis of ALV integrations in CEFs, using the ensembl Gallus gallus 4 genome, we observed a significant bias for ALV integration into genes (approximately 40%) relative to random events (27%) (t test, p-value 0.006) (Fig 1A). We also observed a slight enrichment for integrations near LINE sequences, gene promoters, simple repeat and satellite DNA sequences; however, these were not statistically significant (Table 1). Independent analysis of all these features was consistent for infections with ALV in CEFs, DT-40 chicken B cells and in HeLa cells, suggesting that these preferences are not cell type specific (Table 1).
To study selection of ALV integration sites in tumors, we sequenced 72 tissue samples from 41 different birds (S2 Table). We obtained 17.2 million reads, originating from viral integrations in neoplasms and non-tumor tissues, which were mapped to 71,368 UISs. Similar to the integration pattern in cultured cells, integrations in the ALV-induced tumors, at the primary sites in the bursa or secondary metastases, showed a significant enrichment for genes (approximately 38%), relative to random (27%) (t test, p-value 0.022) (Fig 1A, Table 1).
We analyzed the distribution of ALV integrations within transcriptional units by dividing the gene bodies into 10 equal segments or bins. Then, we calculated the number of integrations within each bin to determine the density of ALV integrations within a given part of a transcriptional unit. Relative to the matched control set (11.99%), there was a significant enrichment of ALV integrations towards the 5’ end of the gene body. This bias is most distinct within the first 10% of the gene body, i.e. in proximity to the transcriptional start site, in both tissue culture (16.25%) (t test, p-value 0.031), and in tumors (16.22%) (t test, p-value 0.004) (Fig 1B).
To determine the pattern of integrations surrounding transcription start sites (TSS), we plotted the observed ALV integrations in cultured cells and in tumors with respect to the nearest TSS, extending over 15 kb on either end (Fig 2A). Relative to the random integration sites within 5 kb of the TSS, we observed a nearly 2-fold enrichment of ALV integration sites in tissue culture (t test, p-value 0.042) (Fig 2B). We observed an even greater frequency of integrations near TSS in tumors, with nearly 3-fold enrichment relative to random events within the 5 kb window (t test, p-value 0.006). This suggests that clonal selection of integration sites in tumors has a strong bias for proximity to TSS, probably to promote induction of aberrant gene expression in tumorigenesis. In addition, consistent with previously reported data from our lab [17], we observed a pronounced drop in the integration frequency in the vicinity of the TSS in tumors (Fig 2A). Interestingly, we also observed this drop in tissue culture (Fig 2A), suggesting it is a result of integration preference and not due to selection in tumors.
Based on our initial HOMER analysis, we observed a 3-fold preference for integrations within CpG islands (approximately 3%) relative to random sites (1.1%) in cultured cells (t test, p-value 0.041) (Fig 3A). In contrast to tissue culture, the percentage of integrations within CpG islands was not enriched in tumors (1.4%), and appeared near random. The same was true for the most clonally expanded integrations (0.96%), with 10 or more breakpoints (see Materials and methods). To further investigate this, we determined the frequency of ALV integrations in the area immediately surrounding CpG islands (Fig 3B). Analyzing ALV integrations in tissue culture, we found that in the 1 kb region surrounding CpG islands, there was a 1.5 fold enrichment of integration relative to random (t test, p-value 0.047). If the window is expanded to 5 kb flanking the CpG island, the enrichment becomes more pronounced. Nearly 33% of all integrations are observed within 5kb of CpG islands, which is a 1.7 fold enrichment relative to the matched random control (21%) (t test, p-value 0.043).
In contrast, this enrichment for integration near CpG islands in cultured cells was not observed in tumors (Fig 3B). Frequency of integration within 1 or 5 kb of CpG islands in tumors was not significant relative to random events. When the same analysis was done for only the most clonally expanded integrations, there is a striking depletion of integrations within 5kb of CpG islands (11.8%) relative to the total integration set in tumors (27.7%) as well as a matched random control (21%) (Fig 3B). This suggests that in tumors, there is an enrichment of integrations away from CpG islands (t test, p-value 0.045).
We next investigated ALV integration as a function of expression levels of the most proximal transcriptional units. In order to determine the background expression levels of host genes, we analyzed RNA-seq data for CEF, DT-40 and HeLa cells. We divided the whole transcriptome into 13 bins and determined the percentage of integrations within each bin, to observe any enrichment above background. While nearly 22.5% of the chicken genes are not expressed in CEFs, only 8.2% of integrations occur in this expression bin. Thus, we observe a significant depletion in the percentage of integrations within or in proximity to unexpressed genes (t test, p-value 0.006) (Fig 4). For expressed genes, there was no preference observed for ALV integration with the level of gene expression, relative to random events. Thus, ALV integrates randomly in proximity to genes with low, intermediate or high levels of expression.
Since integrations can occur in intergenic regions at distant loci from transcriptional units, we asked whether ALV integrations within the transcriptional unit might exhibit a bias towards expressed genes. To address this, we repeated our analysis for only those integrations that occur within the gene body (between the transcription start and termination sites). Overall, we observed a similar integration pattern relative to random, with a more pronounced depletion of integrations within unexpressed genes. In contrast to random events (25.4%), there was a nearly 6-fold decrease in preference for ALV integrations in genes with no expression (3.9%) (t test, p-value 0.003) (S1 Fig). Therefore, while ALV preferentially integrates near or within expressed genes, there is no preference for the level of gene expression.
We also correlated ALV integrations with the gene expression levels via analysis of RNA-seq data for a subset of the ALV-induced lymphomas. Similar to our findings in cultured cells, we observed that in tumors ALV integrations are selected for proximity to or within expressed genes, but there is no distinct bias for varying gene expression levels (S2 Fig).
HIV has been previously reported to preferentially integrate into genes that are highly spliced, i.e. have a greater number of introns [24]. In order to determine whether ALV might display any similar preferences, we correlated ALV integration with the levels of mRNA splicing and alternative splicing. We associated the percent of integrations with the number of introns of the most proximal transcriptional unit, using the ensembl Gallus gallus 4 genome. While by random chance 12.1% of integrations are predicted to occur in unspliced transcriptional units, only 7.7% (t test, p-value 0.004) and 6.4% (t test, p-value 0.002) of ALV integrations in cultured cells and tumors, respectively, occurred within this range. Therefore, there is a significant lack of integration into unspliced genes in tumors. By random chance, the vast majority of the integrations are predicted to occur in transcriptional units with 1–19 introns (71.1%). For this window of splicing, ALV integrations in tissue culture (68.3%) appear close to random as well. On the other hand, nearly 22.1% of ALV integrations in tissue culture fall into highly spliced genes with 20 or more introns, which is a significant enrichment above random events (16.6%) (t test, p-value 0.012). Furthermore, ALV integrations in tumors (47.1%) have a greater enrichment for integration within or proximity to transcriptional units with 10 to 39 introns, relative to random sites (39.3%) (t test, p-value 0.026). Thus, we observe that ALV has a bias for integration into spliced genes with enrichment for higher levels of gene splicing (Fig 5).
We also asked whether the levels of ALV integration are associated with the number of spliced isoforms of the proximal transcription unit. The chicken genome is not well characterized for reporting alternative spliced variants of genes. Since the human genome is better characterized than the chicken genome, we repeated our analysis for ALV integrations in a HeLa cell line (S3 Fig). However, our analysis did not identify any preference for integrations relative to the level of alternative splicing of proximal transcriptional units.
In order to measure the relative abundance, or clonal expansion, of UISs within a tissue, we quantified the number of sonication breakpoints for each site, as described previously [16]. The number of breakpoints reveals the extent of clonal expansion of the UIS. In tissue culture (2–5 days after infection) we identified 16,978 unique sonication breakpoints, i.e. an average of 1.1 breakpoints per integration site (Fig 6A). The vast majority of these integrations (82.9%) had a single breakpoint, suggesting that these integrations were not clonally expanded. On the other hand, in tumors we identified 92,951 unique sonication breakpoints. The average number of breakpoints per integration was 1.3, with the vast majority of integrations (67.6%) showing only a single sonication breakpoint. In contrast to 17.1% of integrations in tissue culture, 32.3% of integrations in tumors had two or more breakpoints, revealing that a large fraction of the infected cells are from expanded clones. Moreover, a large number of integrations (approximately 13,000) in tumors had a very high number (10 or more) of breakpoints.
The distribution of viral integration sites in the neoplasms is depicted in a composite pie chart (Fig 6B). The most highly expanded clones, each with 70 or more breakpoints, are highlighted in a table. The table depicts 28 UISs, indicating the gene most proximal to the integration site, respective tumor, and its corresponding number of breakpoints. The maximum observable number of breakpoints is limited by the length of deep sequencing reads and, in the case of highly abundant integration sites, the probability of repeated sonication at the same genomic position. Thus, it is important to note that our standard breakpoint analysis is an underestimate of the fraction of the infected cells with expanded clones [21].
Cancers exist in a number of stages, characterized by a spectrum of divergent cells and genetic changes. Each cancer is unique, and in any given tumor the clonal structure shifts over time, which involves the clonal selection and expansion of cells. [25]. In cases of ALV-induced B-cell lymphomas, the bursa serves as the primary organ of malignant transformation and site of tumorigenesis [9,26]. Infected chickens typically develop multiple primary neoplastic follicles in the bursa, some of which may eventually form primary tumors. The development of these neoplasms is a multi-stage process [26–28]. In order to examine the clonality of lymphomagenesis, we studied different stages of cancer progression. These include inflammation, neoplastic follicles, primary tumors in the bursa, and metastatic tumors at secondary sites. The stage of neoplastic follicles in the bursa is an early step in tumor progression towards transformation and malignancy [26]. Metastases of primary tumors are observed from the bursa to secondary sites in the liver, spleen, and kidneys.
Analyzing the different neoplasms, we observed an increasing extent of clonality with the advancing stages of tumorigenesis (S4 Fig). This clonal expansion can be represented by a pie chart, where each pie represents an individual tumor. A given slice of a pie represents a UIS, and the size of the slice corresponds to its relative clonal abundance, denoted by the respective number of sonication breakpoints for that UIS.
The most clonally expanded integrations in secondary tumors can be investigated by comparison of metastasized versus non-metastasized neoplasms within an individual bird (Fig 6C). For example, the different tumors in bird A2, all harbor overlapping UISs with an increasing level of clonal expansion in the metastasized neoplasms. For example, the expanded clones of UISs at TAB2, BTBD1 and mir-30a integrations appear in a mix of other clonally expanded integrations in the primary bursa tumor, whereas the secondary liver, kidney and spleen tumors are more homogenous. This suggests increasing clonal homogeneity of the metastasized tumors relative to the bursa. Additional examples of comparisons between primary and secondary neoplasms within an individual bird are depicted in S5 Fig.
In order to empirically estimate the clonality of different neoplasms we made use of an objective parameter called the oligoclonality index (OCI) as defined by the Gini co-efficient index [21,29]. The OCI defines the clonal abundance of a tissue on an objective scale of 0 to 1. In theory, a tissue with perfect monoclonality would have an OCI value of 1. Conversely, an entirely polyclonal tissue would have an OCI value of 0. The OCI values of the neoplastic subtypes, in representative stages of lymphomagenesis, are depicted in Fig 7A. The plot represents the OCI values in ascending order with tumor progression, suggesting an increased magnitude of clonal expansion within these neoplasms. As was previously depicted by the pie charts of bursal tumors (Fig 6C, S4 Fig), the OCI value is lower in these samples compared to the metastases. The OCI for the metastasized tumors was significantly greater, in some cases close to 1.
Additionally, in order to further validate the OCI values, we investigated the ALV integrations and their extent of clonal expansion in different slices of representative neoplasms (Fig 7B). A sample with high clonal homogeneity would exhibit a highly uniform pattern of integrations and corresponding clonal abundance in different slices of the tumor, as depicted for tumor D2L. Conversely, a sample with lower OCI value such as a bursa tumor exhibited more clonal heterogeneity in different portions of the neoplasm (S6 Fig). Additional data for integration patterns in slices of other neoplasms is summarized in S6 Fig.
In addition to the extent of clonal expansion, we also wanted to investigate the average number of proviral integrations within tumors, defined as its proviral load (PVL). Twenty randomly selected tumors across different stages of lymphomagenesis were isolated for this analysis. The PVL varies widely between the tissues, ranging on an average from approximately 2.3 to 4 copies per infected cell of a tumor (Fig 8). We observed a correlation between the PVL and different stages of tumor progression, suggesting that a higher PVL is associated with later stages of tumorigenesis.
ALV env genes are more distinct and offer greater specificity to distinguish ALV proviral genomes from endogenous retrovirus genomes in the host chicken genome. However, proviruses undergo varying amounts of genome rearrangements and deletions during cellular transformation and oncogenesis, especially in the env region, probably to evade cellular immune surveillance. Therefore, we utilized LTR specific primers to estimate the PVL. We assume each provirus contains 2-LTRs therefore, the PVL values described here may be underestimates of the actual PVL among infected neoplasms due to the possibility of solo LTRs. Additional analyses of PVL estimates, via use of gag or env regions of ALV, exhibit similar association of PVL with the progression of tumors (S7 Fig).
Cancers acquire, via mutational and epigenetic changes, a variety of traits that trigger clonal expansion, via proliferation, migration and invasion. These properties are alterations to normal developmental and physiological cellular processes [25]. We wanted to further investigate how ALV integrations near host genes, in certain functional categories, confer oncogenic advantages on the infected B-cell clone. To test this, we used the G:profiler analysis software to analyze the ontology of the nearest host gene upstream or downstream of each integration site. G:profiler allows functional profiling of genes within a neoplasm, in a quantitative fashion [30]. Queries are ordered with more importance given to the more expanded clones within tumors. The results showed a significant overrepresentation of genes in five cellular pathways: cell differentiation, phosphorylation, immune response signaling, proliferation and regulation of apoptosis (Fig 9). These biological processes show a greater enrichment in the secondary tumors (malignant) than in neoplasms with low or intermediate clonality.
These changes reflect the temporal order of genetic alterations and underlying cellular processes acquired in the progression of B-cell lymphomagenesis. These processes were associated with the host gene nearest to the viral integration sites, regardless of its transcriptional orientation relative to the provirus. The genes associated with these processes are listed in S3 Table. Furthermore, when we analyzed the GO terms associated with the most clonally expanded integrations in individual tumors, the above mentioned biological processes appeared repeatedly in many tumors (S4 Fig). This suggests that the gene players involved in these biological processes exhibit a degree of cooperativity to trigger oncogenic transformation. For example, among the most clonally expanded UISs in Fig 6C, TAB2 is known to be involved in immune response signaling, BTBD1 plays a role in cellular differentiation and mir-30a has known roles in regulating cell proliferation, migration and invasion [31–33].
Furthermore, we identified TERT and MYB among the most clonally expanded integrations in 15 independent tumors from 9 different birds, and in 15 independent tumors from 11 different birds, respectively (S4 Fig, S4 Table). TERT, the catalytic component of telomerase, has known roles in immortalization, senescence and apoptotic signaling [34]. MYB, a transcription factor, functions in regulation of cellular differentiation and proliferation [35]. Clonally expanded MYB integrations co-occur in half of the birds with TERT tumors. Additionally, we also observed up regulated MYB expression in TERT tumors without any ALV integrations near or within MYB [15]. This suggests a possible cooperation between MYB and TERT. Genes proximal to other clonally expanded integrations, which occur in the TERT or MYB tumors, might also cooperate with them for inducing oncogenic transformation. Of particular interest, integrations in CTDSPL and CTDSPL2 are frequently clonally expanded, along with TERT and MYB [36]. miR-155 integrations were also frequently seen with MYB in tumors (S4 Table). These putative cooperating gene players are involved in varying biological processes such as differentiation, proliferation, apoptosis, phosphorylation, immune response signaling, immortalization and DNA damage repair (S4 Table).
In order to determine common pathways activated in multiple individual tumors, we also analyzed genes near the clonally expanded integrations within individual tumors. A number of common transcription factor target gene networks were identified as common integration sites in various tumors. Among the common targets of ALV integration, the most enriched are the genes targeted by the E2F, EGR, WT1 and SP families of transcription factors (S5 Table). E2F is a well-characterized protein family that mediates both cell proliferation and apoptosis [37]. EGR (early growth response) is a family of nuclear proteins that function as transcriptional regulators and target genes required for regulating differentiation and mitogenesis [38]. The SP (specificity protein) and WT (Wilms tumor) family of transcription factors are involved in many cellular processes, including cell differentiation, cell growth, apoptosis, immune responses, and response to DNA damage [39–41]. This suggests that the ALV integrations in these genes are at the intersection of events of tumorigenic transformation. These gene players might cooperate to trigger oncogenic characteristics, thus resulting in tumor formation.
We report the analysis of more than 87,000 UISs, leading from early infections (in tissue culture) through to early and late tumor development. With only slight preferences for some genome features, we show that ALV exhibits a relatively random integration pattern. Due to this relatively minimal discrimination, ALV serves as a good insertional mutagenesis tool to study tumorigenesis. We utilize the OCI to empirically define the magnitude of clonality in different stages of tumorigenesis. Consistent with the clonal expansion hypothesis [25], we observe that ALV clonality increases with progressing stages of tumorigenesis We also identify putative cooperating gene players (especially TERT and MYB) and the underlying biological processes of cell differentiation, phosphorylation, immune response signaling, proliferation and regulation of apoptosis involved in tumor progression.
We observed a semi-random integration pattern for ALV. In contrast, MLV and FV exhibit a strong preference for integrations within TSS or CpG islands [42–44]. HIV-1, on the other hand, has a strong preference for integrating into transcriptional units with higher expression levels [5,6]. ALV shows a more random distribution of integration sites, similar to HTLV and MMTV [45,46]. Similar trends in the integration sites of different ALV subgroups were observed via an independent analysis in CEF, DT-40 and HeLa cells, suggesting that the ALV integration preference is not cell type specific.
We report a significant enrichment of ALV integrations within gene bodies as nearly 40% of integrations are found in genes relative to 27% at random integration sites. Barr et al. (2005) reported a similar bias for ALV integrations into transcriptional units relative to matched random sites [5]. They also reported that ALV favors integrations in transcriptional units with higher expression levels [5]; however, our data does not support this. This difference could be explained by our use of different methods to measure gene expression. Since we use RNA-seq data, in lieu of microarrays used previously (with 249 probe sets), our gene expression values might be different for the host genome. Moreover, we analyzed 15,416 proviral integration sites in this study compared to their analysis of 658 integrations, which should offer a more comprehensive analysis.
ALV displays a slight preference for integration near TSSs in tissue culture infections. We observe a further enrichment of integrations near the TSSs in tumors, as a consequence of tumorigenic selection. ALV integrations are also enriched within CpG islands as well as near spliced and expressed genes. There is a significant selection of ALV integrations away from CpG islands in the highly clonally expanded tumor cells (10 or more breakpoints). Since DNA methylation is often observed in CpG islands, ALV integrations near CpG islands may be more susceptible to repression by methylation [47]. Thus, cancer cells are likely enriched for integrations away from CpG islands, where the ALV proviruses are more likely to remain transcriptionally active.
In an emerging picture of B-cell malignancy, understanding tumor progression is an important piece of the puzzle. Here, we show that clonal expansion of ALV-infected B-cells is a key feature of malignant transformation in tumors. Approximately 100 to 500 UISs have been observed for HIV in in peripheral blood lymphocytes [48,49]. On the other hand, nearly 500–5000 UISs have been observed for a typical HTLV-1 host [21]. We observe approximately 700 to 3000 UISs in individual tumors induced by ALV mutagenesis. The most clonally expanded viral integrations appear to be early events in tumorigenesis and are expanded during progression of tumors. Therefore, this pattern of selection and expansion defines the clonal evolution of this cancer.
The distribution of the clone abundances can be quantified by an OCI value. Late stage B-cell neoplasms are associated with higher OCI values than earlier stages, and the PVL is also observed to correlate with the progressing stages of tumorigenesis. Interestingly, while the gag and env ratios appear very similar, LTR ratios are elevated for some individual tumor samples. This suggests that over the course of tumorigenesis, there are likely more deletions and rearrangements acquired in the gag and env regions of the viral genome. Further work will be necessary to identify the epigenetic factors that may influence proviral expression and tumorigenesis.
We observed a correlation between the PVL and different stages of tumor progression. As determined by the PVL analysis, a single cell in a tumor has multiple (2.3 to 4) copies of integrated ALV proviruses, suggesting multiple UISs contribute to oncogenic transformation. Therefore, loss of super-infection resistance could be involved in tumorigenesis. Alternatively, deletions in the env region of proviruses or mutations affecting env expression, identified in some tumors in previous work in our lab [18], might allow cells to overcome super-infection resistance.
Analysis of the ontology of genes flanking integration sites demonstrated a functional overrepresentation of certain pathways that are deregulated in many lymphomas [17]. Consistent with present concepts of oncogenesis and lymphomagenesis, GO analysis revealed that five major gene functions contribute to clonal dominance: regulation of proliferation, differentiation, immune response, apoptosis, and phosphorylation. Of these, cell differentiation and phosphorylation appear to be significantly altered in earlier stages of tumor progression. Interestingly, we also observed possible cooperativity between TERT and MYB, which might function together to induce oncogenic transformation. Further analysis via single cell sequencing would be useful to investigate this cooperativity.
Our work depicts a comprehensive investigation into the role of ALV integrations in lymphomas in chickens. The value of our work can be extended to mammalian systems. B-cell development in chicken and mammals is a very similar process [50]. These similarities are evident at levels of molecular changes and gene regulatory networks [51–53]. Although mice serve as a good mammalian model, in terms of oncogenesis they differ in some fundamental ways from humans. Unlike humans, the mouse telomerase enzyme is active in normal somatic cells [54]. This difference between humans and mice is important because telomerase activation is a critical step in the human oncogenic process, with telomerase activation seen in approximately 90% of human cancers [55,56]. Similar to human expression, chicken telomerase expression is down regulated in most somatic tissues [57]. Furthermore, chicken telomeres shorten with age, and telomerase activity is important for oncogenesis [58]. Therefore, chicken serves as an advantageous model over mouse, to study oncogenic events.
Limited information is available about the molecular mechanisms of lymphomagenesis, and the role of selective clonal expansion. Cells containing certain integration sites can undergo selective expansion in tumors, resulting in abundant clonal populations. We observed that in the course of tumor progression, the more transformed neoplasms contained integrations with a high number of breakpoints [25]. Via our genomic analysis of ALV integrations across progression of B-cell lymphomas, we are able to provide insights into the biological processes associated with initiation, progression, and metastasis of tumors.
Chicken embryo fibroblasts (CEFs) were cultured in medium 199 (Thermo Fisher Scientific) supplemented with 2% tryptose phosphate, 1% fetal calf serum, 1% chicken serum, and 1% antibiotic at 39°C and 5% CO2. Viruses were generated by transfecting CEFs via electroporation, with vectors RCASBP(A) and RCASBP(C) to generate viral titers of subgroups A and C respectively [59]. ALV-J virus [60] was generated from homogenates of tumors with ALV-J integrations, by passing it through a 0.22 micrometer pore size filter. The collected supernatant from tumors was in turn used to infect CEFs. CEFs were grown at approximately 40% confluency and were infected with ALV subgroup A, C or J at an MOI of 1–2. The cells were collected at 48 hours and 120 hours post infection for DNA isolation. DT-40 cells were cultured in Dulbecco’s modified eagle medium (Thermo Fisher Scientific), 10% fetal calf serum, 5% chicken serum, 5% tryptose phosphate, and 1% antibiotic at 37°C and 5% CO2. DT-40s were grown at approximately 40% confluency and were infected with ALV subgroup C at an MOI of 1–2. The cells were collected at 48 hours post infection for DNA isolation. HeLa cells (ATCC) were cultured in Dulbecco’s modified eagle medium, 10% fetal bovine serum (FBS), and 1% antibiotic at 37°C and 5% CO2. To generate ALV pseudo-typed with vesicular stomatitis virus glycoprotein (VSV-G), CEFs were co-transfected via electroporation, with pMD.G (VSV-G envelope plasmid) and RCASBP(C) plasmid [61]. Viral supernatant was collected after 48 h, filtered through a 0.22-micrometer filter, and concentrated by polyethylene glycol (PEG) precipitation (10% PEG8000) [62]. This concentrate of pseudo-typed ALV was used to infect HeLa cells and cells were collected 48 hours post infection for DNA isolation.
5 or 10-day-old chicken embryos were injected with ALV-LR9, ALV- ΔLR9, ALV-G919A, or ALV-U916A as described previously [10,17]. Chickens were observed daily and were euthanized when apparently ill or at 10–12 weeks after hatching. A total of 72 tissues were selected for characterization by high-throughput sequencing (S2 Table). Three uninfected tissues and several non-tumor tissues from infected birds were sequenced to serve as controls. All of the B-cell lymphomas included in the study were rapid-onset lymphomas, arising within 10–12 weeks. LR-9 is an ALV subgroup A recombinant virus consisting of gag, pol, and env genes derived from UR2-associated virus and LTRs derived from ring-necked pheasant virus [63]. ALV-ΔLR-9 contains a deletion in the gag gene, causing increased splicing to downstream genes [11]. ALV-G919A contains a silent mutation in the NRS [10]. Tumors were collected from primary bursal (B) tissue or metastasized liver (L), kidney (K) or spleen (S) tissues.
Five- and ten-day old chicken embryos were injected with virus. Chickens injected include inbred SC White Leghorn line embryos (Hy-Line International, Dallas Center, IA), and SPAFAS embryos (Charles River). Chickens were euthanized at 10–12 weeks post hatching. Institutional Animal Care and Use Committee (IACUC) approval was obtained at the University of Delaware and the Fred Hutchinson Cancer Research Center.
DNA from ALV infected cultured cells or tumor samples were isolated. The sequencing libraries were prepared as described previously [16]. Five micrograms of purified genomic DNA was sonicated with a Bioruptor UCD-200. End repair, A-tailing, and adapter ligation were performed as described previously [21] (adapter short arm, P-GATCGGAAGAGCAAAAAAAAAAAAAAAA, and adapter long arm, CAAGCAGAAGACGGCATACGAGATXXXXXXGTGACTGGAGTTCAGACGTGTGCTCTTCCGATC*T, where “X’s” denote the barcode sequence, “P” denotes phosphorylation, and “*” denotes a phosphorothioate bond). Nested PCR was performed to enrich the library for proviral junctions. The first PCR step had 23 cycles and employed an ALV-specific primer (CGCGAGGAGCGTAAGAAATTTCAGG) between the 3’ LTR and env and a primer (CAAGCAGAAGACGGCATACGAGAT) within the adapter that was attached by ligation in the previous step. In the second round of PCR, a primer (AATGATACGGCGACCACCGAGATCTACACTCGACGACTACGAGCACATGCATGAAG) near the 3’ end of the LTR was used. This primer ended 12 nucleotides short of the junction between viral and genomic DNA. This primer was paired with an adapter-specific primer on the opposite side of the fragment, which overlapped the adaptor’s bar code sequence (CAAGCAGAAGACGGCATACGAGATXXXXXX). Libraries were quantified by quantitative PCR (qPCR) and then under- went single-end 100-bp multiplexed sequencing on the Illumina Hi-Seq 2000. A custom sequencing primer (ACGACTACGAGCACATGCATGAAGCAGAAGG) was used, which hybridized near the end of the viral 3’ LTR, 5 nucleotides short of the proviral/genomic DNA junction. The resulting reads could be validated as genuine integrations by verifying that they began with the last 5 nucleotides of the proviral DNA, CTTCA. The last two nucleotides of the unintegrated proviral DNA, TT, are cleaved by ALV integrase upon integration, so the lack of these 2 nucleotides in the read acted as further validation of a true viral integration.
Reads were first curated with a custom python script to remove sequences that did not begin with the last five nucleotides of viral DNA, “CTTCA” [16,17]. The files were then uploaded to Galaxy [64–66], which was used to perform downstream analyses. In Galaxy, first the quality scores were converted to Sanger format with FastQ Groomer v1.0.4 [67]. CTTCA and adapter sequences were then trimmed using the Galaxy Clip tool v1.0.1. This tool also removed reads containing an N and reads less than 20 nucleotides in length after adapter removal. The remaining reads were mapped with bowtie [66] using the Gallus gallus 4.0 genome (Nov. 2011). Sequences were aligned using a seed length of 28 nucleotides, with a maximum of 2 mismatches permitted in the seed. All alignments for a read were suppressed if more than one reportable alignment existed. This was done to prevent multiple mapping and ensure that reads correspond to only unique integration sites. 100,000 random mapped reads were selected from each sample to be used for further analysis. If less than 100,000 reads were present for a sample, all available reads were used. A custom Perl pipeline developed in the lab was used to analyze the aligned reads output from bowtie [16,17]. This custom pipeline identified unique integration locations, and calculated the number of reads and sonication breakpoints for each integration site. It also identified hotspots of integration and common integration sites among multiple samples. Integrations from two unrelated barcodes on the same sequencing lanes were omitted via our pipeline. The pipeline source code is available upon request. The integration sites identified in our work are deposited at the NCI Retrovirus Integration Database (RID) (https://rid.ncifcrf.gov/) [68].
Reads for the junctions of proviral integration and genomic DNA were mapped with Bowtie [69]. Only reads that mapped uniquely to the genome were utilized for further analysis. This step filtered out reads that originate from repetitive elements. Mapped reads from all samples were then combined into a single file and analyzed with HOMER [23]. HOMER calculates the enriched features at each integration locus as well the proximity to closest transcription start site. A random integration control data set was generated with Bedtools Random [70]. The genomic DNA sequences corresponding to the genomic coordinates obtained from Bedtools Random were extracted from the Gallus gallus 4 genome using the Galaxy tool Extract Genomic DNA. Control sequences were mapped with Bowtie and analyzed with HOMER using the same parameters as for ALV integrations. Proximity to CpG islands was determined using the WindowBed tool in Galaxy [66].
We note that our calculations are subject to certain biases. This includes, but is not limited to, an underestimate of the chicken or human genome sizes due to unsequenced gaps or overlapping sequences. Furthermore, an aberrant karyotype, which might exist in the transformed HeLa [71] or the DT-40 cells [72], was not taken into account for our analysis. However, as previously determined by Narezkina et al. (2004), despite the aberrant karyotype in HeLa cells, the ratio between the genome size and the gene number in HeLa cells is equivalent to that of the normal human genome [4].
The ensembl Gallus gallus 4 genome was utilized to obtain reference information for the transcript count and number of introns for all transcriptional units in the chicken genome. If an integration occurs within a gene, then the corresponding gene is used for all subsequent analysis. If an integration occurs in an intergenic region, then the nearest gene is used for all subsequent analysis. RNA-seq data for analysis of CEFs was downloaded from the public Sequence Read Archive (SRA) database (SRA accession no. SRP107761) [73]. A custom Python script was utilized to associate the expression, transcript count and number of introns of a gene with the number of ALV integrations proximal to or within the given gene. A matched random control set, generated as mentioned above, was used as a control. The Python source code is available upon request.
PVL was measured by quantitative polymerase chain reaction (qPCR) of ALV-LR9 for env (primers CCTGAAACCCAGTGCATAAGG and CTAGCTGTGCAGTTCACCGT), gag (primers GTTTAGAGAGGTTGCCCGAC and GTCAATGATCACCGGAGCCC) and LTR (CGAACCACTGAATTCCGCAT and GAATCAACGGTCCGGCCATC); and HMG14b (primers ACTGAAGAGACAAACCAAGAGC and CCAGCTGTTTTAGACCAAAGAATAC) using Q SYBR green Supermix (Bio-Rad) according to the manufacturer’s protocol on a Bio-Rad C1000 thermal cycler/CFX96 Real-Time System. We assumed a single copy of env and gag and 2 copies each of HMG14b and the LTR per cell. HMG14b is a known single copy gene in the chicken genome and thus, was used as a housekeeping reference gene [74]. Thermal cycling conditions were 95°C for 20 seconds and 40 cycles each of 95°C for 1 second followed by 53°C for 30 seconds. Quantitative PCR (qPCR) was performed in duplicate, with each sample present in technical duplicate during each run. The results were normalized to those for normal bursa using the comparative threshold cycle (CT) method.
Statistical analysis for clonality index was carried out using R version 2.15.2 (http://www.R-project.org/). The oligoclonality index (OCI; Gini coefficient) was calculated using the R package sonicLength (http://soniclength.r-forge.r-project.org/) as described previously [21,22].
Functional profiling of genes and ontology analysis for the clonally expanded integrations was conducted with g:profiler, using an ordered query option (http://biit.cs.ut.ee/gprofiler/) [30].
|
10.1371/journal.ppat.1003747 | A Novel Membrane Sensor Controls the Localization and ArfGEF Activity of Bacterial RalF | The intracellular bacterial pathogen Legionella pneumophila (Lp) evades destruction in macrophages by camouflaging in a specialized organelle, the Legionella-containing vacuole (LCV), where it replicates. The LCV maturates by incorporating ER vesicles, which are diverted by effectors that Lp injects to take control of host cell membrane transport processes. One of these effectors, RalF, recruits the trafficking small GTPase Arf1 to the LCV. LpRalF has a Sec7 domain related to host ArfGEFs, followed by a capping domain that intimately associates with the Sec7 domain to inhibit GEF activity. How RalF is activated to function as a LCV-specific ArfGEF is unknown. We combined the reconstitution of Arf activation on artificial membranes with cellular expression and Lp infection assays, to analyze how auto-inhibition is relieved for LpRalF to function in vivo. We find that membranes activate LpRalF by about 1000 fold, and identify the membrane-binding region as the region that inhibits the Sec7 active site. It is enriched in aromatic and positively charged residues, which establish a membrane sensor to control the GEF activity in accordance with specific lipid environments. A similar mechanism of activation is found in RalF from Rickettsia prowazekii (Rp), with a different aromatic/charged residues ratio that results in divergent membrane preferences. The membrane sensor is the primary determinant of the localization of LpRalF on the LCV, and drives the timing of Arf activation during infection. Finally, we identify a conserved motif in the capping domain, remote from the membrane sensor, which is critical for RalF activity presumably by organizing its active conformation. These data demonstrate that RalF proteins are regulated by a membrane sensor that functions as a binary switch to derepress ArfGEF activity when RalF encounters a favorable lipid environment, thus establishing a regulatory paradigm to ensure that Arf GTPases are efficiently activated at specific membrane locations.
| The intracellular pathogens Legionella pneumophila (Lp) and Rickettsia prowazekii (Rp) inject an effector (RalF) that diverts the host trafficking small GTPase Arf1. In the case of Lp, LpRalF recruits Arf1 to the Legionella-containing vacuole (LCV), where the pathogen replicates. RalF proteins are related to eukaryotic ArfGEFs, from which they depart by a unique auto-inhibitory capping domain that contains localization and functional determinants. In this work, we combined the reconstitution of RalF ArfGEF activity on artificial membranes with cellular and Lp infection assays to uncover how auto-inhibition is released for RalF proteins to function in vivo. We find that LpRalF and RpRalF are activated by membranes by about 1000-fold and map the membrane sensor to a unique motif in the capping domain. This motif is identical to the auto-inhibitory motif, thus establishing a binary switch that controls the ArfGEF activity of RalF in accordance with specific lipid environments. Finally, we show that the membrane sensor drives LpRalF binding to the LCV and timing of Arf activation during Lp infection. These results establish how RalF proteins are derepressed when they encounter a favorable lipid environment, and provide an evolutionary explanation to the presence of RalF in pathogens with diverging lifestyles.
| A number of intracellular pathogenic bacteria can bypass regulatory networks used to control trafficking and cytoskeletal pathways of the infected cell by delivering bacterial effector proteins into the host cytosol that function as illegitimate regulators of small GTPases (reviewed in [1], [2]). One of them, Legionella pneumophila (Lp), the causative agent of a severe pneumonia, the Legionnaire's disease, invades and replicates in macrophages where it survives in a specialized membrane-bound compartment, the Legionella-containing vacuole (LCV) (reviewed in [3]). Maturation of the phagosome into the LCV is driven by an arsenal of effectors delivered by a type IV secretion system called Dot/Icm [4], [5]. Instead of fusing with lysosomes where degradative enzymes would destroy the pathogen, the LCV incorporates membranes from the endoplasmic reticulum (ER), a nutrient-rich compartment that supports multiplication of Lp in high numbers within the macrophage [3], [6], [7]. Over the last decade, a number of Lp effectors have been shown to divert cellular proteins that steer membrane traffic (reviewed in [2], [3], [8]. These include several illegitimate regulators or modifiers of small GTPases of the Arf and Rab families, which are major regulators of cellular traffic in eukaryotes (reviewed in [9], [10]).
One of these effector proteins, RalF, contains a Sec7 homology region [4], which is the catalytic domain in eukaryotic guanine nucleotide exchange factors (GEFs) that is sufficient to activate Arf by stimulating GDP/GTP exchange (reviewed in [11]). Shortly after infection, the LpRalF protein is detected on the cytosolic surface of limiting membranes that defines the LCV [4]. Localization of LpRalF to the LCV is sufficient to mediate the recruitment of cellular Arf GTPases to this organelle by a mechanism that is dependent on a functional Sec7 domain. Arf activity is important for fusion of ER-derived membranes with the LCV [7], although the recruitment of Arf proteins to the LCV is currently of unknown importance (reviewed in [3]). A protein with primary sequence similarity to LpRalF is also encoded by Rickettsia prowazekii (Rp) [4], the bacterial pathogen responsible for epidemic typhus. Rp is unrelated to Lp phylogenetically, and unlike Lp, it lyses the vacuole in which it resides to replicate freely in the cytosol (reviewed in [12]).
Structural studies showed that the C-terminal domain of LpRalF intimately associates with the Sec7 domain to block access to the Arf-binding site [13]. Accordingly, the ArfGEF activities of LpRalF and its homolog from Rickettsia are strongly auto-inhibited in vitro [4], [13], [14]. This domain was thus termed the capping domain. Recently, the capping domains of LpRalF and RpRalF were shown to localize to host membranes when expressed in cells [14]. However, the LpRalF capping domain localized to a perinuclear region reminiscent of the endoplasmic reticulum, whereas the RpRalF capping domain localized to the plasma membrane. In addition, expression of full-length LpRalF and RpRalF in cells resulted in divergent effects, with LpRalF impairing secretion and RpRalF disrupting actin dynamics at the plasma membrane [14]. Thus, the LpRalF and RpRalF proteins represent similar ArfGEF proteins that display different functions and membrane localization properties.
The mechanism by which the LpRalF protein identifies the LCV membrane and restricts its ArfGEF activity to this organelle inside the host cell, and how LpRalF and RpRalf target different membranes, remain important and unanswered questions. In eukaryotes, evidence has accumulated that GEFs not only activate their small GTPase substrates by stimulating GDP/GTP exchange, but also process upstream activating signals, restrict the subcellular localization of active GTPases, and likely convey downstream information (reviewed in [15]). These multiple functions are controlled by sophisticated regulatory mechanisms such as auto-inhibition, feed-back loops and activating interactions with other proteins and/or with membranes, and often involve large conformational changes. Although the auto-inhibitory capping domain of RalF proteins is unrelated to any domain of known structure or function, localization studies in cells have suggested that it has a critical role in RalF localization, and that membrane interactions may be needed to relieve autoinhibition and regulate the ArfGEF activity of the LpRalF protein in vivo.
Here, we investigated the RalF nucleotide exchange reaction using artificial membranes to reconstitute the cellular environment in which the LpRalF and RpRalF proteins function. Combined with cellular expression and Lp infection assays, these data have led to the identification of a membrane sensor in the capping domain of RalF proteins that contributes to membrane localization and spatial regulation of Arf activation on membranes during infection of host cells.
The C-terminal capping domain of LpRalF obstructs the active site of its Sec7 domain, hence must be displaced to bind Arf GTPase substrates (Figure 1A). We first used purified full-length LpRalF and its Sec7 domain alone (LpRalFSec7, residues 1–201) (Figure S1A) to quantify the auto-inhibition of nucleotide exchange by the capping domain. Nucleotide exchange kinetics of human Arf1 lacking its N-terminal α-helix (Δ17Arf1), which is readily activated by ArfGEFs in solution (reviewed in [16]), were monitored by tryptophan fluorescence (Figure 1B). As previously reported [13], [14], LpRalF was essentially inactive in solution (kcat/Km = 3.15±0.17 102 M−1s−1), reflecting strong auto-inhibition. Removal of its capping domain increased exchange efficiency by about 10-fold (kcat/Km = 4.41±0.52 103 M−1s−1). Detection of measurable nucleotide exchange required at least stoechiometric LpRalF amounts, and remained about 1–2 orders of magnitude lower than what is achieved by cellular ArfGEFs [17]. LpRalFSec7 has however the hallmarks of a conventional Sec7 domain, as it readily formed exchange intermediates with Arf1-GDP by mutating the catalytic glutamate to lysine (E103K) (Figure S1B) and with nucleotide-free Arf1 by enzymatic removal of GDP (Figure S1C).
These data suggest that other factors are needed for LpRalF to reach full efficiency on the LCV. The capping domain of LpRalF was previously shown to co-localize with subcellular compartments and to fractionate with membranes when expressed in cells [14], suggesting that membranes could be involved in activation. We first analyzed whether full-length LpRalF has the ability to bind to membranes. Direct binding to liposomes was observed using immunogold-labeling electron microscopy (Figure 1C) and by liposome co-sedimentation experiments (Figure 1D). Next, we conducted ArfGEF assays in vitro in the presence of synthetic liposomes using purified LpRalF and the cellular form of Arf1, which carries a myristoyl lipid attached to its N-terminal helix (myrArf1) (Figure 1E). Using this more physiological ArfGEF assay having both myrArf1 and liposomes the catalytic efficiency of LpRalF increased by 3 orders of magnitude compared to its activity in solution (kcat/Km = 2.44±0.10 105 M−1s−1), thus reaching an efficiency similar to that of eukaryotic ArfGEFs [17]. Importantly, under these physiological conditions, LpRalFSec7 remained essentially inactive (Figure 1E), which indicates that the Sec7 domain alone is unable to efficiently activate myrArf1 on a membrane surface and that the potentiating effect is entirely mediated by the capping domain.
Recent studies highlighted positive feedback regulation of cellular GEFs on membranes, in which initial production of the GTP-bound GTPase potentiates nucleotide exchange efficiency [18]–[20]. To analyze whether LpRalF is controlled by a feedback loop, we either added a large amount of the Arf effector GRAB to the exchange reaction to cancel a feedback effect by depleting Arf1-GTP as it is generated, or we added increasing amounts of myrArf1-GTP prior to measuring exchange rates to maximize such an effect. Neither of these conditions affected the exchange rate of LpRalF (Figures S1D and 1F), indicating that LpRalF is not under a positive feedback control by myrArf1-GTP.
Together, these results suggest that RalF activation comprises a conformational component, which is required to release its auto-inhibition by the capping domain, and a spatial component that is needed for its localization to membranes. These components add up to yield an activation of about 1000-fold, of which the conformational change of the capping domain accounts for about a 10-fold contribution as mimicked in solution by the deletion of the capping domain. Our data also show that neither the Sec7 domain nor membrane-attached myrArf1-GTP contribute to the spatial component, indicating that potentiation of LpRalF activity by membranes is entirely mediated by its capping domain.
The recognition of the capping domain of LpRalF as a membrane-binding domain without homology to any known membrane-binding determinants prompted us to investigate its sensitivity to membranes physico-chemical properties and curvature. This was probed by comparing the exchange rates at fixed concentrations of LpRalF (0.1 µM) and myrArf1 (0.4 µM), using the fluorescence-based nucleotide exchange assay in the presence of liposomes of various charges, curvature and packing (Figure 2A, S2A and S2B). LpRalF had only a weak activity in the presence of uncharged liposomes containing only phosphatidylcholine (PC) and phosphatidylethanolamine (PE). Exchange rates were increased up to 15-fold in the presence of negatively charged phospholipids, with the largest effects resulting from addition of 30% phosphatidylserine (PS) and 5% phosphatidylinositol (4,5) bisphosphate (PIP2). Liposomes of intermediate charge characteristics (50% PC, 19% PE, 5% PS, 10% phosphatidylinositol (PI), 16% cholesterol) extruded through filters of different sizes were used to probe the effect of curvature and packing. Increasing the curvature using filters from 0.4 µm to 0.03 µm resulted only in a modest increase of the exchange rates. In contrast, replacing natural PC, PE and PS lipids by dioleoyl-PC, dioleoyl-PE and dioleoyl-PS lipids, whose unsaturated fatty acid chains form more loosely packed membranes, resulted in a 10-fold increase of the exchange rates. These results indicate that LpRalF is highly sensitive to both the presence of negatively charged lipids and to packing defects but is not sensitive to membrane curvature.
A prominent feature of LpRalFcap is an unusual cluster of aromatic residues, located on two twin α-helices, 248PKSWLSFFTG257 and 280PNIFSKWLFG289, which are wedged into the catalytic groove of the Sec7 domain and must be displaced before Arf GTPases can interact with the Sec7 catalytic site (Figure 1A). Aromatic residues are well-suited for peripheral insertion of proteins into membranes [21], suggesting that the membrane-binding site of the capping domain may encompass these helices. We used Trp fluorescence to measure whether addition of liposomes result in changes in the environment of Trp251 and Trp286, which are located in the cluster. Liposomes induced a marked decrease in the fluorescence signal, to which Trp251 and Trp286 contributed in an additive manner as shown by their mutations to alanines (Figure 2B). These data suggest that the environment of the aromatic cluster is modified by the presence of membranes. However, they do not discriminate between effects that are due to auto-inhibition release from those arising from a potential interaction with membranes. Alternatively, we reasoned that introducing a charged residue in the cluster could reveal direct membrane interactions by monitoring nucleotide exchange in the presence of negatively charged lipids. We chose to mutate Phe255, which has a strategic location in auto-inhibited LpRalF, where it mimics a critical Phe residue in the switch 1 of Arf (Phe51) [22], [23] and is superimposable to an auto-inhibitory Phe residue in cytohesins (Phe 262 in the PH domain of GRP1, [24]). We first assessed whether the F255K mutation would affect the biochemical and structural properties of LpRalF in solution. The SAXS profiles of LpRalF and LpRalFF255K were in good agreement at small Q values (<0.2 Å−1), indicating that LpRalFF255K is mostly in an auto-inhibited conformation in solution (Figure 2C). Consistently, the crystal structure of LpRalFF255K retained an auto-inhibited conformation (Figure S2C), in which the mutation is accommodated by increased local disorder (Figure 2D). In agreement with structural data, LpRalFF255K remained auto-inhibited in solution (kobs = 2.5±0.9 10−4 s−1 measured at 1 µM LpRalFF255K and 1 µM Δ17Arf1, to be compared with kobs = 9±0.9 10−4 s−1 measured for LpRalF under the same conditions). These data establish that the F255K mutation is silent in solution, thus that this protein can be used to investigate auto-inhibition and membrane effects independently. Consistent with the hypothesis, LpRalFF255K was at least 10 times more active than wild type LpRalF in the presence of negatively charged liposomes (kobs = 47.3±12 10−2 s−1 measured at 0.1 µM LpRalFF255K and 0.4 µM myrArf1) (Figure S2D), suggesting that the extra lysine residue facilitates interactions with negatively-charged lipids. Conversely, we reasoned that mutation of F255 to glutamate should introduce repulsive interactions with negatively-charged lipids that should be detrimental to its efficiency. Indeed the nucleotide exchange activity of LpRalFF255E was reduced 4-fold compared to wild-type LpRalF (kobs = 1.3±0.2 10−2 s−1) (Figure S2D). These large effects observed in the presence of membranes but not in solution strongly suggest that the membrane-binding site of the capping domain encompasses the aromatic cluster.
Several species of Rickettsia encode a RalF homolog of unknown function, and these RalF proteins all have a conserved capping domain that includes an aromatic cluster (Figure S3). Unlike Legionella, Rickettsia do not replicate in an intracellular vacuole (reviewed in [12]), raising the issue of whether Rickettsia RalF proteins function as membrane-regulated ArfGEFs. Purified Rickettsia prowazekii RalF (RpRalF) (Figure S1A) displayed a marked decrease of tryptophan fluorescence upon addition of liposomes (Figure 3A), indicating that membranes modify the environment of the unique tryptophan in the aromatic cluster, Trp283. Direct interaction of RpRalF with membranes was further confirmed by its co-sedimentation with liposomes (Figure 3B). To analyze whether the mechanism of activation by membranes observed for LpRalF also applies to RpRalF, we compared its GEF activity in solution and in the presence of membranes. RpRalF was strongly autoinhibited in solution (kobs = 4±1.2 10−4 s−1 measured at 1 µM RpRalF and 1 µM Δ17Arf1, which is very close to the kobs = 8±3.5 10−4 s−1 measured in the absence of GEF). Liposomes containing anionic lipids stimulated nucleotide exchange on myrArf1 at a level similar to that observed for LpRalF (kcat/Km = 2.85±0.09 105 M−1s−1, Figure 3C). Thus, RpRalF is a membrane-interacting ArfGEF that is auto-inhibited in solution and strongly activated by membranes. Given the sequence conservation between LpRalF and RpRalF, these data suggest that RpRalF similarly uses its capping domain for both auto-inhibition and membrane binding, despite the different intracellular lifestyles of these two pathogens. However, RpRalF was not activated by liposomes containing dioleoyl lipids (Figure 3C), in striking contrast with LpRalF under the same conditions, indicating that they have different lipid preferences.
Our in vitro analysis predicts that LpRalF uses the membrane sensor encoded in the aromatic cluster to spatially and temporally regulate Arf activation during the course of LCV maturation. As a first indication, we observed that the F255K mutation in the aromatic cluster was sufficient to displace LpRalF capping domain expressed in HeLa cells (YFP-LpRalFcap) from the reticulate perinuclear pattern reminiscent of the ER observed for wild-type YFP-LpRalFcap [14] to a Golgi pattern (Figure S2E). Consistently, despite its high exchange activity on negatively charged liposomes in vitro, M45-flagged LpRalFF255K failed to recruit Arf1 at the LCV in infected cells, likely due to mislocalization. Interestingly, the capping domains of LpRalF and RpRalF have strikingly divergent localizations when expressed in eukaryotic cells [14] and have different ratio in aromatic and positively charged residues in their aromatic clusters (Figure 4A). Because the aromatic cluster is involved in membrane sensing, divergences between LpRalF and RpRalF aromatic clusters could explain the different localizations of the two capping domains. To test this hypothesis, we introduced reciprocal mutations in LpRalF and RpRalF capping domains (F254K, T279K, Q291K and P292S in LpRalFcap; K252F, K276T and K288Q in RpRalFcap). Recombinant MBP-LpRalFcapmut bound to negatively charged lipids in a lipid overlay assay, unlike MBP-LpRalFcap but reminiscent of MBP-RpRalFcap (Figure S4A, top). Conversely, MBP-RpRalFcapmut lost the ability of MBP-RpRalFcap to bind to negatively charged lipids in this assay (Figure S4B, bottom). When expressed in HeLa cells, YFP-LpRalFcapmut relocated from the perinuclear localization observed for YFP-LpRalFcap to the plasma membrane and to Golgi-like structures (Figure 4B, top). Conversely, YFP-RpRalFcapmut relocated from the plasma membrane to a perinuclear localization (Figure 4B, bottom). These data indicate that the content in aromatic and positively charged residues in the aromatic cluster is a major determinant of the subcellular localization of the capping domain.
To analyze whether the aromatic cluster controls the localization of the full-length RalF protein and the timing of Arf1 activation during the maturation of the LCV, we infected cells with L. pneumophila ΔralF expressing either 3*Flag-LpRalF or 3*Flag-LpRalFmut carrying the F254K, T279K, Q291K and P292S mutations. Both constructs were expressed and translocated to similar levels (Figures S4B and S4C). Strikingly, while LpRalF was still present at the surface of a large fraction of LCVs at 6 h post-infection (26%±2.4), LpRalFmut could not be detected (1.3%±1.15) (Figure 4C), indicating that the aromatic cluster controls LpRalF localization. Next, we compared the timing of Arf1 activation on the LCV (Figure 4D). Both LpRalF constructs were equally efficient at recruiting Arf1 one hour after infection. However, while Arf1 activation by LpRalF continued to rise for another hour and was still detectable after 8 hours, its activation by LpRalFmut decreased after one hour and became undetectable after 4 hours. Altogether, these data indicate that the aromatic cluster functions as a membrane sensor in vivo and that it drives the timing of Arf1 activation during infection in accordance with the ratio between aromatic and positively charged residues.
In addition to blocking the Sec7 catalytic site by its aromatic cluster, the capping domain is also involved in intramolecular interactions with the N-terminus of the Sec7 domain that stabilize the auto-inhibited conformation in LpRalF [13] (Figure 1A). The contact is mediated by a 4-residue motif found in all RalF homologs, 323KATY326 in LpRalF, of which the conserved tyrosine forms a direct interaction with the Sec7 domain (Figures 5A and S3). To investigate the contribution of this motif to RalF regulation and function, we characterized LpRalF constructs carrying a Y326D mutation. We first analyzed the properties of the capping domain carrying this mutation. YFP-LpRalFcapY326D ectopically expressed in HeLa cells showed a perinuclear localization phenotype that was similar to wild-type YFP-LpRalFcap (Figure 5B), which was in contrast to capping domain proteins having mutations in the aromatic cluster (compare with Figures 4B and S2E). Likewise, YFP-LpRalFcapY326D retained the ability to impair secretion (Figure 5C) and to disrupt the Golgi (Figure 5D) previously observed for wild-type YFP-LpRalFcap [14]. Full-length YFP-LpRalFY326D ectopically expressed in HeLa cells, however, did not impair secretion (Figure 5C) or disrupt Golgi architecture (Figures 5D and 5E), unlike full-length wild-type YFP-LpRalF. In addition, recombinant full-length LpRalFY326D was unable to activate Arf1 in the presence of liposomes (kobs = 9±1.2 10−5 s−1 at 0.4 µM myrArf1 and 0.2 µM LpRalFY326D, to be compared to 4.7±1.0 10−2 s−1 for LpRalF under the same conditions) and Legionella ΔralF expressing full-length M45-tagged LpRalFY326D failed to recruit Arf1 at the LCV during infection (Figure 5F). These observations indicate that the mutation is silent in the context of the capping domain alone but is inactivating in the context of the full-length protein, which suggests that the KATY motif is necessary for reorganization of LpRalF to an active conformation.
Previous studies showed that the ArfGEF activity of Legionella RalF is auto-inhibited in solution, which implied that there must be a cellular mechanism for activation that remained unknown [13]. In this work, we demonstrate that interactions between an aromatic cluster of amino acids in the LpRalF C-terminal capping domain and lipid membranes provides a signal that relieves autoinhibition and converts LpRalF to a highly potent ArfGEF. Auto-inhibition and membrane recruitment are mediated by the same region of the capping domain, thus making both states mutually exclusive and establishing a membrane-driven binary switch. Thus, Legionella has evolved a minimal version of the regulatory mechanisms found in eukaryotic GEFs, which enables it to by-pass the cellular regulation of the small GTPase Arf by a single functional site that derepresses the ArfGEF activity when RalF encounters a favorable lipid environment.
Our data show that the membrane sensor of the capping domain of LpRalF contains protruding α-helices that expose aromatic and lysine residues. This atypical composition and structure endows it with a dual sensitivity to packing and electrostatic properties of lipid membranes, but not to curvature. We propose that the aromatic residues encode the sensitivity of LpRalF to packing defects by peripheral insertion into the lipid bilayer, and the lysines recognize negatively charged membrane surfaces by electrostatic interactions. The absence of a recognizable pocket that could accommodate lipid polar headgroups suggests that the recognition of negatively charged lipids is largely non-specific. The highly convex shape of the sensor might also explain why it does not detect membrane curvature. These characteristics make the capping domain a unique membrane-binding determinant with combined membrane sensitivities that depart from those of specialized phosphoinositide-binding domains, such as PH or FYVE domains (reviewed in [25]), or those of curvature-sensing domains, such as ALPS or BAR domains (reviewed in [26]).
Why would LpRalF have evolved to utilize a dual sensitivity membrane sensor? LpRalF is injected rapidly during uptake of Legionella and the ArfGEF activity promotes recruitment of Arf1 to the nascent LCV [13]. LpRalF remains on the LCV after internalization of the bacteria [4]. A major feature of the LCV is that it rapidly converts from a plasma membrane-derived to an ER-like organelle [3], [6]. Accordingly, the LCV is predicted to rapidly loose the negatively charged character of the plasma membrane, and to acquire the characteristics of ER membranes, which are not enriched for negatively charged lipids and are more loosely packed (reviewed in [26]). We propose that the ability of the capping domain to sense these unrelated membrane environments allows LpRalF to be activated soon after its injection on the nascent LCV and to remain attached as the LCV maturates by incorporating ER vesicles (Figure 6A). Hence, the capping domain would directly monitor the duration of Arf activation on the LCV.
Our results strongly suggest that activation of Rickettsia RalF also requires activation through displacement of the capping domain resulting from its recruitment to membranes, according to a scenario similar to that demonstrated in this study for Legionella RalF. However, R. prowazekii is phylogenetically distant from Legionella species and rapidly escapes from the phagosome to replicate in the cytosol (reviewed in [12]). We speculate that the RalF protein family is comprised of xenologs, meaning that an ancestral ralF gene was transmitted between different bacterial species though horizontal gene transfer and this gene then evolved divergently to activate cellular Arf proteins on distinct subcellular membranes. The observation that variations in the aromatic and positively charged residues in the membrane sensor modulate membrane targeting properties suggests that the sensor probably also plays a critical role in Rickettsia infection by restricting RpRalF to specific membranes. The difference in lipid preference between RpRalF and LpRalF and the combined ability of the capping domain of RpRalF to target the plasma membrane and to interfere with the actin cytoskeleton [14] suggest that RpRalF could be involved in bacterial entry into host cell or that it could control actin-dependent processes at the plasma membrane after uptake, thus representing a remarkable case where similar biochemical activity results in unique effects.
The active conformation of RalF proteins has not been resolved structurally. Our analysis suggests that membranes are mandatory to stabilize this open conformation. However, our data provide indirect insight into the nature of the conformational change that are required for derepression of the ArfGEF activity. First, the aromatic cluster should face the membranes, which can be modeled with the twin helices parallel to the membrane (Figure 6B) [21]. Second, the aromatic cluster in membrane-bound RalF should lie in the same plane as the myristoylated N-terminal helix of Arf, so that both can interact with membranes simultaneously. The position of Arf1 bound to RalFSec7 can be modeled from the crystal structures of Arf/Sec7 complexes (Figure 1A) [22], [23], suggesting that the capping domain should swing by at least 90 degrees to release autoinhibition. The short length of the linker that connects the Sec7 and capping domains (191PFELNFVKTSP201 in LpRalF, Figure 1A) and our analysis of the role of the KATY motif in supporting the organization of active LpRalF raises the possibility that the capping domains remains in contact with the N-terminus of the Sec7 domain throughout the exchange reaction, as seen in BRAG ArfGEFs [27]. Structural elucidation of the Arf/RalF complex, which escaped biochemical isolation so far, is now needed to unravel the details of this mechanism.
Legionella pneumophila serogroup 1, strain Lp01 [28], and the ΔralF mutant [4] were used for infection experiments. Legionella strains were grown on charcoal yeast extract (CYE) plates (1% yeast extract, 1% N-(2-acetamido)-2-aminoethanesulfonic acid (ACES; pH 6.9), 3.3 mM l-cysteine, 0.33 mM Fe(NO3)3, 1.5% bacto-agar, 0.2% activated charcoal), supplemented with 10 µg/mL chloramphenicol when required [29].
For in vitro experiments, Legionella pneumophila RalF was subcloned into the pHis-1 vector [13] and used as a matrix for the mutants using the QuickChange site-directed mutagenesis kit (Stratagene). The sequence coding for the Sec7 domain (residues 1 to 201) of LpRalF was PCR amplified and cloned into the Gateway destination vector pDEST14 (Invitrogen). The codon optimized RpRalF sequence (Genescript) was subcloned in the pDEST17 vector (Invitrogen). All clones were confirmed by sequencing (GATC Biotech).
MBP-tagged LpRalF195–374 and RpRalF189–359 were cloned in pMALc5x vector (BamHI/EcoRI). LpRalF1–374, LpRalF192–374 and RpRalF189–359 were subcloned in pYFPC1 (EcoRI/BamHI) for expression in eukaryotic cells. For expression in L. pneumophila, LpRalF was subcloned in pJB1806 (M45-tagged) or pSN85 (3*Flag-tagged) vectors (BamHI/SalI). Site-directed mutagenesis was performed to obtain single point mutants. Plasmids were amplified using two complementary primers containing the desired mutation with Pfu turbo (Stratagene). The product was digested by DpnI for 1 h at 37°C before transformation in DH5α. MBP-RalF constructs were purified as in [14].
All other recombinant proteins were produced in BL21(DE3)Star or Rosetta(DE3)pLysS Escherichia coli strains in 2×TY medium by inducing with 0.5 mM IPTG at 20°C. After centrifugation at 5000 g for 30 minutes, bacterial pellets were resuspended in 40 mL of lysis buffer (50 mM Tris pH 8.0, 300 mM NaCl, 10 mM imidazole, 0.25 mg.mL−1 lysozyme, anti-proteases, except for RpRalF for which Tris pH 8.0 was replaced by Tris pH 9.0 and 600 mM NaCl was used instead of 300 mM) per liter of culture and frozen at −80°C. After thawing, cells were sonicated, cleared by centrifugation at 20 000 g for 30 minutes and the supernatant was filtered over a 0.22 µm filter. Proteins were purified by an affinity step on a 5 mL HisTrap nickel affinity column (GE Healthcare) using 60 mM imidazole for elution, followed by gel filtration on a Superdex 200 column (GE Healthcare) in a buffer containing 10 mM Tris pH 8.0, 150 mM NaCl, and 2 mM β-mercaptoethanol. Protein purity was confirmed by SDS-PAGE, and all proteins were well folded as assessed by circular dichroism, allowing for their accurate kinetics analysis. Purified proteins were concentrated to at least 10 mg.mL−1 before storing at −80°C with 10% glycerol. Human Arf1 truncated of its N-terminal helix (Δ17Arf1) was expressed and purified as described in [23], and was loaded with GDP prior to all experiments. Myristoylated Arf1 (myrArf1) was obtained by co-expression of yeast myristoyltransferase and purified as described in [30]. MyrArf1 expressed in bacteria is readily fully loaded with GDP.
Formation of the complexes between LpRalFSec7 constructs and Δ17Arf1 was analyzed by gel filtration on a Superdex 75 10/300GL (GE Healthcare) and visualized by SDS-PAGE. For the nucleotide-free complex, LpRalFSec7 (residues 1–201) was incubated with an excess of Δ17Arf1 and incubated with agarose bead-coupled alkaline phosphatase (Sigma Aldrich) for 12 hours at 4°C, followed by centrifugation at 16000 g to remove the beads. The complex between Δ17Arf1-GDP and LpRalFSec7/E103K was obtained by incubation in 10 mM Tris pH 8.0, 150 mM NaCl, 2 mM β-mercaptoethanol, 2 mM EDTA, 1 mM MgCl2 for 10 minutes at room temperature as described [23], [31].
All lipids were from Avanti Polar Lipids. Liposomes were prepared as described in [32] and freshly extruded on 0.03, 0.1, or 0.4 µm filters as indicated. Extruded liposomes were stored at room temperature and used within two days.
Tryptophan fluorescence (λexc = 290 nm, λem = 340 nm) was used to follow GDP to GTP exchange as described [32]. All fluorescence measurements were performed using a Varian Carry Eclipse fluorimeter. Samples (800 µL) were thermostated at 37°C and continuously stirred. Nucleotide exchange kinetics in solution were measured with Δ17Arf1-GDP (1 µM) in 50 mM Tris pH 8.0, 50 mM NaCl, 2 mM β-mercaptoethanol, with RalF constructs concentrations in the 0.5–10 µM range for kcat/Km determination, or 1 µM for single kobs measurements. Nucleotide exchange assays with liposomes were done with myrArf1 (0.4 µM) in the presence of 200 µM liposomes, in 50 mM Hepes pH 7.4, 120 mM potassium acetate, 1 mM MgCl2 (HKM buffer) with RalF concentrations in the 0.05–0.3 µM range for kcat/Km determinations, or 0.1 µM for single kobs measurements. Nucleotide exchange was triggered by addition of 100 µM GTP. Activation rate constants (kobs, s−1) were determined by fitting the fluorescence changes to a single exponential using Kaleidagraph software. The catalytic efficiency kcat/Km (M−1s−1) was determined by linear fitting of kobs values as a function of the GEF concentration (in M). All experiments were done at least in triplicate.
For the immunogold electron microscopy analysis, 100 nM of His-tagged LpRalF was incubated with 50 µM extruded liposomes for 5 min in HKM buffer. Samples were applied to carbon-coated 400 mesh Nickel grids (Agar Scientific) for 1 min and subsequently blocked with HKM buffer containing 1% BSA for 45 min. The grids were then incubated for 1.5 hours with a mouse anti-His antibody (Qiagen) at a dilution of 1∶200. After batch washing with HKM buffer, the grids were incubated with 10 nm colloidal gold anti-mouse secondary antibody (Agar Scientific) for 1.5 hours and batch washed with HKM buffer. The grids were then stained with 2% uranyl acetate for 30 seconds. Images were acquired on a JEOL 1400 transmission electron microscope using low-dose conditions at 120 kV with a tungsten filament. Images were recorded using a Gatan 4k×4k CCD camera. We controlled that negatively stained grids containing only soluble RalF were labeled, but not grids containing only liposomes, indicating that immunogold labeling shown in Figure 1C is specific of liposome-bound RalF.
For the tryptophan fluorescence experiments, 2 µM RalF or RalF mutants was incubated with 500 µM liposomes in HKM buffer at 37°C under stirring prior to excitation at 297.5 nm. Tryptophan fluorescence scans were recorded for the buffer without or with liposomes, and were substracted from the scans recorded in the presence of wild type or mutant RalF.
Liposome sedimentation assays were done with RalF proteins (2 µM) incubated in 50 mM Tris pH 7.5, 120 mM NaCl, 1 mM MgCl2, 1 mM DTT for 10 min at room temperature with sucrose loaded fluorescent liposomes (39% PC, 20% NBD-PE, 25% PS, 1% PIP2, 15% cholesterol) extruded on 0.4 µm filter. After centrifugation for 20 min at 400000 g, liposome sedimentation was checked and quantified by fluorescence using a Fuji BioImager equipped with a CCD camera. Pellets were loaded on a 15% SDS-PAGE. Proteins were stained with Sypro-orange.
LpRalF and RpRalF capping domains affinity for lipids was assessed using commercially available membrane lipid strips (Echelon) as described in [14]. Briefly, membranes were incubated for 1 h at room temperature with purified MBP-tagged RalF, and the binding was visualized by chemiluminescence using anti-MBP antibodies.
Crystals of LpRalFF255K mutant were obtained by vapor diffusion in 0.1 M Hepes pH 7–8, 20–30%, PEG1000, 1.5 mM Fos-choline-12 and cryoprotected with a mix of parafin and silicon oils (50∶50 v∶v). A complete diffraction dataset at 3.1 Å resolution was collected at the Proxima-1 beamline (SOLEIL synchrotron, Gif-sur-Yvette, France) and integrated with the program XDS [33]. Crystals belong to space group P3121, with one molecule per asymmetric unit. The structure was solved by molecular replacement with the program Phaser [34] using wild-type Legionella RalF as a model (PDB entry 1XSZ, [13]). Refinement was carried out with the program BUSTER [35], in alternation with graphical building using Coot [36]. Data collection and refinement statistics are reported in Table S1. Atomic coordinates and structure factors have been deposited with the Protein Data Bank with accession code 4c7p. SAXS experiments were conducted on beamline SWING (SOLEIL Synchrotron, Gif-sur-Yvette, France) essentially as described in [37]. The histidine tags were cleaved to avoid noise in the SAXS data. Samples were prepared in 10 mM Tris pH 8.0, 150 mM NaCl, 2 mM β-mercaptoethanol, centrifuged at 16100 g for 30 min before the SAXS experiment and used at four concentrations (10, 5, 2.5 and 1.25 mg.mL−1). SAXS data were reduced with FOXTROT and analyzed with the ATSAS suite (EMBL, Hamburg, www.embl-hamburg.de/biosaxs/software.html). Modeling of LpRalF peripheral insertion in membranes was done using the PPM server (http://opm.phar.umich.edu/server.php) [38].
Legionella were harvested from 2-day heavy patch, and used to infect HEK293 cells stably expressing Arf1-GFP and the receptor Fcγ. This receptor allows L. pneumophila opsonized with anti-Legionella antibodies to be internalized efficiently by non-phagocytic cells [39]. Bacteria were opsonized with rabbit anti-Legionella antibody diluted 1/1000 for 30 min at 37°C. Bacteria were then added to the cells at a multiplicity of infection of 1. The cells were centrifuged 5 min at 1000 rpm and incubated at 37°C. Cells were then fixed with PFA for 20 min at room temperature, and stained for extracellular bacteria with blue anti-rabbit antibodies. Permeabilization was performed by treatment with cold methanol 1 min at RT before staining total bacteria with red anti-rabbit antibodies. The number of LCVs positive for Arf1-GFP was quantified. For kinetics assays of Arf1 recruitment to the LCV, we controlled that RalF and RalFmut were expressed, translocated and present in the host cell 6 hours post-infection at the same level.
Human Embryonic Kidney cells (HEK293) and HeLa cells were maintained in minimal Dulbecco's Modified Eagle's Medium, supplemented with 10% heat inactivated fetal bovine serum, 100 µg.mL−1 penicillin and 10 µg.mL−1 streptomycin at 37°C with 5% CO2. For transfection, Hela cells were plated at a density of 105 cells per well in 24-well tissue culture plates with glass coverslips and transfected the following day using effectene reagent (Qiagen). After transfection, cells were incubated for 24 hours then fixed with 3% PFA for 20 min at room temperature. Cells were permeabilized in Blocking Buffer (0.2% saponin, 0.5% BSA, 1% fetal calf serum in PBS) for 20 min. Coverslips were then washed with PBS and incubated with mouse anti-GM130 (BD Biosciences, diluted 1/1000 in Blocking Buffer) for 1 h at room temperature. Coverslips were then washed with PBS and incubated with anti-mouse TexasRed conjugated antibody at a dilution of 1/250 in Blocking Buffer for 1 h at room temperature. Finally, cells were washed in PBS and mounted on plain microscope slides. Cells were subsequently visualized by fluorescence microscopy using a Nikon Eclipse TE2000-S microscope and a 100×/1.40 oil objective (Nikon Plan Apo). Z-stacks were acquired using a Hamamatsu ORCA-ER camera and 3D max was generated. Images were exported to Image J and deconvoluted for the production of figures.
HEK293 cells were plated in 24-well dishes at a density of 3.104 cells per well. After 18 hours incubation, cells were cotransfected with 200 ng of plasmid encoding the indicated YFP-tagged protein and 300 ng of a plasmid encoding a secreted alkaline phosphatase (SEAP) protein. 24 hours after transfection, cells were washed, and fresh tissue culture medium was added. SEAP activity was measured 7 hours later, in the supernatant and in cells, using the Phosphalight SEAP kit (Applied Biosystems). The ratio of SEAP activity detected in the culture medium to the cells-associated SEAP activity is measured. Data are then normalized and compared to control cells, expressed as percent of control cell activity.
|
10.1371/journal.ppat.1003715 | Bacterial Effector Activates Jasmonate Signaling by Directly Targeting JAZ Transcriptional Repressors | Gram-negative bacterial pathogens deliver a variety of virulence proteins through the type III secretion system (T3SS) directly into the host cytoplasm. These type III secreted effectors (T3SEs) play an essential role in bacterial infection, mainly by targeting host immunity. However, the molecular basis of their functionalities remains largely enigmatic. Here, we show that the Pseudomonas syringae T3SE HopZ1a, a member of the widely distributed YopJ effector family, directly interacts with jasmonate ZIM-domain (JAZ) proteins through the conserved Jas domain in plant hosts. JAZs are transcription repressors of jasmonate (JA)-responsive genes and major components of the jasmonate receptor complex. Upon interaction, JAZs can be acetylated by HopZ1a through a putative acetyltransferase activity. Importantly, P. syringae producing the wild-type, but not a catalytic mutant of HopZ1a, promotes the degradation of HopZ1-interacting JAZs and activates JA signaling during bacterial infection. Furthermore, HopZ1a could partially rescue the virulence defect of a P. syringae mutant that lacks the production of coronatine, a JA-mimicking phytotoxin produced by a few P. syringae strains. These results highlight a novel example by which a bacterial effector directly manipulates the core regulators of phytohormone signaling to facilitate infection. The targeting of JAZ repressors by both coronatine toxin and HopZ1 effector suggests that the JA receptor complex is potentially a major hub of host targets for bacterial pathogens.
| Many Gram-negative bacterial pathogens rely on the type III secretion system, which is a specialized protein secretion apparatus, to inject virulence proteins, called effectors, into the host cells. The type III secreted effectors (T3SEs) directly target host substrates in order to promote bacterial colonization and disease development. Therefore, the identification and characterization of the direct host targets of T3SEs provides important insights into virulence strategies employed by bacterial pathogens to cause diseases. Here, we report that the plant pathogen Pseudomonas syringae T3SE HopZ1a physically interacts with and modifies the jasmonate ZIM-domain (JAZ) proteins in plant hosts. JAZ proteins are components of the receptor complex of the plant hormone jasmonates (JA) and key transcription repressors regulating JA-responsive genes. HopZ1a belongs to the widely distributed YopJ (for Yersinia Outer Protein J) family of T3SEs with a potential acetyltransferase activity. P. syringae producing HopZ1a, but not the catalytic mutant, leads to the degradation of AtJAZ1 during infection. As a result, HopZ1a activates JA signaling and promotes bacterial multiplication in Arabidopsis. This work provides the first example of a bacterial effector that subverts host immunity by directly targeting the receptor complex of a defense-associated hormone in plants.
| A prevailing concept for plant-pathogen interactions highlights the continuing battles between the activation of plant immune responses upon pathogen perception and the subversion of host immunity by virulence factors produced by successful pathogens. One branch of the plant immunity system is based on the recognition of pathogen- or microbe-associated molecular patterns (PAMP/MAMPs), which leads to a signal transduction cascade, and eventually PAMP-triggered immunity (PTI) [1]. PTI, broadly referred as basal defense in plants, restricts the growth of the vast majority of potential pathogens encountered by plants in the surrounding environment [2], [3]. However, successful pathogens produce virulence factors to effectively suppress PTI. For example, Gram-negative bacterial pathogens, such as Pseudomonas syringae, inject type III-secreted effectors (T3SEs) into the host cell to actively inhibit PTI [4], [5]. As a counter-attack strategy, plants have evolved nucleotide-binding leucine-rich repeat (NB-LRR) proteins to perceive specific T3SEs, directly or indirectly, and elicit effector-triggered immunity (ETI), which is often associated with localized programmed cell death at the infection sites [2], [3], [6].
Recent studies suggest that many P. syringae T3SEs suppress PTI and/or ETI by targeting important components of plant immunity [5], [7], [8]. Although the virulence targets of a few T3SEs have been characterized, the molecular mechanisms by which the majority of T3SEs subvert host resistance or facilitate nutrient acquisition remain elusive. HopZ1 is a P. syringae T3SE that belongs to the widely distributed YopJ family of cysteine proteases/acetyltransferases produced by both plant and animal bacterial pathogens [9]. The YopJ-like T3SEs share a conserved catalytic core, consisting of three key amino acid residues (histidine, glutamic acid, and cysteine), which is identical to that of clan-CE (C55-family) cysteine proteases [10]. However, several members of the YopJ effector family have been shown to possess acetyltransferase activity. YopJ and VopA modify their target proteins (mitogen-associated protein kinases and Ikkα/β) in animal hosts and the acetylation of these host targets blocks their phosphorylation and the subsequent defense signal transduction [11], [12]. PopP2 produced by the plant pathogen Ralstonia solanacerum has an autoacetylation activity, which is essential for its recognition in resistant plants; however, whether PopP2 can modify its target proteins in the host remains unknown [13].
Two functional HopZ1 alleles, HopZ1a and HopZ1b, have been identified in P. syringae [9]. HopZ1b is produced by P. syringae pv. glycinea (Pgy) strains, which are the causal agents of bacterial blight disease on soybean (Glycine max) [9]. HopZ1bPgyBR1 (HopZ1b in Pgy strain BR1; hereafter HopZ1b) promotes P. syringae multiplication in soybean; whereas the closely-related HopZ1aPsyA2 (HopZ1a in P. syringae pv. syringae strain A2; hereafter HopZ1a) triggers an HR in soybean cultivar Williams 82 and Arabidopsis thaliana accession Columbia-0 (Col-0, wild-type) [14]. HopZ1 mutants with the catalytic cysteine residues (C216 in HopZ1a or C212 in HopZ1b) substituted by alanines lose the virulence function or the HR-triggering activity, indicating that the functions of HopZ1 alleles require their enzymatic activities [9], [14]. In addition, HopZ1 has a potential N-terminal myristoylation site (Gly2) which directs the proteins to the plasma membrane [14], [15]. This myristoylation site of HopZ1a contributes to its avirulence function in both soybean and Arabidopsis [14], [15]. However, it is not clear whether this myristoylation site is important for the virulence function of HopZ1a. HopZ1 exhibited weak cysteine protease activities in vitro [9]. Recent studies showed that HopZ1a also possessed an acetyltransferase activity and could use tubulin as a substrate in vitro. Modification of tubulin is associated with the disruption of microtubule cytoskeleton, which may contribute to bacterial pathogenesis [16].
To identify potential host targets of HopZ1, we conducted yeast two-hybrid screening using a cDNA library of the natural host soybean and identified several HopZ1-interacting proteins (ZINPs). ZINP1 (2-hydroxyisoflavanone dehydratase, GmHID1) is a key enzyme in the soybean isoflavone biosynthetic pathway and a positive regulator of soybean basal defense. HopZ1 induces the degradation of GmHID1, and hence a decreased isoflavone production in soybean, resulting in increased plant susceptibility to bacterial infection [17]. HopZ1 also enhances bacterial infection in Arabidopsis, which does not have a putative ortholog of GmHID1. To understand the mechanisms underlying the virulence function of HopZ1a in Arabidopsis, we characterized another family of ZINPs, which were identified as jasmonate ZIM-domain (JAZ) proteins. JAZs are key transcriptional repressors of the jasmonate (JA) signaling pathway and major components of the JA receptor complex [18], [19], [20]. JA plays an important role in regulating plant responses to biotic and abiotic stresses. Some P. syringae strains produce the JA-mimicking phytotoxin coronatine, which efficiently activates JA signaling to facilitate bacterial entry into plant apoplastic space and suppress defense [21], [22], [23]. Therefore, HopZ1a may also target the JAZ proteins to promote bacterial infection. Consistent to this hypothesis, HopZ1a was previously reported to induce the expression of the JA/ethylene marker gene AtPDF1.2 in Arabidopsis, indicating that it could activate JA/ethylene signaling [24].
Here, we report that HopZ1a directly interacts with JAZ proteins of soybean and Arabidopsis. We show that HopZ1a induces the degradation of AtJAZ1, and promotes JA-responsive gene expression during P. syringae infection. Furthermore, HopZ1a functionally complements the growth deficiency of a P. syringae pv. tomato mutant that does not produce coronatine. All these activities depend on the intact catalytic core of HopZ1a, which acetylates JAZ proteins in vitro. Taken together, our results suggest that HopZ1a facilitates bacterial infection by manipulating the JA signaling pathway in Arabidopsis.
Using yeast two-hybrid screens, we identified the HopZ1a-interacting proteins (ZINPs) from a soybean cDNA library [17]. Among them, ZINP3 (Gm7g04630) was interesting because it shows significant homology to the Jasmonate ZIM-domain (JAZ) proteins. We designated ZINP3 as GmJAZ1 because it is most similar (51% similarity in full-length amino acid sequences and 62% similarity in the ZIM and Jas domains) to AtJAZ1 in Arabidopsis. GmJAZ1 was then further pursued as a direct target of HopZ1a.
We first confirmed the physical interaction between HopZ1a and GmJAZ1 by in vitro pull-down using recombinant GST-HopZ1a and GmJAZ1-HA proteins over-expressed in E. coli. GST-HopZ1a or GST (empty vector) was purified from whole cell lysate using glutathione resins and then incubated with an equal amount of whole cell lysate of E. coli expressing GmJAZ1-HA. GST-HopZ1a-bound resins, but not GST-bound resins, provided enrichment of GmJAZ1-HA (Fig. 1A), suggesting that HopZ1a interacted with GmJAZ1 in vitro. The catalytic mutant HopZ1a(C216A) also interacted with GmJAZ1, similar to wild-type HopZ1a (Fig. 1A).
We next examined the sub-cellular localization of GmJAZ1 to determine whether it co-localizes with HopZ1a in plant cells. GmJAZ1-YFP was expressed in Nicotiana benthamiana using Agrobacterium-mediated transient expression. Yellow fluorescence was examined in the pavement cells of the infiltrated leaves at 48 hours post inoculation (hpi) using confocal microscopy. Fluorescence was detected both on the plasma membrane and in the nucleus (Fig. S1). Previous studies reported that HopZ1a mainly locates on the plasma membrane with a sub-pool of HopZ1a in the nucleus [17]. These results suggest that GmJAZ1 and HopZ1a could co-localize in plant cells. To further confirm that HopZ1a indeed enters the nucleus, we performed nuclear fractionation of N. benthamiana cells expressing HopZ1a(C216A). The catalytic mutant HopZ1a(C216A) was used in this experiment, because the expression of the functional HopZ1a triggers cell death in N. benthamiana [9], [14]. Consistent with the previous confocal microscopy data [17], we detected the presence of HopZ1a(C216A) from both cytosolic and nuclear fractions (Fig. S2). These data confirmed that HopZ1a and GmJAZ1 co-localize in N. benthamiana cells.
We further used the bimolecular fluorescence complementation (BiFC) assay to determine the interaction between HopZ1a and GmJAZ1 in planta. HopZ1a(C216A) and GmJAZ1 were fused to the nonfluorescent N-terminal domain of YFP (1–155 aa, nYFP) and the C-terminal domain of YFP (156–239 aa, cYFP), respectively, at their C-termini. When the fusion genes were co-expressed in N. benthamiana, fluorescence was detected on the plasma membrane and in the nucleus (Fig. 1B), consistent with the subcellular localization of GmJAZ1 and HopZ1a. Taken together, these experiments demonstrate the interaction of HopZ1a and GmJAZ1 in vitro and in planta.
We have previously observed HopZ1-mediated degradation of another HopZ1-interacting protein GmHID1 when GmHID1 and HopZ1 were transiently co-expressed in N. benthamiana [17]. Therefore, we examined whether HopZ1a can also induce the degradation of GmJAZ1. GmJAZ1-FLAG and HopZ1a-HA were co-expressed in N. benthamiana, and the abundance of GmJAZ1 was determined at 20 hpi before the onset of visible cell death symptoms, which usually starts at 30 hpi. We chose 20 hpi because the expression level of GmJAZ1 was too low for protein analysis at earlier time points. A significant reduction of GmJAZ1 protein level was observed in N. benthamiana leaves co-expressing wild-type HopZ1a-HA, compared to leaves expressing the HopZ1a catalytic mutant or infiltrated with Agrobacterium carrying the empty vector (Fig. 2A). These results suggest that HopZ1a induces the degradation of GmJAZ1 in plant cells and the degradation requires the enzymatic activity of HopZ1a.
Incubation of GmJAZ1 and HopZ1a proteins purified from E. coli did not lead to observable changes in the abundance of GmJAZ1 (Fig. 1A). We suspected that a plant factor(s) might be required for this process and therefore performed a semi-in vitro degradation assay by incubating proteins extracted from N. benthamiana tissues expressing GmJAZ1 or HopZ1a individually. Total proteins extracted from leaves expressing GmJAZ1 or HopZ1a were mixed and incubated at 4°C for six hours before the abundance of GmJAZ1 was examined using western blots. Again, a significant decrease in GmJAZ1 protein level was observed in the presence of wild-type HopZ1a, but not the catalytic mutant HopZ1a(C216A) (Fig. 2B). These data suggest that HopZ1a induces GmJAZ1 degradation in plant cells.
To exclude the possibility that the reduced GmJAZ1 protein levels might have been resulted from cell death triggered by wild-type HopZ1a in N. benthamiana, we performed two control experiments. Firstly, we co-expressed the green fluorescence protein (GFP) with HopZ1a-HA or HopZ1a(C216A)-HA in N. benthamiana. The GFP protein levels remained unchanged in the presence of either wild-type or the catalytic mutant of HopZ1a (Fig. S3A). Secondly, we performed the semi-in vitro degradation assay of GmJAZ1 using AvrRpt2, which also elicits cell death in N. benthamiana [25]. Incubation with plant protein extracts expressing AvrRpt2 did not change the abundance of GmJAZ1 (Fig. S3B). This suggests that the reduced abundance of GmJAZ1 was not a result of HopZ1a-induced cell death in N. benthamiana.
Because GmJAZ1 is an ortholog of Arabidopsis JAZ proteins (AtJAZs), we examined whether HopZ1a also targets AtJAZs. Arabidopsis produces twelve JAZ orthologs (Fig. S4). Among them, seven were tested for their interactions with HopZ1a using in vitro pull-down. Our data showed that AtJAZ2, AtJAZ5, AtJAZ6, AtJAZ8 and AtJAZ12 interacted with HopZ1a in vitro (Fig. 3A). Although AtJAZ1 shares the highest sequence similarity with GmJAZ1 (Fig. S4), the interaction of AtJAZ1 with HopZ1a could not be determined because we were unable to express AtJAZ1 in E. coli at a level suitable for the pull-down assay.
We next confirmed the interaction between HopZ1a and AtJAZ6 in planta using BiFC. Similar to HopZ1a-GmJAZ1 interaction, yellow fluorescence was observed from plasma membrane and nucleus in cells co-expressing HopZ1a(C216A)-nYFP and AtJAZ6-cYFP (Fig. 3B). AtJAZ6 by itself was mainly located in the nucleus, but could also be detected in cytosol (Fig. S2). These data suggest that HopZ1a and AtJAZ6 co-localize and interact in plant cells.
Several effectors from the YopJ family, including HopZ1a, have been shown to possess acetyltransferase activities. To determine whether JAZs are substrates of HopZ1a, we performed in vitro enzymatic assay using C14-labeled acetyl-CoA. Recombinant HIS-SUMO-HopZ1a or HIS-SUMO-HopZ1a(C216A) proteins were expressed in E. coli and purified using nickel column. The HIS-SUMO tag was then removed by ubiquitin like protease 1 (ULP1). Tag-free HopZ1a or HopZ1a(C216A) proteins were incubated with purified HIS-GmJAZ1or MBP-AtJAZ6-HIS proteins in the presence of the cofactor inositol hexakisphosphate (IP6), and the acetylation of HopZ1a, GmJAZ1 and AtJAZ6 was detected by autoradiography as previously described [26]. Our experiments showed that both GmJAZ1 (Fig. 4A) and AtJAZ6 (Fig. 4B) were acetylated by wild-type HopZ1a, which also exhibited autoacetylation. The acetylation of GmJAZ1 appeared to be weaker in the autoradiograph compared to that of AtJAZ6. This is due to the low expression level of GmJAZ1 in E. coli, which only allowed us to use a much lower amount (1 µg), compared to AtJAZ6 (10 µg) in the reactions. Nonetheless, we consistently detected the acetylated form of GmJAZ1 when it was incubated with HopZ1a, but not HopZ1a(C216A), suggesting that GmJAZ1 and AtJAZ6 are both substrates of HopZ1a.
We sometimes could observe a background level of acetylation in tagged AtJAZ6 (MBP-AtJAZ6-HIS) when it was incubated with HopZ1a(C216A). Although this background acetylation was very weak compared to the acetylation of MBP-AtJAZ6-HIS, we decided to use the tag-free AtJAZ6 proteins to further confirm its acetylation by HopZ1a. Again, we observed strong acetylation of AtJAZ6 by HopZ1a, but not by HopZ1a(C216A) using only 5 µg of AtJAZ6 in the reaction (Fig. S5). These results demonstrate that GmJAZ1 and AtJAZ6 are acetylation substrates of HopZ1a.
JAZ proteins share three conserved domains: the C-terminal Jas motif [27], the ZIM domain in the central region [28], and a weakly conserved N-terminal region [19]. Because the conserved Jas domain is essential for the instability of JAZs in response to JA and the JA-mimicking phytotoxin coronatine [18], [19], we examined the impact of the Jas domain in the interaction between JAZs and HopZ1a. We constructed the mutant AtJAZ6ΔJas by deleting ten highly conserved amino acids (from seine191 to lysine200) within the Jas domain. In vitro pull-down assay showed that AtJAZ6ΔJas did not bind HopZ1a (Fig. 4C). Furthermore, AtJAZ6ΔJas was not acetylated by HopZ1a in vitro (Fig. 4D) or degraded by HopZ1a when these two proteins were co-expressed in N. benthamiana (Fig. 4E). These results demonstrate that HopZ1a-induced JAZ degradation requires direct interaction of HopZ1a with AtJAZ6, which is mediated by the Jas domain.
Although we observed the degradation of GmJAZ1 and AtJAZ6 when they were co-expressed with HopZ1a in N. benthamiana, it is important to examine whether HopZ1a can promote JAZ degradation during bacterial infection. For this purpose, we inoculated transgenic Arabidopsis plants expressing 35S-HA-AtJAZ1 with P. syringae producing HopZ1a or HopZ1a(C216A). The Arabidopsis pathogen Pseudomonas syringae pv. tomato strain DC3000 (PtoDC3000) is well-known to induce AtJAZ degradation through the production of coronatine, which acts as a JA mimic [22]. The mutant PtoDC3118 is deficient in coronatine production and therefore no longer degrades JAZs [29]. Importantly, PtoDC3118 expressing HopZ1a from its native promoter also significantly reduced the abundance of AtJAZ1 at 6 hpi (Fig. 5A). The level of AtJAZ1 remained unchanged in tissues infiltrated with PtoDC3118 carrying the empty vector or expressing the catalytic mutant HopZ1a(C216A). These data strongly suggest that HopZ1a can induce AtJAZ1 degradation during bacterial infection.
Because HopZ1a elicits HR in Arabidopsis ecotype Col-0, we performed two experiments to exclude the possibility that HopZ1a-triggered AtJAZ1 degradation was a result of plant cell death. First, we examined whether another effector AvrRpt2 could induce AtJAZ1 degradation. Although AvrRpt2 also triggers HR in Arabidopsis Col-0, the abundance of AtJAZ1 was unchanged when the HA-AtJAZ1-expressing plants were inoculated with PtoDC3118 expressing AvrRpt2 (Fig. 5A). Next, we generated the transgenic Arabidopsis line expressing 35S-HA-AtJAZ1 in the zar1-1 mutant background, which is abrogated in HopZ1a-triggered HR [30]. Again, the AtJAZ1 protein level was significantly reduced by HopZ1a (Fig. 5B), confirming that HopZ1a delivered by P. syringae leads to AtJAZ1 degradation in a cell death independent manner.
A major regulatory mechanism of JAZs in the presence of JA or coronatine is through COI-dependent ubiquitin-proteasome degradation. COI1 is an F-box protein that determines the substrate specificity of a Skp/Cullin/F-box (SCF) E3 ubiquitin ligase-SCFCOI1 [31]. Together with JAZ, COI1 is also a critical component of the JA co-receptor complex [19], [20], [22]. We examined whether COI1 is required for HopZ1a-induced JAZ degradation using the transgenic Arabidopsis line expressing 35S-HA-AtJAZ1 in the coi1-30 (SALK_035548) mutant background. As expected, PtoDC3000, which induces JAZ degradation through coronatine production, was unable to reduce the abundance of AtJAZ1 in the absent of COI1. Interestingly, PtoDC3118 expressing HopZ1a also no longer induced the degradation of AtJAZ1 in the coi1 mutant plants (Fig. 5C). These data suggest that, similar to coronatine- and JA-mediated AtJAZ degradation, COI1 is required for the degradation of AtJAZ1 by HopZ1a.
In Arabidopsis, JAZ proteins are repressors of JA transcription factors (e.g. AtMYC2) that are involved in the expression of JA-responsive genes [32], [33], [34]. Since HopZ1a induces the degradation of AtJAZ1, we examined whether it could induce the expression of JA-responsive genes during bacterial infection. Real-time RT-PCR was carried out to determine the transcript levels of JA-responsive genes in Arabidopsis. Five-week old zar1-1 plants were inoculated with PtoDC3118 expressing HopZ1a or HopZ1a(C216A) at OD600 = 0.2 (approximately 2×108 cfu/mL). The transcript levels of two early JA-responsive genes, AtJAZ9 and AtJAZ10 [34], were analyzed at 6 hpi. Both genes were induced approximately ten fold in plants infected by PtoDC3118(HopZ1a), whereas their expression was not changed in tissues infected by PtoDC3118 expressing HopZ1a(C216A) (Fig. 6A). The level of gene induction by HopZ1a was lower than that by coronatine, as shown by the approximately 40-fold induction of AtJAZ9 and AtJAZ10 in plants infected with PtoDC3000. This is consistent with the partial vs. complete degradation of AtJAZ1 by HopZ1a or coronatine during bacterial infection. Nonetheless, these experiments suggest that bacterium-delivered HopZ1a can activate JA signaling.
Recent findings showed that coronatine can suppress salicylic acid (SA) accumulation, probably as a consequence of the activation of JA signaling [35]. Because SA-associated defense confers resistance against biotrophic and hemibiotrophic pathogens, reduced SA accumulation would lead to defense suppression. In particular, coronatine is able to repress the expression of the SA synthetic enzyme isochorismate synthase gene 1 (AtICS1) in Arabidopsis. We then examined the impact of HopZ1a on the expression of AtICS1. Arabidopsis zar1-1 plants were inoculated with PtoDC3000 or PtoDC3118 carrying empty vector, HopZ1a, or HopZ1a(C216A) at OD600 = 0.2 (approximately 2×108 cfu/mL). Consistent with the prior findings, the transcript abundance of AtICS1 was reduced in plants infected with PtoDC3000 when compared to PtoDC3118 carrying the empty vector or HopZ1a(C216A) (Fig. 6B). The expression of AtICS1 was also reduced in plants inoculated with PtoDC3118 expressing HopZ1a, to a similar level as that in leaves inoculated with PtoDC3000. These data confirmed that, like coronatine, HopZ1a activates JA signaling and represses SA accumulation during bacterial infection.
Coronatine facilitates the infection of PtoDC3000 by activating JA signaling in Arabidopsis [21]. The coronatine-deficient mutant PtoDC3118 exhibits a significant reduction in bacterial population especially when the plants are infected by dipping inoculation [36]. Since HopZ1a also activates JA signaling, we examined whether HopZ1a could complement the growth deficiency of PtoDC3118. The Arabidopsis zar1-1 mutant plants were dipping-inoculated by PtoDC3000 or PtoDC3118 carrying the empty vector, HopZ1a, or HopZ1a(C216A). Three days post infection (dpi), the bacterial populations of PtoDC3118 carrying the empty vector or expressing HopZ1a(C216A) were approximately 200 fold lower than that of PtoDC3000 (Fig. 7A). Importantly, PtoDC3118 expressing wild-type HopZ1a multiplied to a significantly higher level (about 10 fold) than PtoDC3118 or PtoDC3118 expressing HopZ1a(C216A) (Fig. 7A). Although the population of PtoDC3118(HopZ1a) is lower than that of PtoDC3000, this partial complementation of the growth deficiency of PtoDC3118 is consistent with the partial degradation of AtJAZ1 (Fig. 5A) and the lower levels of JA-responsive gene induction (Fig. 6A) by PtoDC3118(HopZ1a) compared to PtoDC3000.
To further confirm that the function of HopZ1a is specifically related to its ability to activate the JA pathway, we introduced HopZ1a into PtoDC3000 and performed the same bacterial growth assay. HopZ1a was previously shown to enhance the infection of PtoDC3000 [30]. However, despite numerous trials, we did not observe any enhancement of HopZ1a on in planta multiplication of PtoDC3000. In fact, we consistently detected a decrease in the population of PtoDC3000(HopZ1a) compared to PtoDC3000 carrying the empty vector (Fig. 7A). Thus, HopZ1a can partially substitute for coronatine to promote bacterial infection.
Because COI1 is required for HopZ1a-induced degradation of AtJAZ1, we then examined whether COI1 is also required for the virulence activity of HopZ1a in Arabidopsis. For this purpose, we generated coi1-1, zar1-1 double mutant Arabidopsis plants, which were inoculated by PtoDC3118 carrying the empty vector, HopZ1a or HopZ1a(C216A). The bacterial populations of these three strains remained the same (Fig. 7B), confirming that the virulence activity of HopZ1a requires COI1. To further confirm that HopZ1a was unable to activate JA signaling in the mutant plants, we also determined the transcript levels of the JA-responsive genes AtJAZ9 and AtJAZ10, as well as the SA-biosynthetic gene AtICS1 after bacterial inoculation. Similar to PtoDC3000, PtoDC3118 expressing HopZ1a was also unable to induce the expression of AtJAZ9 and AtJAZ10 or suppress the expression of AtICS1 (Fig. S6). These data suggest that both the phytotoxin coronatine and the effector HopZ1a activate JA signaling in a COI1-dependant manner.
HopZ1a contains a potential N-terminal myristoylation site (Gly2) which may direct the association of HopZ1a with plasma membrane in plant cells [14], [15]. We therefore examined whether the Gly2 residue was required for the virulence function of HopZ1a. The bacterial population of PtoDC3118 carrying HopZ1a(G2A) was similar to that of PtoD3118 carrying wild-type HopZ1a in Arabidopsis zar1-1 plants (Fig. S7). This result demonstrates that HopZ1a(G2A) retained the virulence activity to promote PtoD3118 multiplication in Arabidopsis, indicating that the potential myristoylation site is not required for the virulence function of HopZ1a.
Taken together, our experiments showed that HopZ1a enhanced P. syringae infection in Arabidopsis in a manner similar to coronatine, which is a potent activator of JA signaling during bacterial infection.
T3SEs manipulate a variety of cellular processes in eukaryotic hosts for the benefit of pathogen infection. Emerging data suggest that bacterial pathogens have evolved various effectors to manipulate the signaling of JA and SA, which are important plant hormones that regulate defense response [37]. SA-dependent defense plays a major role in plant immunity against biotrophic and hemibiotrophic pathogens, such as Hyaloperonospora arabidopsidis and P. syringae [35], [37]. The P. syringae effector HopI1 directly targets Hsp70 in choloroplasts to suppress SA accumulation and thereby facilitate bacterial infection [38]. In addition, the Xanthomonas campestris effector XopD was also shown to repress SA signaling during bacterial infection of tomato [39], [40], [41]. Here, we report that the P. syringae effector HopZ1a represses SA accumulation, probably as an indirect effect of the activation of JA signaling.
In this study, we report that HopZ1 directly targets JAZs, the key negative regulators of JA signaling [18], [19]. Because JA signaling pathway is antagonistic to SA-dependent defense, activating JA signaling would be an attractive bacterial strategy to suppress host defense and facilitate pathogenesis of these pathogens. Indeed, recent studies have shown a remarkable example in which the P. syringae phytotoxin coronatine structurally mimics the active form of JA and targets the JAZ repressors for degradation to efficiently activate JA signaling [20], [22]. However, it has been rather puzzling that only a few P. syringae strains produce coronatine [42]. A previous study showed that a T3SE, AvrB, was also able to promote JA signaling, apparently through an indirect mechanism via the activation of MPK4 [43], [44]. We show here that the effector HopZ1a directly interacts with JAZs and at least some JAZs can be used by HopZ1a as substrates for acetylation. Importantly, HopZ1 mediates the degradation of AtJAZ1 in Arabidopsis and promotes bacterial infection in a COI1-dependent manner. This new finding raises the exciting possibility that JAZ repressors (hence the JA receptor complex) may be a common hub of host targets for diverse bacterial virulence factors. Consistent with this notion, oomycete pathogens were also found to produce effectors that interact with AtJAZ3 [45]. These pieces of evidence suggest that the JA receptor complex might be the Achilles' heel in plant defense system during the arms race with microbial pathogens.
HopZ1a enhances the in planta multiplication of PtoDC3118, but not that of PtoDC3000, in Arabidopsis. A weak growth enhancement of PtoDC3000 by HopZ1a was reported previously [30]. We were unable to replicate the published data, probably due to differences in experimental conditions. Our experiments, including the JA-responsive gene expression, JAZ protein degradation and bacterial in planta multiplication, consistently suggest that HopZ1a activates the JA signaling pathway in a manner similar to coronatine. However, HopZ1a only partially complements the function of coronatine. This could be because HopZ1a is not as potent as coronatine in inducing the degradation of JAZs. Coronatine has dual functions during the pre-entry and post-entry stages of bacterial infection [23], whereas the type III secretion genes are generally believed to be expressed after the bacteria have entered the apoplast [46]. Although it remains to be determined whether HopZ1a could promote stomata opening at the pre-entry stage, it is possible that HopZ1a mainly substitutes coronatine function inside the plant tissues. In addition, proteins might not be as stable as metabolites in planta, which may also explain the partial complementation of HopZ1a for the virulence deficiency of the coronatine mutant PtoDC3118.
As transcription regulators, JAZs are believed to function in the nucleus. However, HopZ1a was previously shown to mainly locate on the plasma membrane and this localization was mediated by a potential myristoylation site in the N-terminus [15]. Our protein-protein interaction and localization analyses showed that HopZ1a is also located in the nucleus and it interacts with JAZs both in the nucleus and on the cytosol/plasma membrane. Importantly, the mutant HopZ1a(G2A), which is abolished for its localization on the plasma membrane, was still able to promote PtoDC3118 infection. These data demonstrate that the membrane localization of HopZ1a is only important for host recognition [14], [15], but not required for virulence activity. This is consistent with the primary localization of JAZs in the nucleus as transcription repressors.
YopJ-like T3SEs produced by plant pathogens appear to have various enzymatic activities. AvrXv4 of Xanthomonas campestris was reported to be a SUMO protease [47]. AvrBsT, also from Xanthomonas, exhibited a weak cysteine protease activity in vitro [48]. PopP2 in R. solanacearum has autoacetylation and trans-acetylation activities in vitro, but it does not seem to acetylate its host target proteins [13]. Recently, HopZ1a was demonstrated to acetylate tubulin [16]. Our experiments showed that GmJAZ1 and AtJAZ6 are also substrates of HopZ1a. Importantly, we found that HopZ1a induces the degradation of AtJAZ1 during bacterial infection. In the presence of the active form of JA or coronatine, the F-box protein COI1 interacts with JAZs via the C-terminus Jas motif and recruits JAZs to the 26S proteasomes for degradation [18], [19], [22]. The fact that HopZ1a-mediated AtJAZ1 degradation is dependent on COI1 suggests that this degradation could also be dependent on the 26S proteasomes as a consequence of JAZ modification by HopZ1a. Post-translational modifications, including acetylation, have been shown to induce or repress proteasomal degradation. For example, in mammalian cells, the acetyltransferase ARD1 acetylates Hypoxia-inducible factor 1α (HIF-1α), which promotes its ubiquitination and proteasomal degradation [49]. Further investigations are needed to determine how HopZ1a-mediated acetylation of JAZs could facilitate COI1-dependent degradation of JAZ repressors.
Pseudomonas syringae, Agrobacterium tumefaciens and Escherichia coli strains were grown as described [50]. Bacteria strains and plasmids used in this study are summarized in Table S1.
To construct plasmids for bimolecular fluorescence complementation (BiFC) assay, full-length cDNA of GmJAZ1 or AtJAZ6 and hopZ1a(C216A) were cloned into the vectors pSPYCE and pSPYNE [51], respectively. To examine the subcellular localization of GmJAZ1, full-length cDNA was cloned into the vector pEG101[52]. The recombinant plasmids were introduced into Agrobacterium tumefaciens strain C58C1(pCH32), which were then used to infiltrate 3–4 week old Nicotiana benthamiana plants using a protocol described previously [14]. Functional fluorophore were visualized in the infiltrated leaves using a Leica SP2 Laser Scanning Confocal Microscope (Leica Microsystems) at 48 hpi for subcellular localization and BiFC. DAPI was used to stain the nucleus in plant cells [53], [54].
To construct GST-fusion plasmids, the full-length hopZ1a gene was inserted into the vector pGEX4T-2 (GE Healthcare Life Science). GmJAZ1-HA gene was cloned into the vector pET14b (Navogen), which has 6×His in the N-terminus. The AtJAZ genes were cloned into the plasmid vector pET-Mal with maltose binding protein (MBP) in the N-terminus and 8×His in the C-terminus [55]. In vitro pull-down assays were carried out using GST pull-down protein∶protein interaction kit (Pierce) according to the manufacturer's instruction. Briefly, GST or GST-HopZ1a was expressed in E. coli strain BL21(DE3). Soluble proteins were incubated with 50 µL glutathione agarose beads (Invitrogen) for one hour at 4°C. The beads were washed (20 mM Tris-HCl (PH 7.5), 150 mM KCl, 0.1 mM EDTA and 0.05% Triton X-100) five times and then incubated with equal amount of bacterial lysates containing JAZ proteins at 4°C for overnight. The beads were washed five times again, and the presence of the JAZ proteins on the beads was detected by western blots using anti-HA or anti-His antibodies conjugated with horseradish peroxidase (HRP) ) (Santa Cruz Biotechnology Inc.).
Pseudomonas syringae expressing the hopZ1a-HA genes was induced in M63 minimal medium containing 1% fructose at room temperature overnight [50]. HopZ1 expression was detected by western blots using the anti-HA antibody.
For JAZ degradation assay in N. benthamiana, hopZ1a-HA or 3×FLAG-hopZ1a were co-expressed with GmJAZ1-FLAG or AtJAZ6-YFP-HA using Agrobacterium-mediated transient expression as previously described [9], [56]. Leaf disks were collected at 20 hpi, and then grounded in 2×Laemmli buffer. The abundances of GmJAZ1 and AtJAZ6 were analyzed by western blots.
For the semi-in vitro protein degradation assay, GmJAZ1-FLAG and 3×FLAG-HopZ1a were over-expressed individually in N. benthamiana using Agrobacterium-mediated transient expression. Total proteins were extracted from the infected leaf tissues at 20 hpi using an extraction buffer containing 200 mM NaCl, 50 mM Tris (pH 7.6), 10% Glycerol, 0.1% NP-40, protease inhibitor cocktail (Roche), 10 mM DTT, 1 mM PMSF. Protein extracts were mixed in equal volume for six hours at 4°C with gentle agitation before the abundance of GmJAZ1 was examined by western blots.
For the in vivo JAZ degradation assay, six-week-old 35S-HA-AtJAZ1 transgenic Arabidopsis plants were hand infiltrated with bacterial suspensions of PtoDC3000 or PtoDC3118 carrying the empty vector (EV), expressing HopZ1a, HopZ1a(C216A) or AvrRpt2 at OD600 = 0.2 (approximately 2×108 cfu/mL). Leaves infiltrated with water were used as a negative control. Six hours post inoculation, total proteins were extracted from four leaf discs (0.5 cm2) in 100 µL of 2×SDS sample buffer. The lysates were incubated at 95°C for 10 minutes followed by centrifugation at 14,000 rpm for 5 minutes. The abundance of AtJAZ1 was then analyzed by western blots. Homozygous coi1-30 mutant plants were selected on 1/2×Murashige & Skoog (MS) medium supplemented with 50 µM JA. Seedlings that were insensitive to JA treatment, i.e. without inhibited root growth symptoms, were transplanted in soil and infected with P. syringae after six weeks.
HIS-GmJAZ1, HIS-SUMO-HopZ1a, HIS-SUMO-HopZ1a(C216A), HIS-SUMO-AtJAZ6, MBP-AtJAZ6-HIS, and MBP-AtJAZ6ΔJas-HIS were over-expressed in the E. coli strain BL21(DE3) and then purified using nickel resins. HIS-SUMO-HopZ1a and HIS-SUMO-AtJAZ6 proteins were then cleaved by ULP1 protease, producing protein mixtures with both HIS-SUMO and either tag-free HopZ1a or AtJAZ6. The protein mixtures were incubated with nickel resin again and the tag-free HopZ1a or AtJAZ6 proteins were collected from the flow through. In a standard acetylation assay, 2 µg purified HopZ1a or HopZ1a(C216A) was incubated with 10 µg MBP-AtJAZ6, 5 µg AtJAZ6 or 1 µg GmJAZ1 at 30°C for one hour in 25 µL of reaction buffer containing 50 mM HEPES (pH 8.0), 10% glycerol, 1 mM DTT, 1 mM PMSF, 10 mM sodium butyrate, 1 µL [14C]-acetyl-CoA (55 µci/µmol,) and 100 nM IP6. The reaction mixtures were then subjected to SDS-PAGE and acetylated proteins were detected by autoradiography as previously described [11], [26], [57] after exposure at −80°C for five days. After autoradiograph, the protein gels were removed from the filter paper and stained with Coomassie blue as a loading control.
The transcript abundances of AtJAZ9, AtJAZ10 or AtICS1 in Arabidopsis leaf tissues were analyzed by real-time RT-PCR using SYBR Green Supermix (BioRad Laboratories) and an CFX96 Real-Time PCR Detection System (BioRad Laboratories). Total RNA was isolated from three independent biological replicates using Trizol, and DNA was removed using DNase I (Fermentas). Reverse transcription was performed using M-MLV Reverse Transcriptase (Promega) with 1 µg of total RNA in a 25 µL reaction. The cDNAs were then used as templates for real-time PCR using gene-specific primers, which are listed below. AtActin was used as internal standard when compared the expression of AtJAZ9 and AtJAZ10 in different treatment. AtUBQ5 was the internal control used for the normalization of AtICS1 expression.
AtActin: 5′-GGTGTCATGGTTGGTATGGGTC-3′ and 5′-CCTCTGTGAGTAGAACTGGGTGC-3′
AtJAZ9: 5′-ATGAGGTTAACGATGATGCTG-3′ and 5′-CTTAGCCTCCTGGAAATCTG-3′
AtJAZ10: 5′-GTAGTTTCCGAGATATTCAAGGTG-3′ and 5′-GAACCGAACGAGATTTAGCC-3′
AtUBQ5:5′-GACGCTTCATCTCGTCC-3′ and 5′-GTAAACGTAGGTGAGTCC-3′
AtICS1: 5′-GGCAGGGAGACTTACG-3′ and 5′-AGGTCCCGCATACATT-3′
Arabidopsis plants were planted as previous described [30], [36]. The leaves of five-week old plants were dipped into the bacterial suspensions at an OD600 = 0.2 (approximately 2×108 cfu/mL) for 15 seconds. The inoculated plants were then transferred to a growth chamber (22°C and 16/8 light/dark regime, 90% humidity). Bacterial populations were determined as colony forming units (cfu) per cm2 using a previously described procedure [50].
Statistical analyses were performed using JMP 8.0 (SAS Institute Inc.).
|
10.1371/journal.ppat.1005052 | Latent KSHV Infected Endothelial Cells Are Glutamine Addicted and Require Glutaminolysis for Survival | Kaposi’s Sarcoma-associated Herpesvirus (KSHV) is the etiologic agent of Kaposi’s Sarcoma (KS). KSHV establishes a predominantly latent infection in the main KS tumor cell type, the spindle cell, which is of endothelial cell origin. KSHV requires the induction of multiple metabolic pathways, including glycolysis and fatty acid synthesis, for the survival of latently infected endothelial cells. Here we demonstrate that latent KSHV infection leads to increased levels of intracellular glutamine and enhanced glutamine uptake. Depletion of glutamine from the culture media leads to a significant increase in apoptotic cell death in latently infected endothelial cells, but not in their mock-infected counterparts. In cancer cells, glutamine is often required for glutaminolysis to provide intermediates for the tri-carboxylic acid (TCA) cycle and support for the production of biosynthetic and bioenergetic precursors. In the absence of glutamine, the TCA cycle intermediates alpha-ketoglutarate (αKG) and pyruvate prevent the death of latently infected cells. Targeted drug inhibition of glutaminolysis also induces increased cell death in latently infected cells. KSHV infection of endothelial cells induces protein expression of the glutamine transporter, SLC1A5. Chemical inhibition of SLC1A5, or knockdown by siRNA, leads to similar cell death rates as glutamine deprivation and, similarly, can be rescued by αKG. KSHV also induces expression of the heterodimeric transcription factors c-Myc-Max and related heterodimer MondoA-Mlx. Knockdown of MondoA inhibits expression of both Mlx and SLC1A5 and induces a significant increase in cell death of only cells latently infected with KSHV, again, fully rescued by the supplementation of αKG. Therefore, during latent infection of endothelial cells, KSHV activates and requires the Myc/MondoA-network to upregulate the glutamine transporter, SLC1A5, leading to increased glutamine uptake for glutaminolysis. These findings expand our understanding of the required metabolic pathways that are activated during latent KSHV infection of endothelial cells, and demonstrate a novel role for the extended Myc-regulatory network, specifically MondoA, during latent KSHV infection.
| KSHV is the etiologic agent of KS, the most common tumor of AIDS patients worldwide. Currently, there are no therapeutics available to directly treat latent KSHV infection. This study reveals that latent KSHV infection induces endothelial cells to become glutamine addicted, similarly to cancer cells. Extracellular glutamine is required to feed the TCA cycle through glutaminolysis, a process called anaplerosis. KSHV induces protein expression of the glutamine transporter SLC1A5 and SLC1A5 expression is required for the survival of latently infected cells. KSHV also induces the expression of the proto-oncogene Myc and its binding partner Max, as well as, the nutrient-sensing transcription factor, MondoA and its binding partner Mlx. MondoA regulates SLC1A5 and glutaminolysis during latent KSHV infection, and its expression is required for the survival of latently infected endothelial cells. These studies show that glutaminolysis and a single glutamine transporter, under the regulation of MondoA, are required for the survival of latently infected cells, providing novel druggable targets for latently infected endothelial cells. This work supports that a cancer-like metabolic signature is established by latent KSHV infection, opening the door to further therapeutic targeting specifically of KSHV latently infected cells.
| Kaposi’s Sarcoma-associated Herpesvirus (KSHV) is a human γ-herpesvirus and the etiologic agent of several malignancies, including two B-cell lymphomas, primary effusion lymphoma (PEL) and Multicentric Castleman’s Disease (MCD), as well as Kaposi’s Sarcoma (KS), an angioproliferative tumor[1, 2]. KS is the most common tumor of AIDS patients worldwide and also commonly occurs in non-AIDS patients in central Africa and the Mediterranean[2–4]. KS is a highly vascularized tumor comprised predominantly of spindle cells of endothelial origin. In both KS spindle cells and endothelial cells in culture, KSHV establishes a primarily latent infection, with only a small percentage of the tumor cells undergoing lytic replication[5, 6].
How KSHV alters endothelial cells to lead to cancer is still an open question. Previous work from our lab and others has demonstrated that KSHV, similarly to cancer cells, induces several major metabolic pathways. These alterations in cellular metabolism are imperative to the survival of cells latently infected with KSHV[7–9]. During latent KSHV infection, glucose uptake is induced and lactate production is significantly increased[7]. This switch to aerobic glycolysis is characteristic of the Warburg effect, a hallmark of cancer cell metabolism[10]. Interestingly, KSHV-infected endothelial cells require the Warburg effect for their survival, as latently infected endothelial cells are extremely sensitive to drug inhibition of glycolysis[7]. Recent evidence supports that the viral miRNAs expressed during latency are sufficient for the induction of the Warburg effect in KSHV-infected cells[11].
Our lab has also shown that KSHV induces the production of lipids via fatty acid synthesis (FAS) during latent infection[8]. Over half of the long-chain fatty acids detected in our metabolomics screen were elevated following latent KSHV infection. Lipid droplet organelles were also increased by latent KSHV infection of endothelial cells, evidence of increased fatty acid synthesis. Inhibition of FAS leads to apoptosis of KSHV-infected cells, which was rescued with supplementation of palmitate, a downstream metabolic intermediate of FAS. These data indicated that downstream intermediates of FAS are required for endothelial cell survival during latent infection. Induction of both glycolysis and FAS are also required in primary effusion lymphoma cells where KSHV is present[9].
Both the Warburg effect and increased FAS are metabolic signatures found in most cancer cells[12]. In these cells, glucose is primarily being utilized to produce lactic acid and fatty acids and is therefore diverted away from the tri-carboxylic acid (TCA) cycle. The TCA cycle metabolizes carbon to produce both bioenergetic and biosynthetic precursors. Importantly, glutamine carbon can be utilized to replenish the TCA cycle through a process termed anaplerosis[13]. Glutamine is the most abundant amino acid available to mammalian cells. Cancer cells induce glutamine uptake to support a glutamine requirement that exceeds the amount that cells can synthesize. Cancer studies have shown that transformed cells become glutamine addicted, or dependent on this exogenous glutamine and its catabolism via glutaminolysis for their survival[14, 15]. Recent evidence demonstrates that glutamine addiction in some cancers is enabled by the extended Myc network. Together, Myc-Max, with MondoA, a nutrient-sensing transcription factor, and its heterodimerization partner, the Max-like protein X (Mlx), facilitate the reprogramming of cellular metabolism in Myc-overexpressing cells [16–18].
A number of lytically replicating viruses also require glutamine for maximal viral replication[19–21]. Previous studies have shown that poliomyelitis virus and human cytomegalovirus depend on both glucose and glutamine for efficient virus replication[19, 21]. Interestingly, during vaccinia virus infection, glucose is completely dispensable for viral replication, but viral infection is reliant on glutamine for maximal virion production[20]. However, no studies have examined glutamine dependence during de novo KSHV infection.
We show that latent KSHV infection of endothelial cells induces glutamine uptake and that infected cells are dependent on the catabolism of glutamine for their survival. In the absence of exogenous glutamine, a significant percentage of KSHV-infected endothelial cells undergo apoptosis unless supplemented with TCA cycle intermediates such as alpha-ketoglutarate (αKG) or pyruvate. Targeted drug inhibition of glutamine uptake or glutaminolysis during latent infection recapitulates the findings from the glutamine-deprived conditions.
Additionally, we show that KSHV infection induces protein expression of c-Myc, its dimerization partner Max, MondoA, and its dimerization partner, Mlx. KSHV infection also induces protein expression of the glutamine-transporter protein SLC1A5. c-Myc coordinately with MondoA/Mlx is essential for regulation of glutaminolysis in cancer cells[16, 18] and is also necessary for the induction of SLC1A5 in KSHV-infected endothelial cells. Inhibition of MondoA or SLC1A5 induces cell death in KSHV-infected cells, but not mock-infected cells, and can be rescued with supplementation of αKG. Therefore, latent KSHV infection induces and requires glutamine uptake and subsequent glutaminolysis, regulated by MondoA and the glutamine transporter SLC1A5, for the survival of latently infected endothelial cells.
A global metabolomics screen identified that glutamine levels are significantly elevated at both 48 and 96 hours post latent KSHV infection[8]. Intracellular glutamine abundance is elevated 2.2 fold at 48 hours post infection (hpi) and 2.7 fold at 96 hpi, as compared to mock-infected cells (Fig 1A). To determine if the increased levels of glutamine in infected cells was due to increased uptake during latent infection, a radiolabeled glutamine molecule, [3H]-Glutamine, was added to the media of mock- and KSHV-infected Tert-Immortalized Microvascular Endothelial (TIME) cells at 96 hpi. Intracellular radiolabeled glutamine levels were then determined 10 min post treatment by scintillation. Latent KSHV infection induces glutamine uptake by approximately 35% compared to mock-infected cells (Fig 1B). These data validate that elevated levels of glutamine during latent KSHV infection are a result of an increase in exogenous glutamine uptake.
To determine if exogenous glutamine is a required carbon source for the survival of endothelial cells latently infected with KSHV, we quantified cell death over time in the presence or absence of exogenous glutamine. TIME cells were mock- or KSHV-infected and allowed to establish latency for 24 hours. Cells were re-seeded into 24-well plates, and overlaid with replete media, which contains 4mM glutamine, or glutamine-free media. Both treatment medias were prepared with dialyzed FBS, depleted of small molecules, including glutamine, and experiments were performed in triplicate. Average cell death over time was measured using the live-cell Essen Bioscience IncuCyte imaging system, which records both phase-contrast as well as fluorescent images over time. Dead cells were identified using the fluorescent nuclear dye YOYO-1, a cell impermeable dye that only enters cells with compromised membranes. Total cell number was determined by using SytoGreen24, a cell permeable dye that enters all cell nuclei. Percent cell death was calculated by dividing the total number of dead cells (YOYO-1 positive) by the total number of cells (SytoGreen24 positive). Cell death was monitored for 48 hours (24 hpi through 72 hpi). Fig 2A shows the average percent cell death recorded every 2 hours over 48 hours of monitoring for three biological replicate infections. The bar graph shows the average cell death at 0, 24, and 48 hours post treatment for each condition. In mock-infected cells, with replete or glutamine-deprived media, there is less than 5 percent cell death over the time monitored (Fig 2A). KSHV-infected cells in replete media have a slight increase in cell death over the time course. However, glutamine starvation of KSHV-infected cells induces a significant increase in cell death, approximately 25–30% after 48 hours of treatment (72 hpi). Microscopy images were analyzed for positive cell nuclei based on size and fluorescence intensity for both YOYO-1 and SytoGreen24. Representative images of YOYO-1 positive cells at 48 hours post treatment are shown in Fig 2B.
To ensure that glutamine addiction is not simply due to virus binding and entry, we repeated the experiments with UV-irradiated KSHV. UV-irradiated virus is able to bind and enter cells, but does not support viral gene expression. KSHV-infected cells show a 10-fold increase in cell death upon glutamine starvation, whereas UV-irradiated KSHV-infected cells show similar levels of cell death to mock-infected cells (Fig 2C). Therefore, KSHV viral gene expression is required to induce the dependence on glutamine and establish a state of glutamine addiction in endothelial cells.
To show that KSHV glutamine addiction was not limited to TIME cells, we conducted similar cell death experiments upon depletion of glutamine in mock- and KSHV-infected primary human dermal microvascular endothelial cells (1° hDMVECs). Mock- and KSHV-infected 1° hDMVECs were overlaid with glutamine depleted media at 48 or 72 hpi. Fourty-eight hours post glutamine depletion, YOYO-1 and SytoGreen24 counts were measured on a Typhoon scanner to determine relative fluorescence for cell death. These experiments reveal a significant increase in cell death only in KSHV-infected 1° hDMVECs, substantiating that KSHV infection of 1° hDMVECs also induces glutamine addiction (Fig 2D).
We have previously shown that inhibition of glycolysis and fatty acid synthesis leads to cell death via apoptosis of endothelial cells latently infected with KSHV[7, 8]. It has also been shown that glutamine deprivation leads to apoptosis of cancer cells[17]. To determine if glutamine starvation induces cell death via activation of an apoptotic pathway, we performed the previously described cell death assay using YOYO-1 and SytoGreen24 counts in the presence or absence of the pan-caspase inhibitor QVD. Upon supplementation with QVD, KSHV-infected cells deprived of glutamine are rescued from cell death (Fig 3A), indicating that cell death is due to caspase-dependent apoptosis.
To confirm that cells are dying via apoptosis, we utilized a fluorogenic Caspase-3/7 substrate, which contains the caspase cleavage site, a short four amino acid peptide (DEVD), conjugated to a nucleic acid binding dye. This cleavage site is specifically targeted by activated executioner caspases 3 and 7. When caspase 3 and/or 7 are activated during apoptosis, the DEVD site is cleaved, resulting in the release of the DNA dye, translocation to the nucleus and fluorescence. For these experiments, the Caspase-3/7 substrate was added to mock- and KSHV-infected cells in the presence or absence of glutamine at 24 hpi. After 48 hours of treatment, plates were scanned for relative fluorescence using a Typhoon 9400 variable mode imager. Caspase-3/7-mediated relative fluorescence was normalized to SytoGreen24 relative fluorescence from the same experiment. Only KSHV-infected cells starved of glutamine showed significant detection of fluorescence from the Caspase-3/7 substrate, indicating that latent infection induces Caspase 3 and/or 7 activation, which in turn results in an elevated level of DEVD cleavage and nuclear fluorescence (Fig 3B). We also included samples supplemented with QVD. These samples showed no increased fluorescence even in the absence of glutamine (Fig 3B). Representative microscopy images at 48 hours post treatment were captured with the Cellomics ArrayScan Vti (Fig 3C). Overall, these data indicate that when deprived of glutamine, KSHV-infected endothelial cells activate apoptosis in a Caspase-3/7 dependent manner.
Upon entering the cell, glutamine is catabolized via glutaminolysis. Glutaminolysis consists of two consecutive deamination steps. First, glutamine is converted to glutamate by glutaminase (GLS). Second, glutamate is converted to αKG by one of three enzymes: glutamate dehydrogenase (GDH), glutamate pyruvate transaminase (GPT) or glutamate oxaloacetate transaminase (GOT)[14, 22]. At this stage, αKG can enter and replenish the TCA cycle. To determine if glutamine is required to maintain the TCA cycle in KSHV-infected cells, we added either membrane-soluble αKG or pyruvate, both of which can enter the TCA cycle, to the treatment medium of glutamine-deprived cells during latent KSHV infection. After mock- or KSHV-infection of TIME cells for 24 hours to allow the establishment latency, cells were re-seeded as before and overlaid with replete media, glutamine-free media, or glutamine-free media supplemented with either 3.5 mM αKG or 8 mM pyruvate. Supplementation with αKG completely rescues the glutamine-deprived KSHV-infected cells from cell death and supplementation with pyruvate significantly rescues cell death in the glutamine-deprived infected cells (Fig 4A). These metabolite rescue data support the model that the exogenous glutamine taken up by KSHV-infected endothelial cells is necessary to support glutaminolytic metabolism for replenishment of the TCA cycle.
BPTES is a specific inhibitor of GLS, the first enzyme of glutaminolysis. When treated with BPTES in the presence of 4mM glutamine (replete media), KSHV-infected endothelial cells died at similar levels to those deprived of glutamine, while having little effect on mock-infected cells (Fig 4B). These data recapitulate our findings with glutamine-deprived media. Taken together, these data validate that glutamine is essential for glutaminolysis in KSHV-infected cells.
Glutamine metabolism is regulated by oncogenic c-Myc in many cancer cells[16, 17, 23]. Additionally, there is evidence that c-Myc is regulated by latent KSHV infection[24, 25]. Recently, it was shown that c-Myc[26], and N-Myc[18], manipulate metabolic gene expression coordinately with the Myc-bHLHZ superfamily members MondoA, a nutrient-sensing protein, and its dimerization partner, Mlx. MondoA/Mlx or the paralogue ChREBP/Mlx constitute the “nutrient-sensing” arm of the extended Myc network[27]. Protein expression of c-Myc, Max, MondoA and Mlx are increased during latent KSHV infection of TIME cells, as determined by immunoblot analysis of whole cell lysates harvested at 48 hpi (Fig 5). Additionally, a known target of activated MondoA/Mlx, TXNIP, is upregulated at the protein level during latent KSHV infection (Fig 5).
It has been shown that Myc/MondoA controls glutamine metabolism by inducing the expression of the major glutamine transporter, SLC1A5[18]. SLC1A5 is a neutral amino acid transporter which localizes to the cellular membrane, and is known to primarily import glutamine[28]. SLC1A5 is upregulated in many cancer cells [16, 18, 28]. There is a small, but reproducible, increase in SLC1A5 protein in TIME cells latently infected with KSHV when whole cell lysates are compared by immunoblot analysis at 48 hpi (Fig 5). Together, these data suggest that latent KSHV infection induces changes to the Max/Mlx-regulation network consistent with coordinate regulation of metabolism, including glutamine uptake through SLC1A5.
To determine the role of the glutamine transporter SLC1A5 during latent infection, the SLC1A5 specific inhibitor L-γ-Glutamyl-p-nitroanilide (GPNA) was used [29]. Mock- and KSHV-infected TIME cells were re-seeded at 24 hpi and overlaid with replete media or replete media treated with 5mM GPNA. YOYO-1 or SytoGreen24 were added to compare the relative florescence of dead cells and the relative fluorescence of total cells, respectively, at 48 hours post treatment. These experiments were conducted using the Typhoon 9400 variable mode imager to measure relative fluorescence of all samples. GPNA treatment leads to increased cell death only in KSHV-infected cells but not their mock counterparts (Fig 6A). Importantly, when supplemented with 3.5 mM αKG, cell death induced by GPNA treatment of KSHV-infected endothelial cells was rescued to KSHV replete control treatment levels, indicating that the drug-induced cell death was due to the requirement of glutamine metabolism via glutaminolysis and not off-target effects.
To further confirm the drug studies, a validated siRNA set directed to SLC1A5 was used to knockdown SLC1A5 expression[18]. SLC1A5 expression was reduced by approximately 70% in TIME cells transfected with a mix of four siRNAs specific for SLC1A5 (siSLC1A5), as compared to cells transfected with a scrambled non-target control (siControl) (Fig 6B). Twenty-four hours post transfection with the SLC1A5 or control siRNA, cells were either mock- or KSHV-infected and subsequently provided replete media containing YOYO-1 for cell death or SytoGreen24 to identify all cells. Plates were scanned at 48 hours post treatment (72 hpi) for relative fluorescence on the Typhoon 9400 variable mode imager. Minimal cell death was observed in both mock- and KSHV-infected cells treated with siControl. KSHV-infected cells, but not mock-infected cells, transfected with the siSLC1A5 show an increase in cell death. The fold change in relative fluorescence for cell death of KSHV-infected cells over mock-infected cells is increased in cells transfected with siSLC1A5 compared to cells transfected with siControl (Fig 6B). Together, these data support that KSHV-infected endothelial cells rely on the expression of the glutamine transporter SLC1A5 for survival.
SLC1A5 is directly regulated by the nutrient-sensing Myc extended network member MondoA in many human cancer cells[18]. To determine if MondoA controls SLC1A5 expression during latent KSHV infection of endothelial cells, we examined the expression of SLC1A5 upon siRNA knockdown of MondoA in mock- and KSHV-infected endothelial cells. MondoA protein expression was significantly reduced in both mock and KSHV-infected TIME cells transfected with a mix of four siRNAs specific for MondoA (siMondoA), as compared to cells transfected with a scrambled non-target control (siControl) (Fig 7A). While SLC1A5 protein levels are elevated by KSHV infection, loss of MondoA results in a reduction in detected SLC1A5 in all samples. Additionally, protein levels of Mlx, a co-stabilized MondoA binding partner[18], and TXNIP, a known downstream target of MondoA/Mlx regulation, are also reduced upon loss of MondoA. These data support the hypothesis that MondoA is directly regulating SLC1A5, the major glutamine transporter, during latent KSHV infection of endothelial cells.
To determine if MondoA is required for endothelial cell survival during latent KSHV infection, we examined cell death in the presence of control siRNA or siRNA directed against MondoA. As shown in Fig 7B, only KSHV-infected endothelial cells in the absence of MondoA show a significant increase in cell death at 48 hpi, indicating that MondoA is indeed required for the survival of latently infected cells. Importantly, this significant increase in cell death is fully rescued upon supplementation with αKG, indicating that the cell death that occurs in KSHV-infected cells where MondoA is knocked down is due to a loss of TCA cycle intermediates and not an unrelated function of MondoA.
Transformed cells were first described as ‘glutamine addicted’ in the 1950’s[15]. It is now well established that glutamine, the most abundant amino acid in plasma, is ‘conditionally essential’ for cancer cell growth and survival[13]. More recent evidence shows that lytically replicating viruses orchestrate specific cellular metabolic modifications to support the unique requirements for their viral replication[19, 20, 30–32]. We demonstrate that latent infection with KSHV, an oncogenic virus, induces glutaminolysis in endothelial cells. In addition to showing that latent KSHV infection enhances glutamine uptake during infection, we have shown that a significant percentage of latently infected endothelial cells become glutamine addicted, and that glutaminolysis is required for the survival of these cells. Deprivation of glutamine in both TIME cells and 1°hDMVECs leads to significant increases in apoptosis unless they are supplemented with TCA cycle intermediates.
Glutaminolysis is an important anaplerotic reaction that produces αKG, which can enter the TCA cycle (Fig 8). TCA cycle intermediates support the production of both bioenergetic and biosynthetic precursors; therefore, glutamine is potentially required for a variety of downstream cellular processes including ATP and NADPH production and fatty acid synthesis[33]. There is substantial evidence in cancer biology that glutamine metabolism is required to replenish the TCA cycle when glucose is being metabolized to lactic acid as part of the Warburg effect[13]. Previous research from our lab has shown that induction of the Warburg effect is required for the survival of endothelial cells during latent KSHV infection. Therefore, we were interested in the role glutamine metabolism may play in KSHV-infected endothelial cells. Human cytomegalovirus and vaccinia virus require glutamine to support the TCA cycle for maximal virus replication and media supplemented with TCA cycle intermediates, such as αKG or pyruvate, rescued replication in the absence of glutamine[19, 20]. Our data supports that glutamine is a vital carbon source during latent infection with KSHV.
A recent study reported an increase in glutamate secretion during latent KSHV infection[25]. Glutamate is produced intracellularly through the deamination of glutamine (Fig 8). When glutamate secretion was inhibited, cell proliferation was reduced; however, apoptosis was not reported upon treatment with glutamate secretion inhibitors. Therefore, the increase in glutamine uptake that we observe during latent KSHV infection could be supporting the pleiotropic role of glutamine during infection to support multiple cellular processes, including anaplerosis to support the TCA cycle as well as signaling to the extracellular environment.
We demonstrate that the glutamine transporter SLC1A5 is upregulated during latent KSHV infection of endothelial cells, and that specifically the latently infected cells are dependent upon SLC1A5 for survival. This was of specific interest because previous studies have shown that oncogenic c-Myc, or N-Myc, induces increased expression of the glutamine transporter SLC1A5, and dependency upon it for survival in Myc-activated cells[26]. Multiple studies have reported that c-Myc is upregulated during KSHV infection[25, 34]. We observed an upregulation in c-Myc during infection of endothelial cells, but also identified a significant upregulation in the related proteins MondoA and Mlx. These proteins are a part of the expanded Myc network, known as the Max/Mlx network. MondoA and Mlx form an important glucose-responsive heterodimer that participates in regulating cellular metabolism, specifically glucose, lipid and glutamine metabolism in collaboration with c-Myc or N-Myc. It was recently described that both Myc overexpression and MondoA expression are required to induce the expression of glutamine transporters, including SLC1A5, as well as induce glutaminolysis[18]. We find that MondoA regulation is required for the survival of latently infected endothelial cells and that supplementation with αKG, the immediate downstream intermediate of glutaminolysis and TCA cycle metabolite, promotes cell survival, similarly to our findings upon glutamine deprivation. This is the first evidence of the requirement for MondoA metabolic regulation during human viral infection.
While we have delineated the cellular mechanism of KSHV-induced glutamine addiction, the latent viral gene or set of genes sufficient to induce the MondoA-mediated metabolic switch to glutamine addiction has not yet been determined. Previous research has identified that the latent KSHV protein LANA collaborates with Myc to stabilize and activate the transcriptional regulator during infection[34]. However, this story may be more complicated. It was recently shown that expression of the latent KSHV miRNA cluster is sufficient to induce glucose uptake and glycolysis[11]. If alterations in glucose and glutamine metabolism are interconnected, such as a requirement for glucose to activate MondoA/Mlx, it is likely that multiple viral genes are involved and more work is needed to identify which latent factors are necessary to activate the overall metabolic signature that is required during latent KSHV infection of endothelial cells.
Several major metabolic switches are required during latent KSHV infection; however, the question remains whether induction of cancer cell metabolism is pre-adapting cells for a cancer microenvironment or if these alterations are helping drive oncogenesis when cells are placed in the correct microenvironment. Our findings are in agreement with metabolic signatures described by many cancer studies, which would be predicted if latent KSHV infection is indeed predisposing cells for oncogenesis. However, these models are not necessarily mutually exclusive. Comparing induced metabolic phenotypes, such as the Warburg effect and glutamine addiction in a viral system, where we can include mock controls, provides a unique model to identify the initial drivers of oncogenesis as well as characterize the suitable microenvironment established. Glutamine addiction may be induced early in oncogenesis, yet also be a characteristic of long-term tumor maintenance. Drug inhibitors specifically targeting glutamine-addicted cells could also provide novel therapeutic treatments to specifically target endothelial cells latently infected with KSHV.
Tert-Immortalized Microvascular Endothelial (TIME) cells [35] and primary human dermal microvascular endothelial cells (1° hDMVECs) (Lonza, MD) were maintained as monolayer cultures in EBM-2 media (Lonza or Cellgro) or EndoGrow (Millipore) supplemented with a bullet kit containing 5% FBS, vascular endothelial growth factor, basic fibroblast growth factor, insulin-like growth factor 3, epidermal growth, and hydrocortisone. Millipore EndoGrow media, supplemented with dialyzed FBS (depleted of small molecules including glucose and glutamine) was used for all experiments that compare replete (4 mM glutamine) and glutamine-free media. KSHV inoculum from induced BCBL-1 cells was titered and used to infect TIME cells or 1° hDMVECs as previously described[36]. Infections were performed in serum-free EBM-2 media and subsequently overlaid with complete EBM-2 media. Infection rates were assessed for each experiment by immunofluorescence and only experiments where greater than 85% of the cells expressed LANA, a latent marker, and less than 1% of the cells expressed ORF59, a lytic marker, were used. In a subset of the siRNA transfection experiments where larger quantities of siRNA were used, there was a slight increase in the cells expressing ORF59, but this always occurred in both the control and gene specific siRNA transfections and did not alter the results of the experiments.
YOYO-1 and SytoGreen24 were diluted in DMSO and used at a final concentration of 100 nM and 50 nM respectively (Life Technologies). Dimethyl-α-ketoglutarate (alpha-ketoglutarate) and pyruvate were purchased from Sigma and used at 3.5 mM and 8 mM respectively. Bis-2-(5-phenylacetamido-1,3,4-thiadiazol-2-yl)ethyl sulfide, or BPTES (Sigma) was solubilized in DMSO, subsequently diluted in methanol and used at a final concentration of 2.5 μM. QVD-OPH (SMBiochemicals) and was dissolved in DMSO and used at a final concentration of 20 μM. L-γ-Glutamyl-p-nitroanilide (GPNA) (Sigma), was prepared in DMSO in a 1 M stock solution and used at a final concentration of 5mM.
Twenty-four hours post mock- or KSHV-infection; TIME cells were re-seeded into 12-well plates at equal numbers in triplicate. At 96 hpi, cells were overlaid with serum-free media for 2 hours. Cells were then washed three times with PBS before the addition of 1mL of serum-free media containing 0.5μCi (10 pmol) of [3H]-L-glutamine (Perkin Elmer #NET551). Cells were incubated for 10 minutes at 37°C. Following incubation, the medium was removed and each well was washed twice with 1mL of ice-cold DPBS and 200μL of lysis buffer (1% SDS in PBS) was added to each well and incubated at room temperature with occasional agitation for 5 minutes. Lysates were transferred to microcentrifuge tubes and mixed by vortexing. 150μL of each lysate was transferred to a vial containing 4mL of Biofluor Plus scintillation fluid (Perkin Elmer). Each vial was mixed by vortexing and counted in a Beckman LS6500 liquid scintillation counter. The remaining lysate was quantified by BCA Protein Assay Reagent Kit (Pierce) for normalization.
At 20 hpi, mock- and KSHV-infected TIME cells were re-seeded into 24-well plates. At 24 hpi, cells treated with Replete (4 mM glutamine), glutamine-free media or replete media with 2.5 μM BPTES in triplicate. Of note, no changes in latent or lytic infection rates were observed after glutamine starvation. YOYO-1, to identify dead cells, or SytoGreen24, to mark all cell nuclei, were added at this step. For rescue studies, supplementation with 3.5 mM αKG, 8 mM pyruvate or 20 μM QVD were added at this step. Plates were then placed on the IncuCyte (Essen Biosciences), a live-cell phase-contrast and fluorescent imaging system and recorded for cell death and total cell number for 48 hours (24 hpi through 72 hpi). GPNA experiments were prepared according to the same protocol, but scanned on the Typhoon 9400 variable mode imager (GE Healthcare) and analyzed with ImageJ software for relative fluorescence at 48 hours post treatment. Apoptosis experiments conducted with the apoptosis marker, Caspase-3/7 substrate, were prepared according to the same protocol, but the Caspase-3/7 Cell Event reagent was added, plates were scanned at 48 hours post treatment on the Typhoon 9400 and ImageJ software and normalized to relative florescence for Styogreen24. Primary hDMVEC experiments were re-seeded into 24-well plates and at 48 or 72 hpi were treated with Replete (4 mM glutamine) or glutamine-free media and 48 hours post treatment (96 or 120 hpi) were scanned on the Typhoon 9400 variable mode imager (GE Healthcare) and analyzed with ImageJ software for relative fluorescence.
All cells were lysed in RIPA and protein was quantified using BCA Assay (Pierce). 30–50ug were subjected to SDS-PAGE in 1xMES Buffer (Life Technologies) on a 4–12% NuPAGE Bis-Tris Gel (Life Technologies) then transferred to 0.2um nitrocellulose membrane (Bio-Rad). The membranes were blocked in 5% Non-Fat Dry Milk in TBS with 0.1% Tween (TBST) for at least an hour then probed with the indicated primary antibodies diluted in 5% milk in TBST for 2 hours at RT, or overnight at 4C (anti-c-Myc (Abcam), anti-Max (Santa Cruz Biotechnology), anti-MondoA (Proteintech), anti-Mlx (Santa Cruz Biotechnology), anti-SCL1A5 (Cell Signaling) and anti-TXNIP (MBL, JY1). Blots were washed 3 times in TBST, then probed with HRP-conjugated secondary antibody (Cell Signaling) diluted in 5% milk in TBST for 1 hour at RT. Blots were washed 3 times in TBST, then subjected to chemiluminescence and exposed to blue autoradiography film (GeneMate) and processed in an autoprocessor.
Total RNA was isolated from TIME cells 72 hours post siRNA transfection using the Nucleospin RNA II Kit (Macherey-Nagel). Two-step quantitative real-time reverse transcription PCR (BioRad) was used to measure expression levels of SLC1A5 and the housekeeping gene GAPDH. iScript Reverse Transcription Supermix and SsoAdvanced SYBR Green Supermix (BioRad) were used according to manufacturer’s protocols. The primers used were: SLC1A5-F ‘5-TTATCCGCTTCTTCAACTCCTT-3’, SLC1A5-R ‘5-ACATCCTCCATCTCCACGAT-3’, or GAPDH-F: ‘5-GGACTCATGACCACAGTCCA-3’, GAPDH-R ‘5-CCAGTAGAGGCAGGGATGAT-3’. Relative levels of SLC1A5 mRNA were normalized by the delta threshold cycle method to the abundance of GAPDH mRNA.
A set of four siRNAs specific to the glutamine transporter SLC1A5 (siSLC1A5) and MondoA (siMondoA) were purchased (Qiagen, Flexitube GeneSolution #GS6510 and #GS22877 respectively). A negative-control siRNA (siControl) was designed and synthesized by Ambion. TIME cells were transfected with siRNA at a final concentration of 200 nM, using the Amaxa Nucleofector Kit by Lonza according to the manufacturer’s protocol. At 24 hours post transfection, cells were mock- or KSHV-infected. Upon completion of the infection, cells were washed and treated with Replete media containing YOYO-1 or SytoGreen24. Relative fluorescence was measured 48 hours post treatment using a Typhoon 9400 variable mode imager (GE Healthcare) and ImageJ software.
|
10.1371/journal.ppat.1005155 | The NLRP3 Inflammasome and IL-1β Accelerate Immunologically Mediated Pathology in Experimental Viral Fulminant Hepatitis | Viral fulminant hepatitis (FH) is a severe disease with high mortality resulting from excessive inflammation in the infected liver. Clinical interventions have been inefficient due to the lack of knowledge for inflammatory pathogenesis in the virus-infected liver. We show that wild-type mice infected with murine hepatitis virus strain-3 (MHV-3), a model for viral FH, manifest with severe disease and high mortality in association with a significant elevation in IL-1β expression in the serum and liver. Whereas, the viral infection in IL-1β receptor-I deficient (IL-1R1-/-) or IL-1R antagonist (IL-1Ra) treated mice, show reductions in virus replication, disease progress and mortality. IL-1R1 deficiency appears to debilitate the virus-induced fibrinogen-like protein-2 (FGL2) production in macrophages and CD45+Gr-1high neutrophil infiltration in the liver. The quick release of reactive oxygen species (ROS) by the infected macrophages suggests a plausible viral initiation of NLRP3 inflammasome activation. Further experiments show that mice deficient of p47phox, a nicotinamide adenine dinucleotide phosphate (NADPH) oxidase subunit that controls acute ROS production, present with reductions in NLRP3 inflammasome activation and subsequent IL-1β secretion during viral infection, which appears to be responsible for acquiring resilience to viral FH. Moreover, viral infected animals in deficiencies of NLRP3 and Caspase-1, two essential components of the inflammasome complex, also have reduced IL-1β induction along with ameliorated hepatitis. Our results demonstrate that the ROS/NLRP3/IL-1β axis institutes an essential signaling pathway, which is over activated and directly causes the severe liver disease during viral infection, which sheds light on development of efficient treatments for human viral FH and other severe inflammatory diseases.
| The NLRP3 inflammasome and IL-1β play essential roles in mediating the primary inflammatory responses against pathogen invasions in the host. Hyperactivation of this signaling pathway can lead to life-threatening diseases under certain circumstances. However, it is not clear if NLRP3 inflammasome activation participates in the pathogenesis of viral fulminant hepatitis (FH), a clinical severe syndrome characterized by acute inflammation in the liver along with massive necrosis of hepatocytes and hepatic encephalopathy during viral infection. Using a mouse viral FH model by infection with murine hepatitis virus strain-3 (MHV-3), we observed a significant macrophage induction and the serum and liver massive accumulation of IL-1β. Conversely, interruption of IL-1β signals results in attenuation of the MHV-3-induced hepatitis and mortality. Blocking IL-1β activity reduces the virus-induced expression of fibrinogen-like protein-2 (FGL2) in macrophages, and limits the liver recruitment of CD45+Gr-1high neutrophils upon the virus infection. We further show that proIL-1β is bioprocessed by NLRP3 inflammasome. Deletion of the components in the inflammasome complex, including NLRP3 and Caspase-1, leads to reduction in the virus-induced IL-1β production and lessening of disease progression. Further studies show that macrophages in deficiency of nicotinamide adenine dinucleotide phosphate (NADPH) oxidase subunit p47phox, a protein that controls acute ROS production, prevents NLRP3 inflammasome activation and IL-1β secretion, suggesting that the virus-induced ROS production can directly initiate NLRP3 inflammasome activation. Therefore, p47phox-/- mice exhibited certain degrees of MHV-3 resistance. Taken together, these results demonstrate that ROS/NLRP3/IL-1β is the key pathway signaling exacerbated inflammatory responses that cause viral FH in mice, suggesting that mediation of this signal cascade may benefit on the disease treatment.
| Viral fulminant hepatitis (FH) is a clinical syndrome characterized by massive necrosis of hepatocytes along with hepatic encephalopathy during the infections [1]. Despite advances in the development of antiviral drugs, a poor understanding of the immune mechanisms underlying viral FH has largely stalled the identification of effective clinical interventions. Fortunately, the recent development of an animal model of FH using murine hepatitis virus strain-3 (MHV-3) infection has provided insights in understanding the pathogenesis and developing novel therapeutics for the disease [2].
MHV-3 is a single-stranded, positive-sense RNA virus belonging to the coronavirus family [3]. The hallmarks of MHV-3-induced FH in susceptible BALB/cJ and C57BL/6 mice include the appearance of liver sinusoidal thrombosis and hepatocellular necrosis, resulting from over expression of a virus-induced, monocyte/macrophage-specific procoagulant, fibrinogen-like protein-2 (FGL2). Liver accumulation of FGL2 directly activates the coagulation cascades, a phenomenon known as virus induced procoagulant activity [3]. MHV-3-induced FH exhibits a syndrome that is very similar to the clinical manifestations of patients with viral FH, making it a good animal model for exploring mechanisms underlying the pathogenesis of human viral FH.
In addition to FGL2, pro-inflammatory mediators such as TNF-α, IFN-γ and complement C5a have been proposed to accelerate viral FH pathogenesis [4, 5]. Nevertheless, the mechanisms on how the inflammatory signaling events that regulate the disease progression are not well understood. Recently, it has been shown that dysregulated NLRP3 (also known as NALP3 and cryopyrin) inflammasome in macrophages causes the pathogenesis of inflammatory diseases, which highlights the importance of inflammasome in regulating immune-mediated tissue damages [6]. The generation of biologically active IL-1β requires cleavage of the inactive precursor proIL-1β by the NLRP3 inflammasome, a protein-scaffolding complex consisting of NLRP3, Caspase-1, and the adaptor molecule ASC (apoptosis-associated peck-like protein with CARD domain, Pycard) [6, 7]. NLRP3 inflammasome and IL-1β mediate the host protection against pathogen invasions, whereas, the hyperactivation of NLRP3 inflammasome contributes to the pathogenesis of certain inflammatory syndromes, including liver injuries such as nonalcoholic/alcoholic steatohepatitis [8, 9], liver fibrosis [10], and immune mediated liver injuries [11]. However, the role of NLRP3 inflammasome signaling pathway participates in the pathogenesis of viral FH is still unclear.
A variety of danger-associated molecular patterns (DAMPs) and pathogen-associated molecular patterns (PAMPs), including virus RNA, nigericin, ATP, silica crystals, mitochondrial DNA, and aluminum hydroxide, appear to be capable of activating the NLRP3 inflammasome [12]. Nevertheless, the reactive oxygen species (ROS) generated by nicotinamide adenine dinucleotide phosphate (NADPH) oxidase are considered to be one of the major factors that activate NLRP3 inflammasome [13]. It has been shown that pharmacological inhibition of the NADPH oxidase complex (NOX) or the down regulation of the NOX subunit p22phox eliminates NLRP3 inflammasome activation by preventing ROS secretion [13, 14]. However, recent studies have also illustrated that mitochondria-originated ROS (MitoSOX) rather than NOX-derived ROS drive NLRP3 inflammasome activation [15, 16]. Various stress condition, including increased metabolic rates, hypoxia, or membrane damage, all significantly induce MitoSOX secretion [17]. Conversely, it remains uncertain for which of the NOX-derived ROS or MitoSOX is responsible for causing NLRP3 inflammasome- dependent pathology in viral FH development.
Here, we showed that C57BL/6 wild type (WT) mice infected with MHV-3 manifest with high levels of IL-1β in the serum and liver. Conversely, the virus infected IL-1R1-/- mice present with much attenuated pathologies, showing with a significant reduction in macrophage-derived FGL2 expression and less liver infiltration of CD45+Gr-1high neutrophils. Furthermore, we showed that the in vivo bioactivation of proIL-1β during MHV-3 infection is mediated by NLRP3 inflammasome activation, thereafter, both the NLRP3-/- mice and the Caspase-1-/- mice display substantial resistance to MHV-3-induced IL-1β production. Mechanistically, MHV-3 infection triggers an acute release of NOX-derived ROS. Blocking ROS with Diphenyleneiodonium chloride (DPI) inhibits Caspase-1 activation and IL-1β maturation in vitro. Furthermore, NOX subunit p47phox- deficient mice also exhibited a delayed and moderate viral pathogenesis due to reduction in NLRP3 inflammasome activation in vivo. These results reveal that the ROS/NLRP3/IL-1β axis is a critical signaling pathway leading to the pathogenesis of viral FH.
To examine the status of IL-1β activation in macrophages in response to MHV-3 infection, primary peritoneal exudative macrophages (PEMs) and the macrophage line-RAW264.7 cells were infected with the virus in vitro. A time course data showed a significant induction of the activated form of IL-1β (IL-1β p17) within 12 hours, sustaining to 48h (Fig 1A). Assessment of the PEMs isolated from the 24h of virus infected C57BL/6 WT mice also revealed a significant increase in proIL-1β mRNA expression (Fig 1B). Moreover, proIL-1β mRNA expression in the infected livers appeared to be markedly augmented at 48h (p = 0.0231), sustaining to 72h (p = 0.0356, Fig 1B). In accordance, western-blotting showed with increases in proIL-1β and IL-1β p17 protein expression at corresponding time points in the infected livers (Fig 1C). Flow cytometry further validated the patterns of proIL-1β protein induction in the PEMs isolated from the virus-infected mice (Fig 1D). In agreement, the infected mice also showed significant accumulation of serum IL-1β during the infection (Fig 1E). In contrast, serum IL-1α concentration exhibited little change in MHV-3 infected mice (Fig 1F). These results suggest that IL-1β significantly elevate in the liver and periphery during viral FH.
IL-1β amplifies the pro-inflammatory response via the type-I of IL-1 receptor (IL-1R1) [18]. To further investigate whether IL-1β signaling affects the pathogenesis of viral FH, we infected IL-1R1-/- mice with MHV-3 (100 PFU) via intraperitoneal (i.p.) injection. Interestingly, IL-1R1-/- mice displayed with a significant increase in survival rate with 60% staying alive for 20 days, as compared to a 100% death of the WT littermates within 5 days of the viral infection (Fig 2A). IL-1R1-/- mice manifested a significant reduction in hepatocellular damage and a decrease in serum ALT/AST levels during the infection (Fig 2B). The expression of biliary glycoprotein-1 (Bgp1), the receptor for MHV-3 [19], appeared to be significantly lower in the virus infected IL-1R1-/- livers comparing to that in the WT controls (Fig 2C), concurring with the plaque assay data showing with limited virus entrance and amplification in the livers 72h post-infection (Fig 2D). In support, the MHV-3 infection efficiency in IL-1R1-/- PEMs dropped more significantly than in the WT counterparts in vivo (Fig 2E). Obviously, recombinant mouse IL-1β protein (20 ng/ml) is able to significantly induce Bgp1 expression in PEMs and RAW264.7 cells in vitro (Fig 2F), and in concurrence, IL-1β treated RAW264.7 cells appear to produce more virus than the PBS treated controls post-infection (Fig 2F). In validation, we injected the virus-infected WT mice with IL-1R antagonist (IL-1Ra, 10 mg/kg/day), a naturally occurring cytokine that blocks IL-1β biologic response [18], and observed a significant limitation of IL-1β secretion (p = 0.0007, S1A Fig), inhibition of Bgp1 expression (S1B Fig) and reduction of virus titers (S1C Fig), suggesting the existence of an IL-1R-dependent positive regulation on the virus receptor that directly associate with virus propagation in the host. These combined data clearly demonstrate that IL-1β promotes viral amplification and exacerbates the progression of hepatitis.
FGL2 plays an essential role in inducing hepatocellular necrosis following MHV-3 infection [3]. We firstly examined FGL2 expression in PEMs isolated from MHV-3 infected IL-1R1-/- mice and observed substantial lower levels of FGL2 as compared to the WT controls (Fig 3A). The limited FGL2 expression in macrophages correlates with the low concentrations of FGL2 observed in the virus infected IL-1R1-/- liver and serum (Fig 3B and 3C). Therefore, in response to MHV-3 viral infection, IL-1R1-/- mice responded with limited fibrinogen formation, leading to a down modulation of liver coagulation and necrosis (Fig 3D). Similarly, IL-1Ra-treated WT mice displayed with reduction of FGL2 and fibrinogen deposition in liver tissues, which was followed with decrease in liver damages and enhance the survival time (S1B, S1D and S1E Fig).
Neutrophils and CD4+Foxp3+ regulatory T cells (Tregs) have been well recognized as important players in viral FH [20, 21]. To determine the role of IL-1β in regulating these cells during viral FH, we firstly examined liver neutrophil infiltration status. Flow cytometry showed that in the liver-tissue samples from 48 and 72h post MHV-3 infection, the infiltration of CD45+Gr-1high neutrophils was substantially higher in the WT livers than that in the IL-1R1-/- littermates (Fig 3E and 3F). The number of CD4+Foxp3+Treg in the virus-infected livers appeared to increase significantly after MHV- 3 infection, nevertheless, little difference was observed between IL-1R1-/- mice and their WT controls (S2 Fig). Similarly, serum concentration of C5a, a cytokine that deteriorates the pathogenesis of MHV-3-mediated FH [5], was not changed dramatically between virus infected IL-1R1-/- mice and their WT controls (S3A Fig). These results suggest that attenuation of viral FH by IL-1R1 deficiency could be the consequence of both ineffective FGL2 production by macrophages and limited CD45+Gr-1high neutrophil infiltration in the affected liver.
A reduction of FGL2 expression was observed in IL-1R1-/- mice in response to MHV-3 infection, together with IL-1β and FGL2 were co-expression in PEMs (Fig 4A), implying that IL-1β/IL-1R1 interactions may directly regulate FGL2 expression in macrophages. To address the issue, we treated RAW264.7 cells, a macrophage line capable of expressing FGL2, with the recombinant mouse IL-1β protein (20 ng/ml) in vitro. qPCR and western-blotting data showed that IL-1β alone is incapable of stimulating FLG2 expression, nevertheless, it synergistically enhances TNF-α-induced FGL2 levels (Fig 4B and 4C).
The expression of FGL2 has been proposed to be mediated through the activation of NF-κB and mitogen-activated protein kinase (MAPK) signaling pathways under inflammatory conditions [5, 22]. To further investigate the molecular mechanisms through which IL-1β promotes FGL2 production, we examined these signaling pathways in IL-1β-treated RAW264.7 cells. Results showed that either IL-1β or TNF-α treatment alone, had a minimum stimulation on phosphorylation of the NF-κB chaperone IκBα (p-IκBα) and the NF-κB subunit p65 (p-p65), appearing only at extended incubation time point (12h). However, synergistic effects of IL-1β and TNF-α (IL-1β+TNF-α) seemed to be significant for which substantial increases in phosphorylation of IκBα and p65 can be detected as early as 3h post infection (Fig 4C). Furthermore, the inhibition of NF-κB activation by Pyrrolidinedithiocarbamic Acid (PDTC) successfully prevented FGL2 upregulation after IL-1β+TNF-α treatment (Fig 4D). The combination of IL-1β and TNF-α is capable of potently stimulation the phosphorylation of MAPKs, including extracellular signal-related kinase (p-ERK1/2) and p38 (pp38) (Fig 4C). Nevertheless, the ERK inhibitor-PD98059 and the p38-MAPK inhibitor-SB203580 seemed to be incapable of blocking FGL2 upregulation. Moreover, blocking all of these three pathways did not show additive effect on inhibition of FGL2 expression (Fig 4D). These results suggest that NF-κB rather than the MAPK pathways is responsible for IL-1β+TNF-α-mediated FGL2 upregulation in viral infected macrophages.
It has been established that the Caspase-1-mediated bio-activation of proIL-1β is under the control of NLRP3 inflammasome [6]. MHV-3 infected PEMs and RAW264.7 cells exhibited with a significantly enhanced NLRP3, ASC, pro-Caspase-1 and its activated form (Caspase-1 p20) within 12h of MHV-3 infection (Fig 5A). In accordance, qPCR analyses illustrated that the mRNAs for Nlrp3 and proCaspase-1 were significantly higher in the virus infected livers, this correlates with observation that these virus infected livers also manifest with higher expression of the respective protein (Fig 5B). Next, we infected Nlrp3-/- mice and Caspase-1-/- mice with MHV-3 to address the importance of NLRP3 inflammasome in the causing the virus-induced liver injuries. Remarkably, a 72h viral infection largely failed to induce IL-1β expression in the livers, which was associated with significant reductions in liver FGL2 accumulation (Fig 5C), fibrinogen deposition and local tissue damages, along with significant decreases in serum ALT/AST enzymes as compared with the infected WT mice (Fig 5D). In agreement with these results, we also observed that Bgp1 expression was significantly lower in NLRP3-/- and Caspase-1-/- livers during infection (Fig 5C). Meanwhile, NLRP3-/- mice and Caspase-1-/- mice appeared to produce much less viruses at 72h of infection as compared to the WT controls (Fig 5E). Finally, NLRP3-/- and Caspase-1-/- mice presented with considerably prolonged survival rates toward MHV-3 infection in comparing to the WT controls (Fig 5F). The serum C5a in the viral infected NLRP3-/- and Caspase-1-/- animals was also significantly increased but no different from the WT control mice (S3B Fig), indicating that C5a up-regulation during the viral infection, appears to either additively or synergistically work with other inflammatory factors to cause viral FH. Together these observations further validate that the NLRP3/Caspase-1-inflammasome regulates the bio-processing of proIL-1β for causing the MHV-3 mediated viral FH.
Assembly and activation NLRP3 inflammasome, being critical for bio-processing and activation of IL-1β, has been suggested to also involve in the bio-activation of IL-18, another member of the IL-1 superfamily [23]. The MHV-3-infected mice showed a significant up-regulation of proIL-18 mRNA in PEMs and livers (Fig 6A), as well as enhanced IL-18 protein in serum (Fig 6B). However, the recombinant mouse IL-18 protein (50 ng/ml) alone, or in the combination with TNF-α and INF-γ, was unable to stimulate fgl2 mRNA transcription in RAW264.7 cells or SVE-10 endothelial cells in vitro (Fig 6C). Moreover, MHV-3 induced liver FGL2 production remained high in IL-18-/- mice (Fig 6D), showing with consequentially high levels of fibrinogen deposition, liver damages and hepatocyte necrosis (Fig 6E). Additionally, liver tissues isolated from IL-18-/- mice appear to up-regulate Bgp1 expression after MHV-3 infection. In accordance, these mice also manifested with high virus duplication (Fig 6F). Overall, IL-18-/- mice are still sensitive to MHV-3 infection (Fig 6G), suggesting that IL-18 is not essential in MHV-3-mediated fulminant hepatitis.
Many factors contribute to activating the NLRP3 inflammasome and among which, ROS is lately gaining particular attentions [13]. In order to examine the role of ROS in NLRP3 inflammasome hyperactivation, we first detected the release of NADPH oxidase-derived ROS by using a permeable dichlorohydrofluorescein (DCFH) upon MHV-3 infection. Flow cytometry showed that the releasing of DCFH from MHV-3 infected PEMs and RAW264.7 cells significantly increased, especially at 12h and 24h post-infection (Fig 7A). This result correlates with the up-regulation of gp91phox, p47phox and NOX4, the subunits that are essential for acute ROS secretion in RAW264.7 cells (Fig 7B). However, the DCFH level dropped dramatically at 48h and 72h post the viral infection (Fig 7A), most likely owing to death of cells under these conditions (S4 Fig).
In addition to NADPH oxidase-derived ROS, mitochondria may provide an alternate source of ROS [15]. We therefore assessed the functional mitochondrial pool in MHV-3 infected cells. The viral infection in PEMs and RAW264.7 cells caused an increase in mitochondrial damage, especially at 48h and 72h post-infection, as detected by MitoTracker Green FM, a dye that stains mitochondria with no influence on their membrane potentials (Fig 7A). Similarly, electron microscopy showed with swollen mitochondria in the MHV-3 infected Raw264.7 cells at 48h and 72h (Fig 7C). This sign of mitochondrial damage seemed to strongly correlate with the increase in MitoSOX release within the same time frame (Fig 7A).
To further elucidate the role of ROS in NLRP3 inflammasome hyperactivation, we treated MHV-3 infected RAW264.7 cells with a ROS inhibitor Diphenyliodonium chloride (DPI), which is capable of preventing both NOX-dependent ROS and MitoXOS secretion [16]. NOX-originated DCFH was successfully inhibited by DPI in a dose dependent manner (Fig 7D). However, MitoXOS release was not prevented by the DPI treatment, even at a very high dose (50μM) (Fig 7D). The efficiency of NOX-originated ROS inhibition by DPI appeared to correlate with the reduction in IL-1β activation in the infected RAW264.7 cells and PEMs in dose dependent manners (Fig 7E). Together, these results suggest that the hyperactivation of NLRP3 inflammasome in macrophage is partially mediated by MHV-3 induced, NOX-derived ROS.
Cells in deficiency of p47phox exhibit a reduced capacity in generating ROS [24]. To further investigate the role of NOX-originated ROS in regulating NLRP3 inflammasome hyperactivation, we infected p47phox-/- mice with MHV-3 and examined the severity of liver pathology. As anticipated, PEMs isolated from MHV-3 infected p47phox-/- mice showed with limited DCFH (Fig 8A). Interestingly, the p47phox-/- mice also displayed considerable resistance to MHV-3 infection, presenting with reduced disease severity within the prolonged survival time as compared with the WT controls (p = 0.0175, Fig 8B). The lack of virus-induced ROS response, which leads to prohibition of NLRP3/caspase-1 activation and thus reduction in IL-1β production, seems to be responsible for this effect (Fig 8C and 8D). As a result, the virus infection is unable to generate significant FGL2 accumulation in the liver and serum (Fig 8C and 8D). Therefore, these mice manifested with less severe fibrinogen deposition, liver injury and hepatocyte necrosis, accompanying with low levels of AST/ALT enzymes released by the liver (Fig 8E). However, the limitation of IL-1β secretion in these p47phox-/- mice only slightly affected liver Bgp1 expression (Fig 8C), and therefore live virus titers were still high at 72h of infection (Fig 8F). Conversely, the administration of IL-1β (100 ng/mouse/day) in MHV-3 infected p47phox-/- mice was able to reinstate all aspects of disease severity typical in viral FH (Figs 8G and S5). Taken together, these results clearly indicate that the ROS/NLRP3/IL-1β axis plays a critical role in the pathogenesis of viral FH.
In the present work, we report that mice infected with MHV-3, an animal model for viral FH, have significantly elevated levels of IL-1β in the serum and liver. The accumulation of IL-1β accelerated liver pathology through synergistically acting with TNF-α, one of the key inflammatory cytokines that has been previously shown to be essential for causing viral FH [4, 18], IL-1R1 signaling is responsible for stimulation of FGL2 expression in macrophages and enhancing infiltration of the inflammatory CD45+Gr-1high neutrophils in the livers. Interestingly, MHV-3 infection in IL-1R1-/- mice, or in WT mice treated with IL-1β signaling inhibitors, such as using IL-1Ra, rescue the otherwise susceptible animals from the viral FH status, presenting with limited virus replication, attenuated disease progression and reduced mortality. We have also shown that the bioprocess of IL-1β maturation is under the control of a key signaling pathway, involving a MHV-3 virus inducible, ROS-dependent NLRP3 inflammasome activation. Animals lacking of NLRP3, Caspase-1 or NADPH oxidase subunit p47phox that controls acute ROS secretion, all exhibited with reduced IL-1β bio-processing that results in prevention of the MHV-3 mediated disease severity. To the best of our knowledge, these data provide evidence for the first time showing that the ROS/NLRP3/IL-1β axis is an essential contributor for the virus-induced FH.
Although macrophage-mediated inflammation has been speculated to be critical for gauging the pathological susceptibility of viral FH caused by MHV-3 infection [25], the mechanisms underlying the pathogenesis are not well understood. IL-1β and IL-18 are two key inflammatory cytokines produced by macrophages which play a pivotal role in antimicrobial immunity [7, 23]. Previous studies have showed that IL-1R1-/- mice appear to have markedly reduced inflammatory pathology in the lung, presumably due to the impaired neutrophil recruitment upon influenza virus infection [26]. Conversely, Ramos et al. reported that IL-1R1-/- mice exhibited with a higher accumulation of the West Nile virus (WNV) in the central nervous system due to a restrained activation of the virus-specific effector CD8+ T cells [27]. Similarly, IL-1β-/- mice are more susceptible to herpes simplex virus 1 (HSV1)- mediated encephalitis due to an increase in viral load [28]. We here further explored the role of IL-1β in MHV-3 mediated FH. Interestingly, IL-1R1-/- animals display a significant reduction in viral duplication, amelioration of liver damage and a prolonged survival rate against MHV-3 infection (Fig 2A and 2B). These effects are probably due to IL-1R1 deficiency lead to limit liver recruitment of CD45+Gr-1high neutrophils and decrease in production of the macrophage-derived FGL2, which mediates sinusoidal fibrin deposition and hepatocellular necrosis in response to MHV-3 infection [3]. Bgp1 (also called carcinoembryonic cell adhesion antigen 1a,CEACAM1a) is the specific receptor for the mouse hepatitis virus (MHV), and down-regulation of Bgp1 by IFN-γ is related to the antiviral state and resistance to mouse hepatitis virus 3 infection [29]. However, Bgp1 does not appear to be involved in IL-6 and TNF-α secretion from MHV-3 infected macrophages [30]. In contrast to IFN-γ treatment, we here showed that the expression of Bgp1 drops significantly in the IL-1R1-/- liver during the viral infection, suggesting Bgp1 expression in macrophages is induced by IL-1β/IL-1R1 signaling, and lacking the pathway may compromise virus entrance and amplification. These unpredicted data implies that IL-1β has double-edge effects on the immune system, in which proper balancing with its signaling extent becomes essential for the host in protection against various invading viruses and meanwhile, in prevention of the potential collateral damage.
The molecular mechanisms that are responsible for triggering the expression of FGL2 prothrombinase, which plays a critical role in the development of MHV-3 mediated FH, are still unclear. McGilvray et al. found that both ERK and p38-MAPK proteins are activated in MHV-3 infected PEMs, and only inhibition of p38-MAPK can abolish FGL2 induction and its functional activity [31]. Jia et al. have illustrated that TNF-α upregulates FGL2 expression via activation of NF-kB and p38-MAPK in cardiac microvascular endothelial cells [22]. Our recent work also have showed that the inhibition of ERK1/2 and p38-MAPK efficiently block C5a-mediated FGL2 upregulation [5]. Ning et al., have demonstrated that the hepatocyte nuclear factor-4 (HNF4) cis-elements and its cognate transcription factor, HNF4α, are necessary for MHV-3-induced fgl2 gene transcription [32]. Based on these studies, we further examined the molecular mechanisms underlying IL-1β-mediated FGL2 expression. The results show that IL-1β and TNF-α synergistically induce NF-κB, ERK and p38-MAPK tyrosine-phosphorylation (Fig 4C). However, the inhibition NF-κB pathway, but not the ERK, or p38-MAPK signals, markedly prevented FGL2 expression (Fig 4D), suggesting that the NF-κB pathways are responsible for IL-1β+TNF-α-mediated FGL2 augmentation.
The NLRP3, RIG-I and the AIM2 are three main types of inflammasome complexes that have been shown to control caspase-1 activity and IL-1β maturation. It seems that AIM2 is responsible for detecting DNA viruses, while both NLRP3 and RIG-I associate with recognition of RNA viruses by cells [33, 34]. Recent evidences suggest that the host protective immunity requires the NLRP3 inflammasome for fighting against various kinds of viruses, including Influenza A virus, modified Vaccinia virus Ankara, Sendai virus, Respiratory Syncytial virus, Encephalomyocarditis viruses, as well as Adenoviruses [35]. Our study shows that the MHV-3 triggered NLRP3, ASC and Caspase-1 mRNA as well as protein expression in PEMs and RAW264.7 cells in vitro (Fig 5A). Nevertheless, loss of either NLRP3 or Caspase-1 in macrophages reduces IL-1β secretion upon MHV-3 challenge (Fig 5C). Additionally, NLRP3-/- and Caspase-1-/- mice essentially pheno-copied the manifestations of IL-1R1-/- mice in response to MHV-3 infections, these mice evidenced with reduction in MHV-3 virus-induced IL-1β production and lessening of disease progression (Fig 5C–5F). These combined data suggest that NLRP3-inflammasome acts as a predominant pathway for triggering IL-1β maturation by MHV-3, and probably also by other corona viruses. Previous study showed that RAW264.7 cells do not release mature IL-1β because they do not express ASC [36]. Conversely, we here show that MHV-3 promotes IL-1β secretion from virus infected RAW264.7 cells through inducing ASC expression. Together with the recent work demonstrated that NLRP3/ASC/caspase-1 axis participates in the regulation of the generation of IL-1β in RAW264.7 cells, indicating that ASC is inducible in the macrophage line RAW264.7 cells under circumstances, especially during MHV-3 infection [37].
ROS plays an essential role in mediating NLRP3 inflammasome activation [13]. Many different viruses, such as Influenza virus, Respiratory Syncytial virus, and Hepatitis C virus, trigger NLRP3 inflammasome activation through ROS-dependent mechanisms [38–40]. NOX is an enzymatic complex consisting mainly of five subunits (p22phox, p40phox, p47phox, p67phox and gp91phox) and two GTP-binding proteins (RAC1/RAC2). We here show that MHV-3 triggers NOX-derived ROS secretion in macrophages by inducing NOX-subunits, including GP91phox, p47phox and NOX-4 expression in the very early stages of the viral infection (Fig 7A and 7C). Additionally, preventing NOX-derived ROS through DPI appeared to successfully down modulate NLRP3 hyperactivation and IL-1β maturation in vitro (Fig 7F). Furthermore, virus infected p47phox-/- macrophages manifested with significant reduction in ROS secretion, leading to the control of NLRP3 hyperactivation, which results in attenuation in severity of the viral FH (Fig 8). These results are inconsistent with previous reports that have shown that NADPH oxidase-derived ROS are not involved in activating NLRP3 inflammasome [41, 42]. One of the discrepancies is the different cell models are used in studies. Silica crystals, LPS, and uric acid crystals act as the stimulators in these studies, while MHV-3 virus is the activator in our research. Conversely, it is worth mentioning that not all p47phox-/- mice are completely resistant to MHV-3, and these animals eventually still died from the infections (Fig 8B), together with some virus infected mice still produce high levels of IL-1β and virus titers, suggesting the presence of other mediators that in response to the virus challenge, are capable of activating NLRP3 inflammasome in vivo. One of the potential activators is MitoSOX [43, 44]. We have also observed a very high level of the MitoSOX production in the MHV-3 infected RAW264.7 cells at 48h and 72h post-infection in vitro, along with high frequency damage and destruction of mitochondria might simultaneously occur. However, the release of MitoSOX was unable to be successfully blocked by ROS inhibitor- DPI (50 uM) (Fig 7). Additionally, DPI is harmful to animals and unsuitable in vivo experiments [45]. The incapable of completely blocking ROS production by using high dose of DPI in vitro suggests the existence of other sources of ROS for activating NLRP3 inflammasome. Interestingly, reduced mortality and pathology were seen in MHV-3 infected p47phox-/- mice compared to WT littermates despite a lack of significant reduction in virus replication, suggesting that MHV-3-mediated pathology is due to inflammation and not direct virus infection. Recent studies by Warner Greene’s group demonstrate that HIV can trigger caspase-1 activation and pyroptosis, a highly inflammatory form of programmed cell death in which dying cells release their cytoplasmic contents, including inflammatory cytokines into the extracellular space where the virus infected CD4+ T-cells recite [46]. A similar environment might also explain for the MHV-3 induced FH status.
IL-18 is another member of the IL-1 superfamily that has been indicated to be important in the pathogenesis of mouse models of Influenza virus, HBV, Rhinovirus and Vaccinia virus infection [47]. For example, IL-18R-/- mice appeared to be protected from Influenza viral initiated inflammatory lung damages [48]. Consistent with previous reports, we have detected significantly high levels of matured IL-18 in the serum of MHV-3 infected WT mice. However, IL-18 deficiency does not prevent Bgp1 expression, virus amplification and FGL2 accumulation in the liver following MHV-3 infection, and as the consequence, these mice stay high with fibrinogen deposition, liver damage and hepatocyte necrosis (Fig 6). These results suggest that IL-18 is not essential for causing MHV-3 mediated acute hepatitis.
In conclusion, our study elucidates that NLRP3 inflammasome-dependent IL-1β production, a primary inflammatory signaling pathway of the host for mounting conventional immunity against pathogen invasions, plays a double-edged role in the host immune system. Hepatotropic virus, like MHV-3 infection in mice, can induce exaggerated inflammation in the liver and cause life-threatening viral FH. These results shed lights on a novel strategy, for which the properly modulation of the IL-1β signaling pathway, in combination with blocking other inflammatory factors, might benefit the treatment of viral FH and other severe inflammatory diseases in human.
The p47phox-deficient (p47phox-/-, #004742), NLRP3-/- (#017970), Caspase-1-/- (#016621), IL-18-/- (#004130), IL-1R1-/- (#003245) and wild type (WT) mice were on C57BL/6 background and were purchased from the Jackson Laboratory (Bar Harbor, Maine, USA). Mice were maintained in micro-isolator cages, fed with standard laboratory chow diet and water, and housed in the animal colony at the animal center of the Third Military Medical University (TMMU). Mice approximately 12 weeks of age were used for these experiments. All animals received humane care according to the criteria outlined in the "Guide for the Care and Use of Laboratory Animals" prepared by the National Academy of Sciences and published by the National Institutes of Health (NIH publication 86–23 revised 1985).
RAW264.7 cells were provided by the Cell Institute of the Chinese Academy of Sciences (Shanghai, China). Peritoneal exudative macrophages (PEMs) were harvested as described previously [5]. Cells were cultured in 6-well plates and propagated in DMEM supplemented with 10% FBS, 100 U/ml penicillin, and 100 μg/ml streptomycin.
MHV-3 viruses were expanded in murine 17CL1 cells to a concentration of 1×107 plaque forming unit (PFU)/ml. The virus-containing supernatants were stored at -80°C until use. Macrophages were infected with MHV-3 (multiplicity of infection, MOI = 1) in vitro and mice were injected with 100 PFU of MHV-3 via i.p. In some experiments, the virus infected mice were treated with IL-1R antagonist (IL-1Ra, 10 mg/kg/day) or recombinant mouse IL-1β protein (100 ng/day/mouse) every day. Mice were euthanized on the indicated days and the virus titers in liver were determined by plaque assay as described previously [25]. The sources of antibodies and other reagents are detailed in S1 Text.
Paraffin-embedded liver tissue blocks were cut into 3 μm slices and mounted onto poly-lysine-charged glass slides, and tissue injury was stained by hematoxylin and eosin (H&E). Cellular apoptosis was measured by TUNEL staining according to the manufacturer's instructions (Roche, Berlin, Germany). The expression of fibrinogen and FGL2 was detected by immunohistochemistry as described previously [25]. Sections were scored in a blinded fashion for histological diagnosis.
Total RNA was extracted from cultured cells or liver tissues with TRIzol reagent according to the manufacturer's instructions (Invitrogen, NY, USA). First-strand cDNA was synthesized with the PrimeScript RT-PCR Kit (Takara, Dalian, China). The expression of mRNA encoding for NLRP3, Caspase-1, proIL-1β and proIL-18 was quantified by real-time quantitative PCR with the SYBR Premix ExTaq kit (Takara) and was normalized to the expression of β-actin. Sequences of the primers are provided in S1 Table. Results were calculated and compared by the 2−ΔΔCt method.
Serum C5a, FGL2, IL-18 and IL-1β levels were measured by ELISA. The expression of FGL2, proCaspase-1, Caspase-1-p20, NLRP-3, p47phox, p90phox, p67phox, Nox-4, Bgp1, proIL-1β and IL-1β-p17 in MHV-3 infected livers or macrophages was detected by western-blotting described previously [25].
The release of IL-1β/ROS from virus infected macrophages, liver infiltration of CD45+GR-1high neutrophil and CD4+Foxp3+ regulatory T cells (Treg), all were detected by flow cytometry (FACsAria cytometer, BD, Franklin Lakes, NJ, USA). The death cells were excluded firstly by staining with LIVE/DEATH Fixable Near-IR Ded Cell Stain Kit (Life technologies, Eugene, Oregon, USA). The secretion of NOX-derived ROS was detected by means of an oxidation-sensitive fluorescent probe-DCFH according to the manufacturer's instructions (Beyotime, Shanghai, China). Moreover, the mitochondria-derived ROS was measured in cells stained with MitoSOX (5 μM, Invitrogen) for 20 min. To measure mitochondrial damage, cells were stained for 20 min with MitoTracker Green FM (20 nM) and MitoTracker Deep Red FM (20 nM), two kinds of dye that stain mitochondria with no influence on their membrane potentials (Invitrogen). A total of 10,000 live cells were analyzed. All the FACs data were analyzed using CellQuest Pro software.
RAW264.7 cells or primary PEMs isolated from MHV-3 infected mice were fixed with 4% (v/v) glutaraldehyde. Sample preparation was conducted as described previously [49]. Mitochondrial morphology and virion was observed with JEOL JEM2100HC transmission electron microscopy.
All data were analyzed using GraphPad Prism 4.03 software. An unpaired Student’s t-test (two-tailed) was used to assess comparisons between two groups when the data met the assumptions of the t-test. Survival curves were generated by log-rank test. p<0.05 was considered a significant difference.
All animal experiments were performed in strict accordance with the Guide for the Care and Use of Laboratory Animals issued by the Ministry of Science and Technology of the People's Republic of China. The protocol was approved by the Third Military Medical University Institutional Animal Care and Use Committee.
|
10.1371/journal.pgen.1006528 | Genome-wide physical activity interactions in adiposity ― A meta-analysis of 200,452 adults | Physical activity (PA) may modify the genetic effects that give rise to increased risk of obesity. To identify adiposity loci whose effects are modified by PA, we performed genome-wide interaction meta-analyses of BMI and BMI-adjusted waist circumference and waist-hip ratio from up to 200,452 adults of European (n = 180,423) or other ancestry (n = 20,029). We standardized PA by categorizing it into a dichotomous variable where, on average, 23% of participants were categorized as inactive and 77% as physically active. While we replicate the interaction with PA for the strongest known obesity-risk locus in the FTO gene, of which the effect is attenuated by ~30% in physically active individuals compared to inactive individuals, we do not identify additional loci that are sensitive to PA. In additional genome-wide meta-analyses adjusting for PA and interaction with PA, we identify 11 novel adiposity loci, suggesting that accounting for PA or other environmental factors that contribute to variation in adiposity may facilitate gene discovery.
| Decline in daily physical activity is thought to be a key contributor to the global obesity epidemic. However, the impact of sedentariness on adiposity may be in part determined by a person’s genetic constitution. The specific genetic variants that are sensitive to physical activity and regulate adiposity remain largely unknown. Here, we aimed to identify genetic variants whose effects on adiposity are modified by physical activity by examining ~2.5 million genetic variants in up to 200,452 individuals. We also tested whether adjusting for physical activity as a covariate could lead to the identification of novel adiposity variants. We find robust evidence of interaction with physical activity for the strongest known obesity risk-locus in the FTO gene, of which the body mass index-increasing effect is attenuated by ~30% in physically active individuals compared to inactive individuals. Our analyses indicate that other similar gene-physical activity interactions may exist, but better measurement of physical activity, larger sample sizes, and/or improved analytical methods will be required to identify them. Adjusting for physical activity, we identify 11 novel adiposity variants, suggesting that accounting for physical activity or other environmental factors that contribute to variation in adiposity may facilitate gene discovery.
| In recent decades, we have witnessed a global obesity epidemic that may be driven by changes in lifestyle such as easier access to energy-dense foods and decreased physical activity (PA) [1]. However, not everyone becomes obese in obesogenic environments. Twin studies suggest that changes in body weight in response to lifestyle interventions are in part determined by a person’s genetic constitution [2–4]. Nevertheless, the genes that are sensitive to environmental influences remain largely unknown.
Previous studies suggest that genetic susceptibility to obesity, assessed by a genetic risk score for BMI, may be attenuated by PA [5, 6]. A large-scale meta-analysis of the FTO obesity locus in 218,166 adults showed that being physically active attenuates the BMI-increasing effect of this locus by ~30% [7]. While these findings suggest that FTO, and potentially other previously established BMI loci, may interact with PA, it has been hypothesized that loci showing the strongest main effect associations in genome-wide association studies (GWAS) may be the least sensitive to environmental and lifestyle influences, and may therefore not make the best candidates for interactions [8]. Yet no genome-wide search for novel loci exhibiting SNP×PA interaction has been performed. A genome-wide meta-analysis of genotype-dependent phenotypic variance of BMI, a marker of sensitivity to environmental exposures, in ~170,000 participants identified FTO, but did not show robust evidence of environmental sensitivity for other loci [9]. Recent genome-wide meta-analyses of adiposity traits in >320,000 individuals uncovered loci interacting with age and sex, but also suggested that very large sample sizes are required for interaction studies to be successful [10].
Here, we report results from a large-scale genome-wide meta-analysis of SNP×PA interactions in adiposity in up to 200,452 adults. As part of these interaction analyses, we also examine whether adjusting for PA or jointly testing for SNP’s main effect and interaction with PA may identify novel adiposity loci.
We performed meta-analyses of results from 60 studies, including up to 180,423 adults of European descent and 20,029 adults of other ancestries to assess interactions between ~2.5 million genotyped or HapMap-imputed SNPs and PA on BMI and BMI-adjusted waist circumference (WCadjBMI) and waist-hip ratio (WHRadjBMI) (S1–S5 Tables). Similar to a previous meta-analysis of the interaction between FTO and PA [7], we standardized PA by categorizing it into a dichotomous variable where on average ~23% of participants were categorized as inactive and ~77% as physically active (see Methods and S6 Table). On average, inactive individuals had 0.99 kg/m2 higher BMI, 3.46 cm higher WC, and 0.018 higher WHR than active individuals (S4 and S5 Tables).
Each study first performed genome-wide association analyses for each SNP’s effect on BMI in the inactive and active groups separately. Corresponding summary statistics from each cohort were subsequently meta-analyzed, and the SNP×PA interaction effect was estimated by calculating the difference in the SNP’s effect between the inactive and active groups. To identify sex-specific SNP×PA interactions, we performed the meta-analyses separately in men and women, as well as in the combined sample. In addition, we carried out meta-analyses in European-ancestry studies only and in European and other-ancestry studies combined.
We used two approaches to identify loci whose effects are modified by PA. In the first approach, we searched for genome-wide significant SNP×PA interaction effects (PINT<5x10-8). As shown in Fig 1, this approach yielded the highest power to identify cross-over interaction effects where the SNP’s effect is directionally opposite between the inactive and active groups. However, this approach has low power to identify interaction effects where the SNP’s effect is directionally concordant between the inactive and active groups (Fig 1). We identified a genome-wide significant interaction between rs986732 in cadherin 12 (CDH12) and PA on BMI in European-ancestry studies (betaINT = -0.076 SD/allele, PINT = 3.1x10-8, n = 134,767) (S7 Table). The interaction effect was directionally consistent but did not replicate in an independent sample of 31,097 individuals (betaINT = -0.019 SD/allele, PINT = 0.52), and the pooled association P value for the discovery and replication stages combined did not reach genome-wide significance (NTOTAL = 165,864; PINT-TOTAL = 3x10-7) (S1 Fig). No loci showed genome-wide significant interactions with PA on WCadjBMI or WHRadjBMI. CDH12 encodes an integral membrane protein mediating calcium-dependent cell-cell adhesion in the brain, where it may play a role in neurogenesis [11]. While CDH12 rs4701252 and rs268972 SNPs have shown suggestive associations with waist circumference (P = 2x10-6) and BMI (P = 5x10-5) in previous GWAS [12, 13], the SNPs are not in LD with rs986732 (r2<0.1).
In our second approach, we tested interaction for loci showing a genome-wide significant main effect on BMI, WCadjBMI or WHRadjBMI (S7–S12 Tables). We adjusted the significance threshold for SNP×PA interaction by Bonferroni correction (P = 0.05/number of SNPs tested). As shown in Fig 1, this approach enhanced our power to identify interaction effects where there is a difference in the magnitude of the SNP’s effect between inactive and active groups when the SNP’s effect is directionally concordant between the groups. We identified a significant SNP×PA interaction of the FTO rs9941349 SNP on BMI in the meta-analysis of European-ancestry individuals; the BMI-increasing effect was 33% smaller in active individuals (betaACTIVE = 0.072 SD/allele) than in inactive individuals (betaINACTIVE = 0.106 SD/allele, PINT = 4x10-5). The rs9941349 SNP is in strong LD (r2 = 0.87) with FTO rs9939609 for which interaction with PA has been previously established in a meta-analysis of 218,166 adults [7]. We identified no loci interacting with PA for WCadjBMI or WHRadjBMI.
In a previously published meta-analysis [7], the FTO locus showed a geographic difference for the interaction effect where the interaction was more pronounced in studies from North America than in those from Europe. To test for geographic differences in the present study, we performed additional meta-analyses for the FTO rs9941349 SNP, stratified by geographic origin (North America vs. Europe). While the interaction effect was more pronounced in studies from North America (betaINT = 0.052 SD/allele, P = 5x10-4, N = 63,896) than in those from Europe (betaINT = 0.028 SD/allele, P = 0.006, N = 109,806), we did not find a statistically significant difference between the regions (P = 0.14).
Physical activity contributes to variation in BMI, WCadjBMI, and WHRadjBMI, hence, adjusting for PA as a covariate may enhance power to identify novel adiposity loci. To that extent, each study performed genome-wide analyses for association with BMI, WCadjBMI, and WHRadjBMI while adjusting for PA. Subsequently, we performed meta-analyses of the study-specific results. We discovered 10 genome-wide significant loci (2 for BMI, 1 for WCadjBMI, 7 for WHRadjBMI) that have not been reported in previous GWAS of adiposity traits (Table 1, S2–S4 Figs).
To establish whether additionally accounting for SNP×PA interactions would identify novel loci, we calculated the joint significance of PA-adjusted SNP main effect and SNP×PA interaction using the method of Aschard et al [16]. As illustrated in Fig 1, the joint test enhanced our power to identify loci where the SNP shows simultaneously a main effect and an interaction effect. We identified a novel BMI locus near ELAVL2 in men (PJOINT = 4x10-8), which also showed suggestive evidence of interaction with PA (PINT = 9x10-4); the effect of the BMI-increasing allele was attenuated by 71% in active as compared to inactive individuals (betaINACTIVE = 0.087 SD/allele, betaACTIVE = 0.025 SD/allele) (Table 1, S2–S4 Figs).
To evaluate the effect of PA adjustment on the results for the 11 novel loci, we performed a look-up in published GIANT consortium meta-analyses for BMI, WCadjBMI, and WHRadjBMI that did not adjust for PA [17, 18] (S22 Table). All 11 loci showed a consistent direction of effect between the present PA-adjusted and the previously published PA-unadjusted results, but the PA-unadjusted associations were less pronounced despite up to 40% greater sample size, suggesting that adjustment for PA may have increased our power to identify these loci.
The biological relevance of putative candidate genes in the novel loci, based on our thorough searches of the literature, GWAS catalog look-ups, and analyses of eQTL enrichment and overlap with functional regulatory elements, are described in Tables 2 and 3. As the novel loci were identified in a PA-adjusted model, where adjusting for PA may have contributed to their identification, we examined whether the lead SNPs in these loci are associated with the level of PA. More specifically, we performed look-ups in GWAS analyses for the levels of moderate-to-vigorous intensity leisure-time PA (n = 80,035), TV-viewing time (n = 28,752), and sedentary behavior at work (n = 59,381) or during transportation (n = 15,152) [personal communication with Marcel den Hoed, Marilyn Cornelis, and Ruth Loos]. However, we did not find significant associations when correcting for the number of loci that were examined (P>0.005) (S16 Table).
In addition to uncovering 11 novel adiposity loci, our PA-adjusted GWAS and the joint test of SNP main effect and SNP×PA interaction confirmed 148 genome-wide significant loci (50 for BMI, 58 for WCadjBMI, 40 for WHRadjBMI) that have been established in previous main effect GWAS for adiposity traits (S7–S12 Tables, S4 Fig). The lead SNPs in eight of the previously established loci (5 for BMI, 3 for WCadjBMI), however, showed no LD or only weak LD (r2<0.3) with the published lead SNP, suggesting they could represent novel secondary signals in known loci (S17 Table). To test whether these eight signals are independent of the previously published signals, we performed conditional analyses [19]. Three of the eight SNPs we examined, in/near NDUFS4, MEF2C-AS1 and CPA1, were associated with WCadjBMI with P<5x10-8 in our PA-adjusted GWAS even after conditioning on the published lead SNP, hence representing novel secondary signals in these loci (S17 Table).
Epigenetic variation may underlie gene-environment interactions observed in epidemiological studies [20] and PA has been shown to induce marked epigenetic changes in the genome [21]. We examined whether the BMI or WHRadjBMI loci reaching P<1x10-5 for interaction with PA (13 loci for BMI, 5 for WHRadjBMI) show overall enrichment with chromatin states in adipose, brain and muscle tissues available from the Roadmap Epigenomics Consortium [22]. However, we did not find significant enrichment (S18 and S19 Tables), which may be due to the limited number of identified loci. The lack of significant findings may also be due to the assessment of chromatin states in the basal state, which may not reflect the dynamic changes that occur when cells are perturbed by PA [23].
We also tested whether the loci reaching P<5x10-8 in our PA-adjusted GWAS of BMI or WHRadjBMI show enrichment with chromatin states and found significant enrichment of the BMI loci with enhancer, weak transcription, and polycomb-repressive elements in several brain cell lines, and with enhancer elements in three muscle cell lines (S20 and S21 Tables). We also found significant enrichment of the WHRadjBMI loci with enhancer elements in three adipose and six muscle cell lines, with active transcription start sites in two adipose cell lines, and with polycomb-repressive elements in seven brain cell lines. The enrichment of our PA-adjusted main effect results with chromatin annotations in skeletal muscle in particular, the tissue most affected by PA, could highlight regulatory mechanisms that may be influenced by PA.
In this genome-wide meta-analysis of more than 200,000 adults, we do not find evidence of interaction with PA for loci other than the established FTO locus. However, when adjusting for PA or jointly testing for SNP main effect and interaction with PA, we identify 11 novel adiposity loci, suggesting that accounting for PA or other environmental factors that contribute to variation in adiposity may increase power for gene discovery.
Our results suggest that if SNP×PA interaction effects for common variants exist, they are unlikely to be of greater magnitude than observed for FTO, the BMI-increasing effect of which is attenuated by ~30% in physically active individuals. The fact that common SNPs explain less of the BMI variance among physically active compared to inactive individuals indicates that further interactions may exist, but larger meta-analyses, more accurate and precise measurement of PA, and/or improved analytical methods will be required to identify them. We found no difference between inactive and active individuals in variance explained by common SNPs in aggregate for WCadjBMI or WHRadjBMI, and no loci interacted with PA on WCadjBMI or WHRadjBMI. Therefore, PA may not modify genetic influences as strongly for body fat distribution as for overall adiposity. Furthermore, while differences in variance explained by common variants may be due to genetic effects being modified by PA, it is important to note that heritability can change in the absence of changes in genetic effects, if environmental variation differs between the inactive and active groups. Therefore, the lower BMI variance explained in the active group could be partly due to a potentially greater environmental variation in this group.
While we replicated the previously observed interaction between FTO and PA [7], it remains unclear what biological mechanisms underlie the attenuation in FTO’s effect in physically active individuals, and whether the interaction is due to PA or due to confounding by other environmental exposures. While some studies suggest that FTO may interact with diet [24–26], a recent meta-analysis of 177,330 individuals did not find interaction between FTO and dietary intakes of total energy, protein, carbohydrate or fat [27]. The obesity-associated FTO variants are located in a super-enhancer region [28] and have been associated with DNA methylation levels [29–31], suggesting that this region may be sensitive to epigenetic effects that could mediate the interaction between FTO and PA.
In genome-wide analyses for SNP main effects adjusting for PA, or when testing for the joint significance of SNP main effect and SNPxPA interaction, we identify 11 novel adiposity loci, even though our sample size was up to 40% smaller than in the largest published main effect meta-analyses [17, 18]. Our findings suggest that accounting for PA may facilitate the discovery of novel adiposity loci. Similarly, accounting for other environmental factors that contribute to variation in adiposity could lead to the discovery of additional loci.
In the present meta-analyses, statistical power to identify SNPxPA interactions may have been limited due to challenges relating to the measurement and statistical modeling of PA [5]. Of the 60 participating studies, 56 assessed PA by self-report while 4 used wearable PA monitors. Measurement error and bias inherent in self-report estimates of PA [32] can attenuate effect sizes for SNP×PA interaction effects towards the null [33]. Measurement using PA monitors provides more consistent results, but the monitors are not able to cover all types of activities and the measurement covers a limited time span compared to questionnaires [34]. As sample size requirements increase nonlinearly when effect sizes decrease, any factor that leads to a deflation in the observed interaction effect estimates may make their detection very difficult, even when very large population samples are available for analysis. Finally, because of the wide differences in PA assessment tools used among the participating studies, we treated PA as a dichotomous variable, harmonizing PA into inactive and active individuals. Considerable loss of power is anticipated when a continuous PA variable is dichotomized [35]. Our power could be enhanced by using a continuous PA variable if a few larger studies with equivalent, quantitative PA measurements were available.
In summary, while our results suggest that adjusting for PA or other environmental factors that contribute to variation in adiposity may increase power for gene discovery, we do not find evidence of SNP×PA interaction effects stronger than that observed for FTO. While other SNP×PA interaction effects on adiposity are likely to exist, combining many small studies with varying characteristics and PA assessment tools may be inefficient for identifying such effects [5]. Access to large cohorts with quantitative, equivalent PA variables, measured with relatively high accuracy and precision, may be necessary to uncover novel SNP×PA interactions.
|
10.1371/journal.pcbi.1000423 | Drug Discovery Using Chemical Systems Biology: Repositioning the Safe Medicine Comtan to Treat Multi-Drug and Extensively Drug Resistant Tuberculosis | The rise of multi-drug resistant (MDR) and extensively drug resistant (XDR) tuberculosis around the world, including in industrialized nations, poses a great threat to human health and defines a need to develop new, effective and inexpensive anti-tubercular agents. Previously we developed a chemical systems biology approach to identify off-targets of major pharmaceuticals on a proteome-wide scale. In this paper we further demonstrate the value of this approach through the discovery that existing commercially available drugs, prescribed for the treatment of Parkinson's disease, have the potential to treat MDR and XDR tuberculosis. These drugs, entacapone and tolcapone, are predicted to bind to the enzyme InhA and directly inhibit substrate binding. The prediction is validated by in vitro and InhA kinetic assays using tablets of Comtan, whose active component is entacapone. The minimal inhibition concentration (MIC99) of entacapone for Mycobacterium tuberculosis (M.tuberculosis) is approximately 260.0 µM, well below the toxicity concentration determined by an in vitro cytotoxicity model using a human neuroblastoma cell line. Moreover, kinetic assays indicate that Comtan inhibits InhA activity by 47.0% at an entacapone concentration of approximately 80 µM. Thus the active component in Comtan represents a promising lead compound for developing a new class of anti-tubercular therapeutics with excellent safety profiles. More generally, the protocol described in this paper can be included in a drug discovery pipeline in an effort to discover novel drug leads with desired safety profiles, and therefore accelerate the development of new drugs.
| The rise of multi-drug resistant (MDR) and extensively drug resistant (XDR) tuberculosis around the world, including in industrialized nations, poses a great threat to human health. This resistance highlights the need to develop new, effective and inexpensive anti-tubercular agents. Unfortunately, conventional approaches have yielded very few successes in the field of anti-infective drug discovery. It is a challenge to design drugs with both efficacy and safety. These challenges are reflected in the high costs involved in bringing new drugs to market. It has been estimated that the cost to launch a successful new drug is in excess of US$800 million. We have developed a novel computational strategy to systematically identify cross-reactivity between different drug target families. In this paper we demonstrate the strength of this approach through the discovery that existing commercially available drugs prescribed for the treatment of Parkinson's disease have the potential to treat MDR and XDR tuberculosis. The protocol described herein can be included in a drug discovery pipeline in an effort to accelerate the development of new drugs with reduced side effects.
| Tuberculosis, which is caused by the bacterial pathogen Mycobacterium tuberculosis (M.tuberculosis), is a leading cause of mortality among infectious diseases. It has been estimated by the World Health Organization (WHO) that almost one-third of the world's population, around 2 billion people, is infected with the disease [1]. Every year, more than 8 million people develop an active form of the disease, which subsequently claims the lives of nearly 2 million. This translates to over 4,900 deaths per day, and more than 95% of these are in developing countries [2]. In 2002, the WHO estimated that if the worldwide spread of tuberculosis was left unchecked, then the disease would be responsible for approximately 36 million more deaths by the year 2020.
Despite the current global situation, anti-tubercular drugs have remained largely unchanged over the last four decades [3]. The widespread use of these agents, and the time needed to remove infection, has promoted the emergence of resistant M.tuberculosis strains. Multi-drug resistant tuberculosis (MDR-TB) is defined as resistance to the first-line drugs isoniazid and rifampin. The effective treatment of MDR-TB necessitates the long-term use of second-line drug combinations, an unfortunate consequence of which is the emergence of extensively drug resistant tuberculosis (XDR-TB) – M.tuberculosis strains that are resistant to isoniazid plus rifampin, as well as key second-line drugs, such as ciprofloxacin and moxifloxacin. XDR-TB is extremely difficult to treat because the only remaining drug classes exhibit very low potency and high toxicity. The rise of XDR-TB around the world, including in industrialized nations, imposes a great threat on human health, therefore emphasizing the need to identify new anti-tubercular agents as an urgent priority [4].
Currently, anti-infective therapeutics are discovered and developed by either de novo strategies, or through the extension of available chemical compounds that target protein families with the same or similar structures and functions. De novo drug discovery involves the use of high throughput screening techniques to identify new compounds, both synthetic and natural, as novel drugs. Unfortunately, this approach has yielded very few successes in the field of anti-infective drug discovery [5]. Indeed, the progression from early-stage biochemical hits to robust lead compounds is commonly an unfruitful process. The identification of both molecular targets that are essential for the survival of the pathogen, and compounds that are active on intact cells, is a challenging task. Even more formidable, however, is the requirement for appropriate potency levels and suitable pharmacokinetics, in order to achieve efficacy in small animal disease models [5]. These challenges are reflected in the high costs involved in bringing new drugs to market. In fact, it has been estimated that the successful launch of a single new drug costs more than US$800 million [6].
Two alternative drug discovery strategies that circumvent some of the challenges associated with de novo drug discovery are the label extension and ‘piggy-back’ strategies, both of which are widely employed for the discovery of novel therapeutics to treat tropical diseases. Label extension is a fast-track approach that involves the extension of the indications of an existing treatment to another disease. Some of the most important anti-parasitic drugs in use today, such as praziquantel for schistosomiasis, were derived from the label extension process. The major advantages of label extension are the significant reductions in cost and time to market that can be achieved. Alternatively, when a molecular target that is present in a pathogen is under investigation for other commercial indications, it is possible to adopt the ‘piggy-back’ strategy by utilizing the identified chemical starting points. Examples of this approach include the anti-malarial screening of a lead series of cysteine protease inhibitors for the treatment of osteoporosis, and histone deacetylase inhibitors for use in cancer chemotherapy [5].
One of the main aims of drug discovery is to develop safe and effective therapeutic agents through the optimization of binding to a specific protein target. In this way, undesirable effects resulting from side scatter pharmacology are minimized. However, the recent and rapid completion of numerous genome sequencing projects has revealed that proteins involved in entirely different biochemical pathways, and even residing in different tissues and organs, may possess functional binding pockets with similar shapes and physiochemical properties [7]. Therefore, chemical matter for one target could be considered as the basis for leads for an entirely different target. Recent work on large scale mapping of polypharmacology interactions by Paolini et al. [8] revealed the extent of promiscuity of drugs and leads across the proteome. They discovered that around 35% of 276,122 active compounds in their database had observed activity for more than one target. Whilst the majority of these promiscuous compounds were found to be active against targets within the same gene family, a significant number (around 25%) had recorded activity across different gene families.
The finding that so many drugs interact with more than one target provided the rationale behind the selective optimization of side activities (SOSA) approach recently developed by Wermuth [9],[10]. The SOSA approach involves the use of old drugs for new pharmacological targets, which is a valuable concept considering the finite number of small molecules that can be safely administered to humans. The process itself involves screening a limited number of structurally diverse drug molecules, and then optimizing the hits so that they show a stronger affinity for the new target and a weaker affinity for the original target(s). In this way, it is possible to derive a whole panel of new active molecules from a single marketed drug. Since the screened drug molecules already have known safety and bioavailability in humans, the overall time and cost of drug discovery is significantly reduced when compared with de novo strategies.
We have developed a novel computational strategy to identify off-targets of major pharmaceuticals on a proteome-wide scale [11]–[15]. Our methodology extends the scope of the SOSA concept effectively and systematically across gene families, and is more likely to be successful in achieving the ultimate goal of providing new drugs from old ones. Our chemical systems biology approach proceeds as follows:
Our approach essentially explores complex protein-ligand interaction networks on a proteome-wide scale. The lead compound can be discovered from all drug targets across different gene families. Moreover, lead optimization can focus on compounds with excellent safety profiles and known clinical outcomes. In this way, our approach has the potential to increase the rate of successful drug discovery and development, whilst reducing the costs involved.
In the present study we demonstrate the efficiency and efficacy of our chemical systems biology approach through the discovery of safe chemical compounds with the potential to treat MDR-TB and XDR-TB. The identified compounds are entacapone and tolcapone. These drugs primarily target human catechol-O-methyltransferase (COMT), which is involved in the breakdown of catecholamine neurotransmitters such as dopamine. They are used as adjuncts to treat Parkinson's disease by increasing the bioavailability of the primary drug levodopa, which is a substrate of COMT. Entacapone and tolcapone are predicted to inhibit M.tuberculosis enoyl-acyl carrier protein reductase (InhA), which is essential for type II fatty acid biosynthesis and the subsequent synthesis of the bacterial cell wall [16]. InhA is the target of the anti-tubercular drugs isoniazid [17] and ethionamide [18]. Similar to newly developed direct InhA inhibitors [3], [19]–[21], entacapone and tolcapone require no enzymatic activation to bind InhA. Thus they may avoid the commonly observed resistance mechanism to isoniazid and ethionamide that is exhibited by many MDR strains. Our computational predictions have been partially validated by demonstrating that the entacapone drug tablet Comtan inhibits the growth of M.tuberculosis at the minimal inhibition concentration (MIC99) of entacapone of 260.0 µM, well below the concentration leading to neuroblastoma cellular toxicity. The direct inhibition of InhA by entacapone is further confirmed by experimental enzyme kinetic assays, in which Comtan is shown to reduce InhA activity by up to 47% at the effective entacapone concentration of 80 µM. Since entacapone has an excellent safety profile with few side effects, it shows potential as a drug lead in the development of a new class of anti-tubercular therapeutics with favorable ADME/Tox properties.
Recently, Xie and Bourne developed a sequence order independent profile-profile alignment (SOIPPA) algorithm [11], which they subsequently used to detect common binding sites among proteins unrelated in sequence and/or function. Their studies implied an evolutionary relationship between the NAD-binding Rossmann fold and several other fold classes, including the SAM-binding domain of the SAM-dependent methyltransferases, through similarities between their co-factor binding sites. It is interesting to note that both nicotinamide adenine dinucleotide (NAD) and S-adenosyl methionine (SAM) include adenine as a common molecular fragment. In fact, previous studies have shown that adenine binding pockets from proteins lacking significant homology will share common physiochemical properties [11]. These findings, plus the versatility of our method, form the basis for the present study.
Entacapone and tolcapone are drugs that block the ligand binding site of COMT, a member of the SAM-dependent methyltransferase superfamily, in the presence of the SAM co-factor. They are used as adjuncts in Parkinson's disease therapy, to prevent the metabolism of levodopa to 3-O-methyldopa, therefore improving levodopa bioavailability and increasing its delivery to the brain [22]. The dopamine precursor levodopa has been the key drug for symptomatic treatment of Parkinson's disease for more than 30 years. Since COMT is a SAM-dependent methyltransferase, it is possible that it may possess a ligand binding pocket similar to those found in protein domains belonging to the NAD-binding Rossmann fold, as their co-factor binding sites are strikingly similar [11]. When entacapone and tolcapone were docked into 215 NAD-binding proteins from multiple species, the InhA enzyme from several different organisms, including M.tuberculosis, was consistently highly ranked (see Tables S1 and S2). Since InhA is the primary target of the anti-tubercular drugs isoniazid [17] and ethionamide [18], entacapone and tolcapone can potentially inhibit the InhA ligand binding site directly. Indeed, alignment of the COMT and InhA binding sites by the SOIPPA algorithm revealed similarities in the positioning of both their co-factors and ligands (Figure 1).
As shown in Figure 2, the existing InhA inhibitor with the greatest 2D similarity to entacapone is 3-(6-aminopyridin-3-yl)-N-methyl-N-[(1-methyl-1H-indol-2-yl)methyl]acrylamide (AYM) [23] (Tanimoto coefficient = 0.155), whereas the existing InhA inhibitor with the greatest 2D similarity to tolcapone is 3-[(acetyl-methyl-amino)-methyl]-4-amino-N-methyl-N-[(1-methyl-1H-indol-2-yl)-methyl]-benzamide (ZAM) [23] (Tanimoto coefficient = 0.173) (see Table S3). Neither of their p-values (AYM; 0.065 and ZAM; 0.205) are significant at the 0.05 level, implying that none of the investigated InhA inhibitors exhibit significant molecular similarity to either entacapone or tolcapone. Therefore, it is unlikely that ligand-based screening methods would be able to identify entacapone and tolcapone as potential InhA inhibitors.
Table 1 shows the predicted binding affinities of entacapone and tolcapone towards InhA. Since they fall within the range of binding affinities exhibited by the known InhA inhibitors, this not only provides a further implication of the cross-reactivity between InhA and COMT, but also suggests that entacapone and tolcapone are able to inhibit InhA directly.
Unfortunately it is difficult to identify entacapone as a lead compound using conventional virtual screening because it is only ranked at 15,892 and 9,719 by eHiTs and Surflex among 20,000 randomly selected drug-like molecules, respectively. While advanced virtual screening techniques such as the relaxed complex scheme [24], which combines a docking algorithm with molecular dynamics simulations, may improve the ranking of these potential lead compounds [25], they demand significant computational resources. Thus, SOIPPA, a ligand binding site similarity based method, provides an efficient way of identifying potential drug-like leads that have well-established pharmacokinetics and pharmacodynamics properties.
Although the 2D similarities between entacapone and existing InhA inhibitors have been shown to be statistically insignificant, entacapone shares a similar molecular size and common functional groups with several of the InhA inhibitors (see Table 1). For example, entacapone and five of the InhA inhibitors (468, 566, 641, 665 and 774) all possess a single benzene and amide moiety. More importantly, the predicted binding poses of the benzene ring and the amide bond of entacapone are similar to those of the InhA inhibitors, with root mean square deviations (RMSDs) of as little as 2.87Å and 1.05Å, respectively (in the case of 566). Although tolcapone does not share the same amide moiety, the predicted binding pose of its benzene ring has an RMSD of only 1.01Å from that of 566, further demonstrating the potential of these drugs as InhA inhibitors.
Previous studies have highlighted the necessity of the interaction between the catechol oxygens of COMT inhibitors with an Mg2+ ion in the active site [26]. From the predicted binding poses of entacapone and tolcapone docked with InhA, three potential interaction sites of Mg2+, including the Asp110, Asp115, and Glu210 residues of InhA, have been identified and are shown in Figure 3. The closest residue, Glu210, is positioned at a distance of 13.23Å from the nitrite group of entacapone, and at a distance of 11.54Å from the nitrite group of tolcapone (see Figure S2). Although it is possible that the conformation of the side chains may adjust to reduce this distance under in vitro or in vivo conditions, such a large distance may hinder the formation of a coordinate bond by the Mg2+ ion between the InhA active site and each of the two drugs. Consistent with the predicted binding pose, enzyme kinetic assays indicate that the addition of an Mg2+ ion has no effect on the inhibition of InhA by entacapone. This therefore provides us with opportunities to optimize entacapone and tolcapone, by reducing or removing their conjugation to the Mg2+ ion, so that they exhibit a weaker affinity for the original target COMT.
The partition coefficient (logP) is the ratio of the concentrations of an unionized compound in the two phases of a mixture of octanol and water at equilibrium, whereas the distribution coefficient (logD) is the ratio of the sum of the concentrations of all forms of the compound (both ionized and unionized) in each of the two phases. Since logD is pH dependent, the pH at which it was measured is specified, as shown in Table 2. Entacapone and tolcapone were discovered to have generally higher logP and logD values than most of the existing anti-tubercular drugs investigated. According to Lipinski's rule of five [27], poor absorption or permeation is more likely for compounds with a logP value of greater than 5.0. Such compounds are considered non-drug-like and are commonly filtered out in the early stages of drug discovery. However, entacapone and tolcapone are prescribed drugs and have been clinically tested with desired pharmacokinetics profiles. Indeed, entacapone can be rapidly absorbed with a Tmax of approximately one hour (http://www.rxlist.com/comtan-drug.htm). The high logP values of entacapone and tolcapone are therefore acceptable regardless of Lipinski's rule. Moreover, they suggest that entacapone and tolcapone are more hydrophobic than existing drugs, and would therefore pass more easily through the M.tuberculosis cell envelope. Although entacapone demonstrates poor solubility, dissolution enhancers such as croscarmellose sodium are used in the formulation of Comtan [28]. This example well illustrates that drug repurposing would accelerate drug discovery and development by bypassing the time consuming steps of ADME/Tox evaluation and drug formulation.
Entacapone and tolcapone were evaluated for their ability to inhibit growth of M.tuberculosis using a 96 well microplate assay. A 99.0% reduction in growth was observed with concentrations of between 62.5 µg/ml and 125 µg/ml for each drug. The sensitivity of M.tuberculosis to entacapone was confirmed by quantitative growth on agar plates containing known amounts of the drug. The minimum inhibitory concentration was between 62.5 µg/ml and 125 µg/ml (Table 3), therefore confirming the result from the microplate assay. These results support the computational predictions that entacapone and tolcapone have inhibitory activity against M.tuberculosis and may therefore be considered as lead compounds.
The ability of entacapone to directly inhibit InhA was evaluated using enzyme kinetics. Due to the strong UV absorbance of entacapone over a wide range of 300–400 nm (see Figure S4), and the poor solubility of entacapone in water, the highest concentration of Comtan that could be used in the assays was 90 µg/ml (the corresponding concentration of pure entacapone is 24.9 µg/ml), at which concentration InhA is approximately 47% inhibited. Fitting of the data in Table 4 to a dose response equation provided an IC50 value of 24±3 µg/ml (79±10 µM) (see Figure S3). Since Mg2+ is critical for entacapone and tolcapone to inhibit COMT, assays were repeated in the presence of 5 mM Mg2+ to explore whether or not metal ion chelation could improve the affinity of the drug for InhA. However, the inclusion of Mg2+ in the assay had no effect on the IC50 value for enzyme inhibition. The progress curve analysis for the inhibition of InhA by Comtan shows that the UV absorbance is linearly time-dependent up to one hour (see Figure S5), indicating that entacapone is not a slow-onset inhibitor.
The sharp contrast between the fatty acid biosynthetic pathway in humans (FAS-I) and that found in prokaryotes (FAS-II) has established this bacterial pathway as an attractive target for the design of new antibacterial agents [20],[29]. The FAS-II pathway in M.tuberculosis is involved in the production of mycolic acids, which, along with peptidoglycan and arabinogalactan, are central constituents of the mycobacterium cell wall [2],[29]. InhA catalyzes the final, rate-determining step in the fatty acid elongation cycle by converting trans-2-enoyl-ACP to acyl-ACP in an NADH-dependent reaction [3]. This crucial regulatory enzyme is the primary molecular target of isoniazid, which has been used as a frontline anti-tubercular agent for the past 40 years [20],[29]. As a pro-drug, the activity of isoniazid is dependent on its activation by KatG, a catalase/peroxidase enzyme. KatG oxidizes isoniazid to an acyl radical that covalently binds to NADH, and functions as a potent inhibitor of InhA. Unfortunately, it is this activation requirement that allows M.tuberculosis to acquire resistance to the drug. Indeed, mutations in the KatG gene account for around half of all isoniazid-resistant clinical isolates [3]. Direct InhA inhibitors that avoid this activation requirement are not susceptible to this resistance mechanism [30]. Triclosan and the diazoborines are well-known InhA inhibitors that do not require activation, but unfortunately they are not suitable for human treatment due to their respective poor solubility and toxicity [3]. However, a new class of high affinity direct InhA inhibitors, consisting of alkyl diphenyl ethers of triclosan derivatives, have been found to exhibit activity against drug-resistant strains of M.tuberculosis [21]. In addition to these alkyl diphenyl ethers, the arylamides [20],[31], indole-piperazines, pyrazole-based inhibitors [3], and indole-based inhibitors [23],[32], have recently been described as other classes of direct InhA inhibitors.
Using a novel computational strategy we have predicted that entacapone and tolcapone will directly inhibit InhA. Our prediction was subsequently confirmed by in vitro antibacterial and enzyme kinetic assays using Comtan tablets containing the active component entacapone. Thus entacapone and tolcapone are promising lead compounds against drug-resistant strains of M.tuberculosis. These drugs are currently in clinical use, although the association of tolcapone with hepatotoxicity has caused the drug to be placed under strict regulation in the United States [33]. Entacapone, which is not associated with the same hepatotoxic risks, is therefore more attractive as a drug lead. Interestingly, a recent study showed that when patients suffering from Parkinson's disease were treated with rifampin and isoniazid, their condition was observed to improve [34]. Since rifampin inhibits DNA-dependent RNA polymerase, this observation implies cross-reactivity between the M.tuberculosis InhA, the target of isoniazid, and the drug targets of Parkinson's disease, which is consistent with our predictions that InhA inhibitors can also inhibit COMT.
Recent studies have shown that triclosan can trigger the upregulation of M.tuberculosis detoxification mechanisms that result in its metabolism or efflux from the cell [21]. For instance, triclosan has been shown to induce the expression of an aromatic dioxygenase involved in the degradation of arenes. Since triclosan consists of a diphenyl ether structure, it is thought that the induction of this enzyme may serve to degrade, and hence detoxify, triclosan [35]. The subsequent modification of triclosan derivatives has led to high affinity alkyl diphenyl ether InhA inhibitors that upregulate neither efflux pumps nor aromatic dioxygenase [21]. Although gene transcription studies are required to determine the ability of entacapone and tolcapone to cause upregulation of the aromatic dioxygenase, it is speculated that the strong electron-withdrawing nitrite groups of these drugs may result in significantly lower reduction potentials, therefore making them less prone to oxidation by this enzyme. The narrow ranges of the MIC99 and the IC50 of Comtan support this hypothesis.
From our current experimental results, it is not possible to determine the effect of the inactive ingredients in the Comtan tablets (magnesium stearate, microcrystalline cellulose, hydroxypropyl methylcellulose, yellow iron oxide and red iron oxide, titanium dioxide, sucrose, mannitol, hydrogenated vegetable oil, polysorbate 80, glycerol 85%, croscarmellose sodium) on the growth of M.tuberculosis. It is unlikely that the major formulations in the Comtan tablets (magnesium stearate, microcrystalline cellulose, hydroxypropyl methylcellulose, iron oxide, titanium dioxide, sucrose) directly affect M.tuberculosis growth because they are the same as the active ingredients found in anti-tubercular drugs such as Rifater (isoniazid/pyrazinamide/rifampin combination tablet), Rifamate (isoniazid/rifampin tablet), and Priftin, whose active component is a rifamycin derivative (http://www.rxlist.com). Other ingredients such as polysorbate 80 and croscarmellose sodium are mainly used to enhance the dissolution and stability of entacapone [28]. It will be particularly interesting if they are active against the growth of M.tuberculosis, as they are commonly used food and drug additives. We have observed that a higher concentration of pure entacapone than that which is present in Comtan is required to achieve the same rate of inhibition in the InhA assay. Additional experiments such as X-ray crystallography and mass spectroscopy need to be conducted in order to investigate the precise mechanisms of action of Comtan and entacapone against M.tuberculosis.
Although we have demonstrated that Comtan is active against M.tuberculosis in vitro, further studies are required in order to transform it into an anti-tubercular therapeutic. The active component of entacapone in Comtan has an MIC99 for M.tuberculosis of approximately 80 µg/ml (262 µM), and an estimated IC50 value for InhA inhibition of above 80 µM. However, it exhibits very low cellular toxicity, with no effect on neuroblastoma cell lines that provide an in vitro model for high throughput toxicity screening [36] at concentrations of up to 500 µM [37], thus making the concentration required for M.tuberculosis inhibition not unreasonable for a lead compound.
The possibility that entacapone inhibits other enzymes besides InhA cannot be excluded. From our initial studies of the ligand binding site similarity network in the M.tuberculosis structural genome, InhA is one of the most promiscuous proteins, having a similar ligand binding site to more than 20 other enzymes [15]. This implies that an InhA inhibitor can potentially interact with multiple targets. If it is proven in future studies that entacapone can inhibit multiple targets simultaneously, the potential of entacapone as an anti-tubercular drug is even more promising, as “dirty” drugs lessen the likelihood of emergent resistance and higher clinical efficacy than exquisitely selective drugs [38],[39].
A further challenge is transforming entacapone into a nanomolar inhibitor without impacting its ADME/Tox profile. A series of direct InhA inhibitors with IC50's ranging from 1 nM to greater than 100 µM are available in the RCSB Protein Data Bank (PDB) [40]. Thus it is possible to build reliable 3D QSAR models in order to guide the lead optimization process. Since entacapone has been in clinical use for many years, the accumulated knowledge of its pharmacokinetics and pharmacodynamics will be invaluable in predicting ADME/Tox properties of the compound and its derivatives.
The continuing emergence of M.tuberculosis strains resistant to all existing, affordable drug treatments means that the development of novel, effective and inexpensive drugs is an urgent priority [3]. Our chemical systems biology approach to drug discovery revealed that Comtan, with the active component entacapone, shows potential for use as an anti-tubercular drug. Entacapone may adopt different inhibition mechanisms from the first- and second-line drugs that result in MDR and XDR M.tuberculosis strains. Moreover, it has an excellent safety profile with few side effects, and is commercially available. Therefore, entacapone can potentially be used as a lead compound to develop a new class of anti-tubercular drugs.
By integrating techniques from ligand binding site similarity, small molecule similarity and protein-ligand docking, our chemical systems biology approach is able to model protein-ligand interaction networks on a proteome-wide scale. The systematic use of small molecules to probe biological systems will provide us with valuable clues as to the molecular basis of cellular functions, and at the same time it will shift the conventional one-target-one-drug discovery process to a new multi-target-multi-drug paradigm.
Previously, the SOIPPA algorithm had revealed a highly significant similarity (p-value = 2.7e-5) between the NAD binding site of the Rossmann fold and SAM binding site of the SAM methyltransferases [11]. In order to identify similar ligand binding sites adjunct to the co-factor binding site, further docking studies were carried out on the ligand binding sites of proteins that bind NAD as a co-factor. Freely available docking software eHiTs 6.2 [41] and Surflex2.1 [42] were selected due to their relatively fast speed, high accuracy and ease of automation in large-scale docking studies. It is worth noting that when preparing PDB files for any of the docking studies, NAD and SAM co-factors were added as one of residues of the protein chain.
215 non-redundant proteins with NAD co-factors were downloaded from the RCSB Protein Data Bank (PDB) [40],[43]. The ideal coordinates for entacapone and tolcapone were downloaded from DrugBank [44]. Both Surflex and eHiTs were used to dock both entacapone and tolcapone onto each of the 215 proteins, and the proteins that produced the highest docking scores were investigated further. Enoyl-acyl carrier protein reductases (InhAs) from several different organisms, including M.tuberculosis and Toxoplasma gondii, were identified as proteins to which entacapone and tolcapone showed favorable binding affinities (see Tables S1 and S2). However, the M.tuberculosis InhA (PDB ID: 2H7M) is the focus of this paper due to its importance as a drug target for the treatment of tuberculosis.
Unfortunately, the human COMT protein that is the drug target of entacapone and tolcapone is absent from the RCSB PDB. The only COMT structure available at the time of writing is that from the brown rat, Rattus norvegicus. A standard protein BLAST [45] search revealed that human and rat COMT share 81% sequence identity without insertion or deletion. Moreover, their functional site residues were found to share 100% identity (see Figure S1). Therefore, rat COMT (PDB ID: 2CL5) was used as an accurate representation of human COMT throughout this study. The SOIPPA algorithm was used to align the 2H7M structure with that of 2CL5 so that their respective NAD and SAM co-factor binding sites were aligned. The aligned proteins were then visualized using Accelrys DS Visualizer (http://www.accelrys.com/products/downloads/ds_visualizer/).
The RCSB PDB was queried for proteins with sequence similarity to chain A of 2H7M, using an e-value cut off of 0.0001. Between them, the resulting proteins bound a total of 22 different InhA inhibitors (including the ligand of 2H7M). OpenBabel (http://openbabel.org) was used to calculate the 2D small molecule similarity between these InhA inhibitors and both entacapone and tolcapone. In order to create background distributions for comparison, all drug-like molecules were downloaded from the ZINC database [46], and a subset of 20,000 molecules was extracted randomly. The 2D similarities of each of these molecules to both entacapone and tolcapone were calculated using OpenBabel, and density distributions of the scores were plotted using R 2.5.0 [47]. P-values corresponding to the 2D similarity scores of the 22 InhA inhibitors were subsequently calculated from both density distributions (see Table S3).
Nine M.tuberculosis InhA structures (PDB IDs: 1P44, 1P45, 2B36, 2B37, 2H7I, 2H7L, 2H7M, 2H7N and 2H7P) were downloaded from the RCSB PDB. The ten inhibitors of these InhAs (Ligands: Pyrrolidine carboxamide s3, GEQ, TCL, 5PP, 8PS, 566, 665, 641, 744 and 468) were downloaded from the RCSB PDB as ideal coordinates. In addition, the ideal coordinates of entacapone and tolcapone were generated using CORINA (http://www.molecular-networks.com/online_demos/corina_demo.html). All twelve molecules were docked onto the nine InhA structures, as well as onto COMT using eHiTs and Surflex. The mean and standard deviation of the docking scores of each molecule with all ten of the InhAs were calculated, and the docking scores were tabulated for comparison.
The predicted binding poses of entacapone and tolcapone with the various different M.tuberculosis InhAs were visualized and analyzed using Accelrys DS Visualizer. Distances between the nitrite groups of entacapone and tolcapone and the surrounding aspartic acid and glutamic acid residues of InhA were measured. RMSDs between the benzene rings and the amide bonds of entacapone and the native ligands were calculated, in addition to the RMSDs between the benzene rings of tolcapone and the native ligands.
The ideal coordinates of entacapone, tolcapone, five first-line anti-tubercular drugs (ethambutol, isoniazid, pyrazinamide, rifampicin and streptomycin) and three second-line anti-tubercular drugs (ciprofloxacin, moxifloxacin and aminosalicylic acid) were downloaded from DrugBank [44]. Both ChemAxon's Calculator from Marvin Beans (http://www.chemaxon.com/marvin) and ChemSilico Predict (http://www.chemsilico.com) were used to calculate a) the partition coefficient (logP) and b) the distribution coefficient (logD) of all ten drug molecules.
In order to prepare stock solutions of each drug, one tablet of Comtan (Sandoz) containing 200 mg of entacapone, and one tablet of Tasmar (Valeant) containing 100 mg of tolcapone were each ground to a fine powder and completely dissolved in dimethylsulfoxide (DMSO). For the microplate assay, serial dilutions of each drug were made in Middlebrook 7H9 media supplemented with ADS (albumin, dextrose, and saline) and Tween 80 [48] in a volume of 100 µl. A culture of M.tuberculosis Erdman was grown to mid log in 7H9 plus ADS and Tween 80, and adjusted to an optical density600 of 0.2. 100 µl of bacteria was subsequently added to each well. The cultures were incubated for 14 days until the control wells containing only medium developed a dense layer of bacteria. Wells were visually scored for the amount of growth in comparison to the control wells. The dilution of drug that produced almost complete inhibition of growth was scored as MIC99. The agar plate assay was carried out as previously described [48] using Middlebrook 7H9 plates supplemented with OADC and containing known amounts of entacapone. In order to determine the MIC99, the number of bacteria that grew in the presence of each concentration of entacapone was compared with the number of bacteria that grew on the plate with no drug.
Comtan (entacapone) tablets were ground into powder, and dissolved in DMSO. Kinetic assays using trans-2-dodecenoyl-Coenzyme A (DD-CoA) and wild-type InhA were performed as described previously [49]. Reactions were initiated by addition of InhA to solutions containing 250 µM NADH, 25 µM DD-CoA, 0 or 5 mM MgCl2 and inhibitor in 30 mM PIPES and 150 mM NaCl, pH 6.8 buffer. IC50 values were calculated by fitting the initial velocity data to equation 1;(1)where I is the inhibitor concentration and y is the percent activity. Data analysis was performed using Grafit 4.0 (Erithacus Software Ltd.). The IC50 curve fitting is shown in the Figure S3.
A progress curve was calculated in order to study the slow-onset mechanism of inhibition of Comtan (entacapone). InhA activity was monitored by adding the enzyme (10 nM) to assay mixtures containing 8% V/V glycerol, 0.1 mg/ml BSA, 2% V/V DMSO, 300 µM DD CoA, 250 µM NADH, 200 µM NAD+ and inhibitors. Reactions were monitored until the progress curve became linear, therefore indicating that the steady-state had been reached. Subsequently, a low enzyme concentration and a high substrate concentration were used to ensure that the depletion of the substrates was minimal and would not affect the reaction rate, so that the progress curve in the absence of inhibitors was linear. Progress curve data were collected for up to 1 hour (see Figure S5).
|
10.1371/journal.ppat.1008009 | A robust human norovirus replication model in zebrafish larvae | Human noroviruses (HuNoVs) are the most common cause of foodborne illness, with a societal cost of $60 billion and 219,000 deaths/year. The lack of robust small animal models has significantly hindered the understanding of norovirus biology and the development of effective therapeutics. Here we report that HuNoV GI and GII replicate to high titers in zebrafish (Danio rerio) larvae; replication peaks at day 2 post infection and is detectable for at least 6 days. The virus (HuNoV GII.4) could be passaged from larva to larva two consecutive times. HuNoV is detected in cells of the hematopoietic lineage and the intestine, supporting the notion of a dual tropism. Antiviral treatment reduces HuNoV replication by >2 log10, showing that this model is suited for antiviral studies. Zebrafish larvae constitute a simple and robust replication model that will largely facilitate studies of HuNoV biology and the development of antiviral strategies.
| Human norovirus (HuNoV) is the number one agent of viral gastroenteritis worldwide. It can infect people of all age groups, resulting in 700 million infections and 219,000 deaths each year. Outbreaks of acute HuNoV gastroenteritis occur often, but chronic infections also happen in people with immune deficiencies. Despite its clinical relevance, studying the virus has been a decades-long challenge because no simple and efficient cultivation system existed. This has recently started to change; we here contribute with an important step forward by establishing an in vivo model system to study HuNoV replication using zebrafish larvae. We inject a HuNoV in the yolk (food reserve) of 3-day-old zebrafish larvae, which results in high virus titers up to 6 days after inoculation. We could detect the virus in the tissues of the infected larvae by a variety of techniques including histology, which showed us in which organs the virus is present. Importantly, by adding an antiviral to the water in which the larvae swim, we could significantly reduce the virus levels in the larvae. This means that testing small molecules and developing the first antiviral therapy for HuNoV will be much easier from hereon.
| Human noroviruses (HuNoVs) are an important cause of epidemic and sporadic acute gastroenteritis worldwide; annually about 700 million people develop a HuNoV infection resulting in ~219,000 deaths and a societal cost estimated at 60 billion US dollars [1]. Large outbreaks of norovirus gastroenteritis are frequent and have a significant impact in terms of morbidity, mortality and health care costs, in particular in hospital wards and nursing homes. Chronic norovirus infections present a problem for a large group of immunodeficient patients, who may present with diarrhea for several months. Furthermore, in countries where routine rotavirus vaccination has been implemented, noroviruses are the most common cause of severe childhood diarrhea resulting in important morbidity and mortality [2]. Knowledge on the biology and pathogenesis of human noroviruses largely depends upon the development of robust and physiologically relevant cultivation systems. The available model systems carry important limitations. HuNoV replication has been reported in large animals such as chimpanzees, gnotobiotic pigs and calves. However, these animals are either not suited for extensive studies or are, in the case of chimpanzees no longer allowed due to ethical reasons [3–5]. Importantly, a HuNoV mouse model was described in BALB/c Rag-γ c-deficient mice, but only a short-lasting replication was achieved, which limits its applications [6]. Standard cell culture models are to date not available, but first steps towards this have been given by establishing that (i) human B-cells are susceptible to HuNoV and that (ii) HuNoV can be cultivated in stem-cell-derived enteroids [7–9]. There is thus an urgent need for simpler, more robust, widely available HuNoV replication models. Such models should contribute to a better understanding of the biology of HuNoV replication and infection, this will significantly facilitate larger-scale research efforts, such as the development of therapeutic strategies.
Zebrafish (Danio rerio) are optically-transparent tropical freshwater fish of the family Cyprinidae that are widely used as vertebrate models of disease. They have remarkable genetic, physiologic and pharmacologic similarities to humans. Compared to rodents, the maintenance and husbandry costs are very low. Zebrafish have high fecundity and using their offspring is in better compliance with the 3Rs principles of humane animal experimentation (EU Directive 2010/63/EU). The immune system of zebrafish is comparable to that of humans; there are B and T cells, macrophages, neutrophils and a comparable set of signaling molecules and pathways [10]. Whereas innate immunity is present at all developmental stages, adaptive immunity develops after 4–6 weeks of life [11, 12]. Host-pathogen interactions can be studied, as zebrafish are naturally infected by multiple bacteria, protozoa and viruses that affect mammals [11]. Infection of zebrafish larvae has been shown with some human viruses (herpes simplex virus type 1, influenza A virus, and chikungunya virus) [13–15] as well as enteric bacteria, e.g. E. coli, Listeria, Salmonella, Shigella and Vibrio [11]. The intestinal tract of zebrafish is comprised of large folds of an epithelial lining, a lamina propria containing immune cells and underlying smooth muscle layers [16]. Enterocytes, goblet cells, enteroendocrine cells, and possibly M-cells are present, but not Paneth cells or Peyer’s patches [12]. Intestinal tuft cells, which were recently shown to be a target cell for the mouse norovirus (MNV) [17], have been described in teleost fish thus are likely present in zebrafish. There is an intestinal bulb (instead of a stomach) and a mid- and posterior intestine [16]. Epithelial cells show a high-turnover from base to tip, with intestinal epithelial stem cells at the base and apoptotic cells at the tips [16]. A resident commensal microbiota is present (comprising most bacterial phyla of mammals) and serve analogous functions in the digestive tract [11, 16]. Here we report a robust replication model of HuNoV in zebrafish larvae.
Zebrafish larvae were injected with a PBS suspension of a HuNoV positive stool sample at 3 days post-fertilization (dpf). At this time point, zebrafish larvae have hatched and organs are formed (including the full-length gastrointestinal tract). Three nL, containing 3.4 x 106 viral RNA copies of HuNoV GII.P7-GII.6 (1.1 x 1013 RNA copies/g of stool), were injected in the yolk of the larvae (which provides nutrition during early larval stage). Each day post-infection (pi), the general condition of the zebrafish larvae was assessed microscopically and these were harvested in groups of 10 for viral RNA quantification by RT-qPCR [14, 15]. To detect input virus, in every independent experiment, 10 larvae were harvested at day 0 pi (specifically 1 h pi). A maximum increase of ~2.5 log10 in viral RNA copies compared to day 0 was detected at day 2 pi (Fig 1A); high levels of viral RNA remained detectable for at least 6 days pi (Fig 1A). When larvae were injected with 3 nL of UV-inactivated HuNoV GII.P7-GII.6, no increase in viral RNA titers was detected (Fig 1A). No obvious signs of distress or disease were observed as a result of the viral replication (e.g. changes in posture, swimming behavior or signs of edema). To determine the 50% infectious dose (ID50) for this strain, larvae were injected with 10-fold dilution series of the virus (Fig 1C). The ID50 was calculated to be 1.8 x 103 viral RNA copies. HuNoV antigens were detected using the commercial enzyme immunoassay (EIA) RIDASCREEN (R-Biopharm), in HuNoV GII.P7-GII.6-injected zebrafish larvae harvested at day 3 pi (Fig 1B).
We next investigated the innate immune response to a HuNoV infection of larvae at multiple time points pi. An increased expression of ifn, mx and rsad2/viperin mRNA was detected, with a 8-fold, 144-fold and 266-fold maximum increase, respectively, when compared to PBS injected larvae (Fig 1E). Two additional controls were included, namely (i) larvae injected with a UV-inactivated virus-containing Stool sample and (ii) larvae injected with a stool sample that was negative for HuNoV (Fig 1E). When UV-inactivated virus was injected, a delayed increase of ifn was observed, but this did not trigger the downstream cascade partners mx and rsad2 (Fig 1E). The level of upregulation reported here is in line with the observed induction of the ifn response following an influenza A infection of zebrafish larvae [14]. These same genes (or the related cytokines) were upregulated in other in vivo models, such as in HuNoV-infected calves or MNV-infected mice [5, 18], or in a HuNoV replicon system in the case of viperin [19]. Altogether, this points out that the antiviral signaling cascades that are activated upon a HuNoV infection of zebrafish larvae are relevant and likely the same as in humans. Zebrafish are thus a suitable model for the study of the innate immune response to a HuNoV infection.
Next, injected larvae were treated with a broad-spectrum antiviral, i.e. the viral polymerase inhibitor 2’-C-methylcytidine (2CMC) of which we showed earlier inhibition of MNV replication in vitro and in mice [20, 21], by immersion (whereby the molecule was added to the water). A 2.4 log10 reduction in viral RNA titers was observed at the peak of replication (Fig 1D). The fact that replication can be significantly reduced with an inhibitor of the viral polymerase provides further evidence that HuNoV replicates efficiently in zebrafish larvae and that the model is suitable for antiviral drug development.
One of the hurdles to develop robust replication models for HuNoV is the need to use a stool sample of an infected patient as inoculum. In order to rule out any potential impact of other agents present in the sample we fully characterized the samples used. A viral metagenomics analysis was performed on the clinical sample containing the HuNoV GII.P7-GII.6 used in this study, together with subsequent clinical samples of the same chronically-infected 2.5-year old transplant patient (Fig 2). The viral population consisted predominantly of HuNoV (with a minor presence of anelloviruses, common in patients undergoing immunosuppressive therapy [22]). Mutations that occurred in the virus over the course of the infection were mostly in the capsid-encoding region (Fig 2, S1 Table) and did not affect the kinetics of virus replication in larvae (Fig 2B–2D). In addition, upon injection of HuNoV GII.P7-GII.6 (week 0) in zebrafish larvae no mutations were detected in the viral genome at the peak of replication (2 days pi).
Injection of zebrafish larvae with other HuNoV genotypes was next performed. Injection with the HuNoV GII.P4 New Orleans-GII.4 Sydney strain, recovered from stool samples of two different patients, resulted in a 3.1 and 3.4 log10 increase in viral replication in both cases. A maximum of ~107 viral RNA copies/zebrafish larva was detected at day 2 pi (Fig 3A), the highest observed in this model. Viral non-structural and structural antigens were detected by western blot (Fig 3B, S1 Fig) and by EIA (Fig 3C), respectively. Viral antigens were no longer detected by EIA in 2CMC-treated HuNoV GII.4-infected larvae (Fig 3C). Injection with HuNoV GII.P16-GII.2 (Fig 3D) and GII.P16-GII.3 (Fig 3E) yielded increasing viral RNA titers, although the replication kinetics of GII.P16-GII.3 reached lower titers and at a later time point than that observed for the other genotypes. Slower kinetics of HuNoV GII.3 replication was also observed in stem-cell-derived enteroids [23]. A GI HuNoV, specifically GI.P7-GI.7, replicated with comparable kinetics to GII viruses (Fig 3F). Injection of zebrafish larvae with MNV (genogroup V) yielded no productive infection (S2 Fig), most likely due to the fact that the receptors CD300lf and CD300ld are not encoded by zebrafish [24].
We next passaged the virus from larva to larva. To that end HuNoV GII.P4-GII.4 injected zebrafish larvae was collected at the peak of replication, homogenized in 50 μL PBS, and the supernatant was used to inject new larvae. This was successful for two passages (Fig 4).
Although a yolk injection adds a known virus inoculum to the food of the larvae and is therefore close to the natural route of infection, we next attempted to infect larvae by adding virus to the swimming water (i.e. via immersion). This was done by immersing 4 or 5 dpf larvae (when the mouth has opened and the gastrointestinal tract fully matured) in 1 mL of a HuNoV GII.P4-GII.4 PBS suspension for 8 h, after which the larvae were washed extensively with Danieau’s. No consistent increase in viral replication was noted up to day 5 pi. To determine the preferential site of replication of HuNoV in zebrafish larvae, HuNoV GII.P7-GII.6-injected larvae were dissected at day 3 pi in 4 different parts (yolk, head, body and tail). Viral RNA titers were detected in every part (Fig 5A), whereby the yolk (the initial site of inoculation) had the lowest titers, implying that HuNoV disseminates past the yolk and intestine. To investigate which tissues are infected, sagittal and coronal histological sections of larvae injected with HuNoV GII.P7-GII.6 were harvested at day 3 pi and stained with HuNoV VP1-specific antibodies (Figs 6 and 7, S3 Fig). Viral antigens were frequently detected in the intestine, pancreas and liver. A strong signal was observed in the caudal hematopoietic tissue (CHT), which contains hematopoietic stem/progenitor cells (HSPCs) that differentiate into multiple blood lineages and by 4 dpf start to migrate to the kidney marrow and thymus [25]. This migration may explain why HuNoV was detected in every section of the larvae. HuNoV has been detected in the intestine and liver of chimpanzees [3], and has as well been reported to replicate in cells of hematopoietic lineage [26]. To determine if viral antigens could be detected at earlier time points, staining’s of larvae were performed at day 1 and 2 pi. At 1 day pi, the intestine (anterior and posterior) was already strongly positive (S4 Fig). Lighter staining was detected in the liver (S4A Fig), pancreas (S4A Fig) and caudal hematopoietic tissue (CHT) (S4G Fig). The same was observed at day 2 pi (S4C and S4E Fig).
Here we describe a robust and very convenient HuNoV replication model in zebrafish larvae. Zebrafish share about ~70% of their genes with humans and 82% of disease-related genes have at least one zebrafish orthologue [27], but there are obvious differences between zebrafish larvae and the natural host. Zebrafish larva are an optically accessible whole-organism with comparable organs and systems as higher vertebrates, thus providing a unique system to identify and study physiologically relevant features of a HuNoV infection. Zebrafish-based models have also been used to study multiple pathogenic bacteria. Such studies have contributed to a better understanding of cellular microbiology, for example by highlighting the occurrence of emergency granulopoiesis for the replenishment of neutrophils after a Salmonella infection [28] and by in vivo imaging of septin cage entrapment of Shigella flexneri leading to autophagy [29].
We here report that HuNoV of multiple genotypes replicate efficiently in zebrafish larvae, with GII.4 viruses yielding the highest fold-increase over background (˃ 3 log10). This is much higher than the 60-fold increase observed following intraperitoneal injection of HuNoV in BALB/c Rag-γ c mice [30]. Also when compared to the use of intestinal enteroids, where infection with HuNoV GII.4 yielded a 1.5–2.5 log10 increase of viral RNA [or a maximum average of 2.8 log10 after the addition of bile] [7], we here consistently detected a stronger increase of viral genomes. We here also report replication of HuNoV GII.6 in zebrafish larvae (this was not reported in intestinal enteroids [23, 31]). In addition, replication of GII.3 viruses was achieved without the need for addition of bile, although it should be noted that the yolk is a lipid-rich structure [32]. In addition, a HuNoV GI (genotype GI.7) replicated efficiently, yielding a comparable 3 log10 increase in viral RNA with kinetics comparable to that observed for GII viruses. Only HuNoV GI.1 was ever shown to replicate and this to a maximum yield of 1 log10 (with addition of bile) [23, 31]. This finding highlights a unique opportunity to study the biology of GI noroviruses.
In addition, we succeeded to passage HuNoV GII.4 from larva to larva, which was possible up to the second passage. The reason this was not possible for more passages is that a minimum volume of PBS (50 μL) is needed to homogenize the zebrafish tissue whereas only 3 nL can be injected in each larva. This dilution after each homogenization (~16,600-fold) results in a too low inoculum in the third passage (below the ~103 viral RNA copies determined as the ID50 for a GII.6 strain). This is thus the main limiting factor, although other unknown restriction factors could exist.
HuNoV replication was most prominent in the intestine of larvae one day pi, as detected by immunohistochemistry. Viral antigens were also detected in the hematopoietic tissue at this point and this evolved to very intense staining at day 3 pi. These findings clearly illustrate virus replication in both tissues thus supporting a notion of dual tropism of HuNoV. The fact that there are common features to the innate immune response triggered by the virus infection in zebrafish further adds value to the model. In addition, their optical transparency allows live imaging studies, for example using transgenic lines with tagged immune or gut cells, which would aid studies of tissue tropism and pathogenesis. The availability of simple genetic manipulation methods facilitates the understanding of gene functions [33, 34], which, combined with the availability of many knockout alleles [35], could significantly enhance our ability to dissect HuNoV-host interactions.
Zebrafish are widely available at universities/research centers and their use is amenable to high-throughput studies. A trained researcher can inject/manipulate hundreds of larvae per day requiring only a microscope, a micromanipulator and injection pump. Moreover, only a few nL of virus is required to inject zebrafish larvae. Consequently, a 100 mg stool aliquot (with a high virus titer) is sufficient to inject about 300,000 larvae, yielding 30,000+ data points. The ability to generate large and homogenous datasets is essential for large-scale efforts such as the development of therapeutics [36–38]. The zebrafish larvae fit in 96- and 384-well plates and small molecules can be simply added to the swimming water. Moreover, only a minute amount of compound is needed to assess a potential antiviral activity, this is in stark contrast to what is needed for studies in mice. Thus the model here validated for antiviral drug studies using GI and GII HuNoVs, now allows to readily assess the potential antiviral activity of novel inhibitors.
Overall, this model is a major step forward in the study of HuNoV replication and provides the first robust small laboratory animal of HuNoV infection.
All zebrafish experiments were approved and performed according to the rules and regulations of the Ethical Committee of KU Leuven (P086/2017), in compliance with the regulations of the European Union (EU) concerning the welfare of laboratory animals as declared in Directive 2010/63/EU. Zebrafish larvae were used from 48 hpf until a maximum of 144 hpf. Human stool samples, positive for human norovirus (HuNoV), were obtained from the existing collection of samples of the University Hospital of Leuven (Belgium) in an anonymous way. In a letter of consent, the patient is informed that leftover stool samples can be used for scientific research; no new samples were collected in light of this study.
Wild type AB adult zebrafish were maintained in the aquatic facility of the KU Leuven (temperature of 28 °C and 14/10 h light/dark cycle). Fertilized eggs were collected from adults placed in mating cages and kept in petri dishes containing Danieau’s solution (1.5 mM HEPES, 17.4 mM NaCl, 0.21 mM KCl, 0.12 mM MgSO4, and 0.18 mM Ca(NO3)2 and 0.6μM methylene blue) at 28 °C until the start of experiments.
Human stool samples, positive for human norovirus (HuNoV), were obtained from the University Hospital of Leuven (Belgium). An aliquot of 100 mg of each stool sample was re-suspended in 1 mL of sterile PBS, thoroughly vortexed and centrifuged (5 min, 1,000 g), supernatant was harvested and stored at -80°C. This virus suspension was used for RNA extractions, quantification by RT-qPCR, sequencing and injections in the zebrafish larvae. HuNoV RNA was extracted from 100 μl of PBS suspension using the RNeasy minikit (Qiagen, Hilden, Germany), according to the manufacturer’s protocol. UV inactivation of the sample was done by 10 min radiation under an UV lamp (UVP, UVG-54 254 nm). The virus samples used in this study where the following: HuNoV GI.P7-GI.7 (1.19 x 1012 RNA copies/g of stool), HuNoV GII.P16-GII.2 (7.67 x 1010 RNA copies/g of stool), HuNoV GII.P16-GII.3 (1.58 x 1011 RNA copies/g of stool), HuNoV GII.P4-GII.4 (5.50 x 1012 RNA copies/g of stool, MN248513), HuNoV GII.P4-GII.4 (1.01 x 1012 RNA copies/g of stool, MN248518), HuNoV GII.P17-GII.6 week 0 (1.12 x 1013 RNA copies/g of stool, MN248514), HuNoV GII.P17-GII.6 week 1 (7.33 x 1010 RNA copies/g of stool, MN248516), HuNoV GII.P17-GII.6 week 5 (8.48 x 109 RNA copies/g of stool, MN248517), HuNoV GII.P17-GII.6 week 12 (1.47 x 1011 RNA copies/g of stool, MN248515), MNV.CW3 (9.3 x 106 TCID50/mL) and MNV.CR6 (9.9 x 108 TCID50/mL).
Three dpf zebrafish larvae were anaesthetized by immersion for 2–3 minutes in Danieau’s solution containing 0.4 mg/mL tricaine (Sigma-Aldrich, Saint Louis, Missouri, stock solution 4 mg/mL in Na2HPO4, pH 7–7.5). Thereafter, the zebrafish larvae were transferred to a petri dish (92x16mm) with grooves of a mold imprint (6 rows, one side of 90° the other of 45°) in 1.5% agarose. A dissection needle was used to gently orient the zebrafish larvae so that they were lying on their dorsal side with the yolk facing upwards. Injection needles were pulled using glass capillaries (WPI, Sarasota, Florida, TW100F-4) and a Micropipette Puller fitted with a heat filament (Sutter Instruments, Novato, California). In every experiment, the injection needle was calibrated to ensure the precision of the injection volume. Microinjection was done using a M3301R Manual Micromanipulator (WPI) and a Femtojet 4i pressure microinjector (Eppendorf, Hamburg, Germany). Each zebrafish larvae was injected with 3 nL of virus (HuNoV GII.2, HuNoV GII.3, HuNoV GII.4, HuNoV GII.6, HuNoV GI.7, MNV.CW3 or MNV.CR6), while negative control zebrafish were injected with 3 nL of PBS. After injection, zebrafish larvae were transferred to 6-well plates with Danieau’s solution and further maintained in an incubator with a 14/10 h light/dark cycle at 32°C. Every day post injection, the general condition of the zebrafish larvae (e.g. posture, swimming behavior or signs of edema) was observed in order to record clinical signs of virus infection, and 10 zebrafish larvae were collected into 2 mL tubes containing 2.8 mm zirconium oxide beads (Precellys/Bertin Technologies, Montigny-le-Bretonneux, France) and stored at -80°C. To determine the 50% infectious dose (ID50), 10-fold dilutions up to 1/10,000 of a HuNoV GII.P7-GII.6 PBS suspension were used to inject zebrafish larvae, as described above. The ID50 was defined as the virus inoculum necessary to result in the detection of a significant increase of viral RNA lasting for more than one day pi in 50% of infected zebrafish larvae.
2′-C-methylcytidine (2CMC) was obtained from Carbosynth Limited, Compton, United Kingdom; a stock solution was prepared in DMSO (VWR Chemicals, Radnor, Pennsylvania). Treatment, via immersion, with 4 mM 2CMC started 1 day prior to injection with HuNoV GII.6 or HuNoV GII.4, thus in 2 dpf embryos (10 per condition, manually dechorionated using 2 fine tweezers), and was replenished every 12 h until the end of the experiment. The potential toxicity of 2CMC was evaluated beforehand and a non-toxic concentration was selected to treat injected zebrafish larvae.
Zebrafish larvae harvested in Precellys tubes were homogenized with 3 cycles of 5 sec (6300 rpm) with rest intervals of 30 sec (Precellys24, Bertin Technologies). Homogenates were cleared by centrifugation (5 min, 9,000 g) and RNA was extracted using the RNeasy minikit (Qiagen), according to the manufacturer’s protocol.
For detection of HuNoV GII or MNV RNA, a one-step RT-qPCR was performed using the iTaq Universal Probes One-Step Kit (Bio-Rad, Hercules, California), primers and probes used are in S2 Table. Cycling conditions were: reverse transcription at 50°C for 10 min, initial denaturation at 95°C for 3 min, followed by 40 cycles of amplification (95°C for 15 s, 60°C for 30 s) [Roche LightCycler 96, Roche Diagnostics, Risch-Rotkreuz, Zwitserland]. For absolute quantification, standard curves were generated using 10-fold dilutions of template DNA of known concentration.
HuNoV isolated RNA was reverse transcribed by a one-step multiplex RT-PCR using the OneStep RT-PCR Kit (Qiagen), according to the manufacturer’s protocol, primers used are in table S2. The cycling conditions were reverse transcription at 50°C for 30 min, initial denaturation at 95°C for 15 min, followed by 40 cycles of amplification (94°C for 30 s, 55°C for 30 s, 72°C for 60 s) and final extension of 10 min at 72°C. The PCR products were run on a 2% agarose gel. All positive PCR samples were purified using ExoSAP-IT (TermoFisher Scientific, Waltham, Massachusetts) and sequenced with the specific GI and GII primer sets. Viral genotypes were determined with the Norovirus Typing Tool Version 2.0 [39].
To generate the cDNA, the ImProm-II Reverse Transcription System (Promega, Madison, Wisconsin) was used. Briefly, a total of 1 μg (ca. 10 μL) of extracted RNA was added to 1 μL of random hexamers and incubated at 70°C for 5 min, followed by 5 min at 4°C. To this reaction mix a total volume of 40 μL containing 8 μl of Improm II 5X reaction buffer, 6 mM MgCl2, 0.5 mM deoxynucleoside triphosphate, 40 units of RNase inhibitor, 1 μL of Improm II reverse transcriptase, followed by an incubation at 25°C for 5 min, 37°C for 1 h, and 72°C for 15 min. A qPCR was performed with 4 μL template cDNA using the SsoAdvanced Universal SYBR green supermix, 600 nM of forward and reverse primers for ifn, mx, rsad2/viperin and the housekeeping genes β-actin and ef1a. Primers sequences were as previously described [40, 41]. Cycling conditions were: polymerase activation at 95°C for 3 min followed by 40 cycles of denaturation at 95°C for 15 s, annealing at 55°C and extension at 72°C for 30 sec (Roche LightCycler 96, Roche Diagnostics). Data was normalized to the housekeeping genes and compared to PBS-injected zebrafish larvae to determine the fold induction of the expression, according to the Livak method [42].
Human fecal samples and zebrafish larvae were prepared using the NetoVIR protocol, with minor modifications [43]. To preserve the fecal sample for the infection experiments, no virus like particle purification was performed. RNA and DNA were extracted using the QIAamp Viral RNA Mini Kit (Qiagen) according to the manufacturer’s instructions, without addition of carrier RNA. First and second strand synthesis and random PCR amplification for 17 cycles were performed using a modified Whole Transcriptome Amplification 2 (WTA2) Kit procedure (Sigma-Aldrich), allowing for amplification of both RNA and DNA [43]. PCR products were purified with MSB Spin PCRapace spin columns (Stratec, Birkenfeld, Germany) as instructed and library preparation was done using a modified Nextera XT DNA kit (Illumina, San Diego, California) protocol [43]. Libraries were quantified with the KAPA Library Quantification kit (Kapa Biosystems) and DNA size of libraries was obtained using Agilent High Sensitivity DNA Kit on a Bioanalyzer 2100 (Agilent, Santa Clara, California). Sequencing of the samples was performed on a NextSeq500 platform (Illumina) for 300 cycles (150 bp paired ends).
Raw Illumina reads were trimmed for quality and adapters using Trimmomatic (version 0.35), S3 Table. The remaining reads were de novo assembled into contigs with SPAdes assembler (version 3.9.0) using the metaspades flag [44]. Contigs were classified using DIAMOND in sensitive mode [45]. A full norovirus genome sequence was obtained for the baseline sample (week 0) which was manually checked by aligning sample reads using BWA-MEM [46]. The assembled human norovirus genome of week 0 was used as reference to align the longitudinal patient samples and the sequence obtained from zebrafish larvae 2 days pi to infer viral changes over time using BWA-MEM [46] and Tablet [47]. Trimmed reads from each sample were mapped to the baseline sample of the patient (HuNoV GII.P7-GII.6 week 0) using BWA 32 to obtain individual sample magnitudes for the analysis of the donor-derived reads. An in-house developed python script (python version 2.7.6) was used to generate a summary of annotated viral genus reads per individual sample. To assess mutations on the viral genome overtime, samples were compared to the sample of week 0 using the package deepSNV from biocLite in R. Sites with less than 10 reads coverage were excluded from the analysis. A site was considered either having a mutation, major or minor variant if respectively ≥ 80%, 50–80% or 10–49% of the reads were different from the reference sequence for a particular nt position. Nucleotide positions with < 100 reads were only included as minor variant if >20% of the reads were mutated compared to the reference.
HuNoV structural antigens were detected in infected zebrafish larvae via the RIDASCREEN Norovirus 3rd Generation (R-Biopharm, Darmstadt, Germany), according to the manufacturer’s instructions. Ten zebrafish larvae were harvested 1 h pi or 3 days pi and deyolked. The zebrafish larvae were smashed in 50 μl of ddH2O with a pestle in a micro centrifuge tube (VWR, Leuven, Belgium) and debris was removed by centrifugation (10 min, 9,000 g). The supernatant was diluted to a volume of 100 μl. In each EIA run, the positive and negative controls of the kit were included for assay validation and cutoff calculation. The optical density (OD) was measured at 450 nm (Spark, Tecan, Männedorf, Switzerland).
Mock and HuNoV injected zebrafish larvae were deyolked at 3 days pi, then lysed with a pestle in RIPA buffer (Thermo Scientific, Waltham, Massachusetts) in the presence protease inhibitor (Merck KGaA, Darmstadt, Germany) and centrifuged at 9,000 g for 5 min at 4°C. Protein concentration of the lysates were determined using the BCA protein assay kit (Thermo Scientific). BHK cells (purchased from ATCC) transfected with GII.4 construct were used as positive controls. Thirty μg of either mock or HuNoV injected zebrafish larvae lysates were analyzed by SDS-PAGE and western blotting. Baby hamster kidney cells expressing T7 polymerase (BSR-T7 cells, received from Klaus Conzelman, Ludwig-Maximilians-Universitat München, Germany) were used to analyze viral protein expressions of HuNoV GII.4. Briefly, BSR-T7 cells infected with poxviruses expressing T7 RNA polymerase at an MOI (based on the virus titer in chick embryo fibroblasts) of 0.5–1.0 PFU per cell, were subsequently transfected with 1 μg construct containing the full length clone of HuNoV GII.4 using Lipofectamine 2000 according to the manufacturer’s instructions (Invitrogen, Carlsbad, California). To analyze viral protein expression, cells were harvested 24 h post-transfection for western blot analysis. Proteins were separated in 12.5% or 17.5% SDS-PAGE and transferred onto a 0.45 μm nitrocellulose membrane (GVS North America, Sanford, Maine). Membranes were then blocked with 5% milk/PBS-T, washed and incubated overnight with primary antibody against viral proteins NS3 or VPg, which were kindly provided by Professor Ian Goodfellow (University of Cambridge). Membranes were washed extensively, incubated with species-specific secondary antibodies (Li-cor, Lincoln, Nebraska) and viral proteins were detected using Li-cor Odessey CLx imaging system. As housekeeping gene, β-actin (Proteintech, 60008-I-Ig, Uden, The Netherlands) was used.
Four and five dpf zebrafish larvae were immersed in a 1 mL HuNoV GII.P7-GII.6 suspension diluted in Danieau’s (containing ~ 1011 viral RNA copies). After 6 h of exposure to the virus, the larvae were washed with clean Danieau’s, and transferred to a well containing new media and further maintained in an incubator with a 14/10 h light/dark cycle at 32°C. Every day pi, the general condition of the zebrafish larvae (e.g. posture, swimming behavior or signs of edema) was observed in order to record clinical signs of virus infection, and 10 zebrafish larvae were collected into 2 mL tubes containing 2.8 mm zirconium oxide beads (Precellys/Bertin Technologies) and stored at -80°C.
Twenty zebrafish larvae were injected with HuNoV GII.P4-GII.4 as described above and were harvested at the peak of replication (2 or 3 days pi). The zebrafish larvae were smashed in 50 μl of sterile PBS with a pestle in a micro centrifuge tube (VWR, Leuven, Belgium) and debris was removed by centrifugation (3 min, 8,000 g). The supernatant was collected and injected in zebrafish larvae at 3 dpf as described before.
HuNoV-infected zebrafish larvae were harvested at 1, 2 or 3 days pi and fixed in 4% formaldehyde overnight at 4°C. The following day the formaldehyde was replaced by 70% ethanol and the zebrafish larvae were embedded in an agarose mold, then processed in paraffin, sectioned and stained as previously described [48]. Immunohistochemistry was performed using antibody TV20 at 1/1000 dilution (kindly provided by Dr. Peter Sander, R-Biopharm) and Anti-VP1 (ab92976, Abcam), 1/500 dilution. Additional staining’s were performed with hematoxylin–eosin (H&E). Microscopy was performed using a Carl Zeiss Axio Imager Z1 microscope at 10, 40 and 100x magnifications and images were captured and processed with the AxioVision 4.8.2.0 software (Zeiss).
Data was analyzed using GraphPad Prism 7 (Graph-Pad Software) and p values were determined with the nonparametric Mann-Whitney test, where ****p<0.0001, *** p <0.001, ** p <0.01, * p<0.05, and ns is p≥0.05.
|
10.1371/journal.pcbi.1000676 | Adaptable Functionality of Transcriptional Feedback in Bacterial Two-Component Systems | A widespread mechanism of bacterial signaling occurs through two-component systems, comprised of a sensor histidine kinase (SHK) and a transcriptional response regulator (RR). The SHK activates RR by phosphorylation. The most common two-component system structure involves expression from a single operon, the transcription of which is activated by its own phosphorylated RR. The role of this feedback is poorly understood, but it has been associated with an overshooting kinetic response and with fast recovery of previous interrupted signaling events in different systems. Mathematical models show that overshoot is only attainable with negative feedback that also improves response time. Our models also predict that fast recovery of previous interrupted signaling depends on high accumulation of SHK and RR, which is more likely in a positive feedback regime. We use Monte Carlo sampling of the parameter space to explore the range of attainable model behaviors. The model predicts that the effective feedback sign can change from negative to positive depending on the signal level. Variations in two-component system architectures and parameters may therefore have evolved to optimize responses in different bacterial lifestyles. We propose a conceptual model where low signal conditions result in a responsive system with effectively negative feedback while high signal conditions with positive feedback favor persistence of system output.
| Bacteria have evolved various mechanisms for surviving unpredictable changes and stresses in the environment, such as nutrient limitation. One common survival mechanism is the two-component system, where a sensor protein responds to a particular type of stress by activating a regulator in the cell. These regulators can in turn activate genes that produce proteins for stress-appropriate responses. The activated regulator often positively regulates transcription of its own operon containing the sensor and regulator genes leading to a feedback loop. This is interesting, because positive feedback is usually associated with a slower response time than negative feedback and therefore negative feedback would often be selected for by evolution. Here we analyze a mathematical model to study the interplay of this feedback and postranslational mechanisms regulating two-component system signaling. We found that modulation of regulator activity by its operon partner can lead to overall negative feedback to result from autoactivation. This happens if (1) the sensor can both activate and deactivate the regulator, and (2) there is some reaction resulting in regulator activation independently of its cognate sensor. As a result our model predicts that two-component systems may be capable of flexibly switching between positive and negative feedback depending on different circumstances, allowing for appropriate responses in a variety of conditions.
| Unpredictably changing environments necessitate appropriate responses for successful survival by bacteria. Bacterial two-component system (TCS) signaling shifts transcriptional programs in response to a variety of external cues affecting bacterial growth such as nutrient availability, osmolarity, redox state, temperature, and concentrations of other important extracellular molecules [1]. The basic TCS core structure includes a sensor histidine kinase (SHK) and response regulator (RR). SHK-modulated phosphorylation of RR results in activation that frequently induces a transcriptional program.
Environmental signals are routed through conformational changes in SHK homodimers, which may participate in up to three biochemical processes. First, each subunit hydrolyzes ATP to trans-phosphorylate a His residue in the other subunit. Second, the phosphorylated SHK subunit transfers its phosphate to an Asp residue on unphosphorylated RR bound to the SHK. Third, in bifunctional TCSs, unphosphorylated SHK can catalyze dephosphorylation of RRP with release of inorganic phosphate. Modulation of phosphatase and/or kinase activities of SHK may therefore induce system responses by shifting the dynamic equilibrium between active (phosphorylated) and inactive (unphosphorylated) forms of RR. Interactions between the RR and exogenous phosphodonors (e.g. cross-talk with non-cognate TCSs or small molecule phosphodonors) may also contribute to TCS activation.
Activated RR may induce expression of multiple operons. This regulon often contains the operon encoding the RR and SHK. Such autoregulation (Figure 1A) is observed in PhoPQ [2], PhoBR [3], VanRS [4], CpxRA [5], CusRS [6], and many other model systems [7]. The physiological significance of this feedback loop is not completely understood.
Previous genomic studies in E. coli have shown widespread positive and negative transcriptional autoregulation [8]–[10]. These studies indicate several E. coli TCSs as examples of positive feedback loops. However, we suggest that the effective sign of feedback in TCSs may depend on the biochemical interactions of the autoregulated proteins. Induction of bifunctional SHK in the same operon as RR may affect the phosphorylation equilibrium, and therefore have either positive or negative effects on the amount of transcriptionally active RR. That is, the signal increases RR phosphorylation, but resultant increases in gene expression may in turn positively or negatively change the amount of phosphorylated RR amplifying or attenuating the original signal. The resulting sign of feedback can be related to the transient dynamics of TCS activation: overshoot kinetics often result from underdamped negative feedback [11], and have been observed in one TCS [12]. The attainment of such overshoot has important implications on the kinetics: it transiently speeds expression of genes in the regulon (e.g. [13] and below) and is necessary for virulence of Salmonella enterica serovar Typhimurium (hereafter, Salmonella) in mice [12].
We used a mathematical model of a generic TCS to demonstrate that post-translational kinetics of SHK-RR interactions can determine the effective sign of feedback. The model explains disparate results relating to transcriptional autoregulation in TCSs including the “learning” effect [3] and feedback-induced surge [12]. Moreover, we show that some systems may display both effectively negative and positive feedback at different signaling levels. The effective feedback sign is determined by kinetic parameters of TCSs, with positive and negative feedback allowing distinct functional advantages in different circumstances. Therefore, differences in post-transcriptional kinetics may have arisen from selective pressure for feedback based on bacterial lifestyle. The accessibility of either type of feedback in the same system also raises the possibility of tuning between “responsive” and “persistent” signaling modes in a single TCS.
To determine the role of transcriptional feedback in TCSs, we constructed an ordinary differential equation model extending previous work [14],[15]. The resulting model is schematically represented in Figure 1A; reaction mechanisms are presented in Figure 1B. The full set of reactions is listed in Table 1.
A signal may modulate the rate of kinase (kap in Reaction 10) or phosphatase (kph in Reaction 18) activity of SHKs [16]. We usually take the kinase activity to be constitutive as with Salmonella PhoPQ [17], but demonstrate the generality of our results to either type of modulation below.
A critical component of this model is the existence of an SHK-independent flux of RR phosphorylation and dephosphorylation, arising from small molecule phosphodonors, autodephosphorylation, or crosstalk with other TCSs [1], [18]–[20]. We assume a Michaelis-Menten form for these fluxes that is biologically consistent with the crosstalk mechanism: (Reaction 19); (Reaction 20). The possibility of small molecule phosphodonors as the source may result in linear phosphorylation/dephosphorylation kinetics, resulting in qualitatively similar results: ;. For brevity we present only the results using the Michaelis-Menten form.
To study TCS induction dynamics with the model, we chose a Monte Carlo parameter sampling approach because no general analytical solution of transient response is possible. Signal level was determined by parameter kph, held at 10/s for the resting steady state and changed to 0.1/s to activate at t = 0 min. The effective sign of feedback in the model is measured by open-loop gain at the activated steady state. That is, we take a no-feedback, or open loop, form of the model in which the operon is activated by putative exogenous activator R0. The gain measures how changes in R0 concentration affect the concentration of transcriptionally active response regulator RRP2 (at a concentration of R0 equal to the activated steady-state RRP2 concentration). Intuitively, the feedback increases total RR and SHK concentrations. However, increases in these concentrations may have a positive or negative effect on the phosphorylated fraction of RR. These conditions respectively correspond to effectively positive and negative feedback. The open-loop gain is notably different from the steady-state signal-response gain of the system. In all cases considered, system response (level of activated RR) increased with increased signal (decreased kph). Nevertheless, the sign of open-loop gain can be either positive and negative for the cases in which transcriptional feedback respectively amplifies or attenuates RRP concentration.
Parameter sampling returns both negative and positive open-loop gains corresponding to overall negative and positive sign of feedback (Figure 2A). In each case with a negative loop, some fraction of RR phosphorylation results from non-cognate sources (i.e., JE/(JE+JS)>10−3 where JE is flux through Reaction 19 and JS is flux through reaction 16 in Table 1). An exogenous RR phosphorylation flux was the only identified mechanism to produce a negative feedback loop in a more generalized TCS model as well (Text S1). More extensive sampling of a simplified model confirms these results (Text S2; Figure S3).
A common dynamical characteristic of negative feedback is overshoot kinetics, which have been shown to occur in PhoPQ of Salmonella [12]. A subset of randomly generated parameters results in the prediction of feedback-induced overshoot in concentrations of RRP and mRNA of genes under its control (Figure 2A, red circles). Notably, the effective sign of feedback is negative for all of them. Figure 2B shows a sample time-course selected to resemble Salmonella PhoPQ compared with experimental results of downstream promoter binding from [12]. We used a genetic algorithm to select parameter sets for systems resembling PhoPQ; one of the representative sets is used as a default example throughout the text (Table S1; Figure 2B). Briefly, we selected for significant feedback-modulated overshoot in RRP and significant increases in both total RR and transcriptionally active RRP2 at the activated steady state. Figure S1 shows further dynamical characteristics of the reference system.
In simple inducible genetic systems with autoregulation, negative feedback is associated with faster attainment of a given steady state as compared to positive feedback [21],[22]. In the TCS model, we compared the response (here, downstream (DS) protein accumulation) between a feedback-regulated model and an open-loop model where constitutive gene expression in the open-loop form equals expression at the activated steady state of the closed loop model (Figure 3A–B). The results are consistent with negative open-loop gain speeding responses and positive open-loop gain slowing responses compared to the case with no feedback going to the same steady state. The advantage of negative open-loop gain is especially pronounced with feedback-induced overshoot. This result confirms that the effective feedback sign as measured by open-loop gain corresponds to the directly observable feedback sign in simpler systems. Text S3 and Figure S4 discuss the small number of cases in Figure 2B that appear to contradict this rule.
Another kinetic effect of feedback in TCSs is “learning” exhibited in E. coli PhoBR, where responses to a second stimulus after a transient signal interruption are faster than the response to the first signal [3]. To explore the “learning” effect, we determined the response time of downstream protein accumulation after a 45 minute signal interruption (Figure 3C–D) for sampled parameter sets and compared this response with the one following the initial signal. The length of the signal interregnum was chosen to reflect prior experiments [3]. The 45 minute interregnum, longer than the timescale of dephosphorylation (seconds to minutes), but shorter than the timescale for protein dilution to near basal levels (hours), is biologically reasonable (Figure 3C simulates this in the TCS model). The results predict that the “learning” effect can be observed in both positive and negative feedback, but is more pronounced for positive feedback. Improvements in downstream response depend on accumulation of SHK and RR protein concentrations during the first activation (Figure 3D; Spearman's rank correlation ∼0.625; p∼0 with double machine precision). High induction capacity attainable by most positive feedback cases amplifies the “learning” effect; it does not arise from bistability or slow response times usually associated with positive feedback [21],[23]. In fact, the phosphorylated fraction of RR responds quickly to changes in signaling; it is the accumulation in protein level that causes this effect (Figure 3C). Therefore, our results correlate TCSs functioning in positive feedback mode with faster recovery following transient signal fluctuations.
The sign of feedback is a function of parameters in the model presented here; it is therefore conceivable that specific parameter perturbations could change the strength or sign of feedback. We found a large number of cases in which effective feedback signs change as a function of input signal strength (parameter kph). The majority of Monte Carlo parameter samples that predict negative feedback for one signaling level exhibit positive feedback for a different signal level (Figure 4A).
If wild-type TCSs are indeed capable of changing feedback sign, the dynamic performance of the system may be capable of meeting different functional criteria depending on the signal level. To determine if feedback sign changes performance in TCSs, we compared, at various signal levels (kph), open-loop gain and response times of downstream protein induction from the basal state (kph = 10/s) using the default parameter set (Figure 4B–C). Response times correspond directly to the effective sign of feedback at each activation level: negative feedback imparts faster response. Furthermore, peak overshoot time approaches infinity as the signal level approaches the positive feedback region (Figure 4C dashed line). Therefore the TCS model predicts that adapting the feedback sign based on transient signal intensity permits TCSs to meet different functional criteria depending on the context.
Unmasking of positive feedback for the high signal region of the dose-response curve may explain the strong effect of feedback in E. coli TCS PhoPQ that only occurs for extreme signaling conditions [24]. We explore other aspects of the E. coli PhoPQ dose response in Text S4 and Figure S5.
TCSs may be subject to modulation of SHK kinase activity (parameter kap in this model) or both kinase and phosphatase activity (kap and kph). We scanned the 2-dimensional kap × kph signal space and found that modulation of either activity may result in feedback tuning (Figure 4D).
Autoregulation by RRP proportionally affects concentrations of both RR and SHK in the genetic architecture assumed here – both genes co-expressed from a single operon (Figure 1). Is there a functional rationale for why feedback affects both genes in many TCSs, as opposed to autoregulation of only RR or SHK? To address this question, we formulated models where expression of one of the TCS proteins, either RR or SHK, is constitutive (outside of the TCS regulon) whereas the other is regulated by RRP (Circuits I and II, Figure 5). The constitutive rate of production is set equal to the production rate in the wild-type system with the signal parameter kph = 0.1/s. This level was chosen to represent a production rate that avoids saturation effects of excessively high or low expression.
When RRP only regulates the RR gene (Circuit I), the response time is typically slow (Figure 5A). When RRP only regulates SHK (Circuit II) the induction range (defined as the difference between high and low signal limits, normalized to the case without feedback) is typically smaller (Figure 5B). This suggests that the wild-type case, with both RR and SHK feedback-regulated from a single operon, exhibits a trade-off, balancing fast response and high induction range. At higher signal levels, the wild-type system exhibits a synergistic effect where the linked genes can attain higher signal output than with feedback to RR alone (Figure 5B).
TCSs are responsible for diverse, specialized, large-scale reprogramming of bacterial transcription in response to many possible signals. Nevertheless, while orthologous TCSs in different species often have drastically different regulons [25], core TCS architectures are remarkably similar. Some TCSs are autoregulated, where transcriptionally active RR induces expression of the TCS operon. This feedback may in itself have diverse effects on the system, including overshoot kinetics [12] and “learning” to respond faster after a previous stimulus [3]. In the interest of generality we focused on a commonly occurring TCS architecture, with linked genes that are autoregulated. Others not considered here may have important functional consequences (e.g. bistability emerging from autoregulation in the B. subtilis DegS-DegU system [26]).
Despite the fact that autoregulation is almost uniformly positive in TCSs (with the exception of TorRS in E. coli [27]), we have shown that the effective feedback sign may be either positive or negative, depending on biochemical characteristics of the system. Why is attainment of different feedback signs physiologically relevant? In most biological systems, negative feedback often reduces noise and speeds responses in well-controlled comparisons [21],[28],[29] while positive feedback leads to phenotypic heterogeneity and bistability [23],[26],[30],[31].
Attainment of negative feedback is not deducible from examining network diagrams such as those in Figure 1, which may bias the observer into assuming that the feedback is necessarily positive. It is ultimately related to bifunctionality of the SHK enzyme that can both increase and decrease the fraction of activated RR. Our results demonstrate a mechanism for creating a negative feedback loop that depends on the existence of a pathway for RR phosphorylation independent of cognate SHK activity. Intuitively, when SHK is the sole source of RR phosphorylation and dephosphorylation the system output is robust and insensitive to SHK concentration [14],[32]. However, when an additional feedback-independent flux of RR phosphorylation exists, upregulation of SHK can disproportionately increase the phosphatase flux resulting in negative feedback. Other mechanisms of attaining negative feedback may exist, but we have failed to identify them (cf. Text S1, Table S2 and Figure S2).
If exogenous phosphorylation is an important mechanism in TCSs, why is it not frequently detected? One proposal is that existence of the phosphatase activity of SHKs results in buffering, or suppression of exogenous phosphorylation and thereby reduces or eliminates crosstalk with other TCSs [33],[34]. As a result, an exogenous phosphorylation flux may not lead to a large effect on the levels of activated regulator and therefore existence of cross-talk may be difficult to detect in wild-type systems [18]. Our modeling predictions support this conclusion: phosphorylated RR can be kept at a low level by phosphatase activity while still maintaining effectively negative feedback enabled by exogenous phosphorylation.
An alternative model for feedback-induced overshoot that does not directly invoke negative feedback is a mechanism for dynamic regulation of SHK kinase or phosphatase activity. We considered two mechanisms: ATP-SHK interactions that alter SHK activities, and a temporal delay in SHK maturation such that the intermediate species has kinase but not phosphatase activity (Text S1). We were unable to produce overshoot kinetics with the ATP-SHK model. The SHK maturation model can produce overshoot, but every case we found requires unrealistic component concentrations and invokes a strong assumption without experimental evidence.
The best characterized system that may be attaining negative feedback is Salmonella PhoPQ. The evidence is compelling: it displays overshoot kinetics [12] and increasing SHK expression from an inducible plasmid can reduce expression of a PhoP-regulated gene [2]. However, deleting functional SHK by insertion of a phage-derived element sometimes results in a low level of RR activity [2],[35], in contrast to what would be predicted by the exogenous phosphorylation model in the absence of other differences between the wild-type and mutant.
A likely explanation is that the SHK insertion somewhat destabilizes the transcript in the mutant. In some genes, insertion of stop codons has this effect [36]. Several TCSs also undergo transcript processing with much greater stability of the monocistronic RR mRNA than the polycistronic or SHK-only mRNAs in E. coli [37]. If the insertion disrupts mRNA processing, the resulting polycistronic mRNA may have a shorter half-life with resultant lower rate of protein synthesis. The default parameter set for our model predicts that ∼2.5-fold increase in mRNA degradation rate is sufficient for basal expression (Figure S6). Other PhoQ-disabling insertions may not have the same effect on mRNA stability, resulting in overexpression of PhoP in these cases. Indeed, PhoP overexpression resulting from different PhoQ mutagenesis experiments has been observed [38].
Our predictions are consistent with experiments showing overshoot kinetics, which are only attainable with negative feedback (Figure 2). Furthermore, there are multiple established mechanisms for exogenous phosphorylation of RRs [16]. Our model offers an explanation for the attainment of diverse experimental results relating to feedback in TCSs, including overshoot kinetics [12], “learning” effects [3], and feedback effects that only occur for extreme signaling conditions [24]. However, more direct experimental tests are still needed to determine if negative feedback commonly emerges as a characteristic of TCSs in nature.
In some cases, the effective feedback sign in TCSs depends on signal level, and the dynamic characteristics of the system follow the reversal of that sign (Figure 4), suggesting on-the-fly reversal of feedback sign as an adaptive signaling mechanism. Tuning the feedback sign in this way allows a fast response to the initial signal with negative feedback for rapid induction of the new transcriptional program. When the signal is high and persistent, the feedback sign is switched to positive, filtering out transient signaling interruptions and increasing the attainable range of signaling.
We also found some cases where the effective sign of feedback reverses between negative and positive more than once (Figure 4A), being positive at low signal then negative at intermediate signals and eventually positive again at very high signals. A physiologically relevant function for positive feedback at very low signal levels is to create a signal threshold, below which the response is slow, and above which the system responds rapidly to a decisive signal.
The functional consequences of transcriptional feedback are surprisingly flexible for a system with only two interacting proteins. Previous models suggest that TCSs without the crosstalk effect are robust, or insensitive to variations of protein concentrations [14],[32]. The flexibility to tune feedback depending on the signal level appears to be a sacrifice to robustness. Explicit determination of robustness is beyond the scope of this work, but sensitivity to perturbations of one parameter does not necessarily imply that other aspects of robustness are lost. Further, possible evolutionary advantages of flexibility are clear: feedback to both TCS genes in this model enables characteristics of negative and positive feedback that would not be attainable with transcriptional feedback to RR or SHK alone (Figure 5).
Tuning of the feedback sign is reminiscent of other results showing a diverse response without explicitly rewiring the network. Some systems have been shown to transition between graded monostable and discrete bistable steady state dose responses [23],[39],[40]. On evolutionary timescales, evolvable motifs may be capable of adapting to many different functions without disrupting the network architecture [41]. TCSs may be similarly adaptable, but on a short biochemical timescale.
We propose a model for TCSs whereby transcriptional feedback shows diverse physiologically relevant effects, including negative feedback that gives rise to fast overshooting responses and positive feedback that better filters transient signal interruptions.
In order to determine the effective feedback sign in vivo, we suggest that a direct test is necessary. A conceptually straightforward way to test the feedback sign is to synthetically engineer an open-loop system under an inducible promoter and find the inducer level for which RR and SHK concentration match their wild-type values. Exploring changes in downstream transcriptional activity as a function of inducer concentration would allow direct determination of the feedback sign. (This was done in E. coli PhoPQ [24], which shows characteristics of effectively zero feedback at small signals and positive feedback at large signals; c.f. Text S4). A similar experimental set-up with just one protein on an inducible promoter can be used to test feedback synergy and trade-off predictions (Figure 5). Alternatively, TCS point mutations may alter transcriptional or translational efficiency (Reactions 1–2 or 6–7 in Table 1). With a reduction in gene expression efficiency, steady state RRP concentrations will diverge depending on feedback sign (Figure S7 shows predicted effects of such an experiment). Predicted quantitative effects of this method are more pronounced with positive than negative feedback. Thus, the low gains in the negative feedback regime may make steady state effects of the feedback difficult to detect. Methods similar to those used in previous studies exploring crosstalk between TCSs [18] may be useful to determine if Salmonella PhoPQ is subject to exogenous phosphorylation.
Many biological networks represent a balance of stimulatory and inhibitory effects. Here we have shown that this balance leads to flexibility and diversity in the functional role of the transcriptional feedback loop. This should guide toward a deeper understanding of how interactions in biological networks may have evolved to allow successful responses to a wide array of conditions.
We constructed a mathematical model extending previous TCS models [14],[15]. As outlined in the Results section, this model includes reactions for production/degradation, regulation of expression from TCS and downstream genes, and phosphorylation/dephosphorylation of RR by an exogenous source. Included reactions are summarized in Table 1.
Models were generated using BioNetGen 2.0.46 [42], and imported for analysis in Mathematica 6.0 with MathSBML 2.7.1-07-Dec-2007 [43]. Monte Carlo parameter sampling was done for parameters varied with log-uniform distributions, having intervals constrained as described in Table S1. Randomly generated parameter sets were tested for physical realizability and consistency with known characteristics of TCSs using the following criteria:
Response times were calculated as elevation time for attaining either 50% (τ50) or 95% (τ95) of the activated steady state. The former is reported for responses where post-transcriptional kinetics dominate a response while the latter is reported when transcriptional induction dominated response times. In practice, qualitative differences between these measures are minimal.
Two sample parameter sets were selected to have similar open-loop induction kinetics for the purposes of illustrating differences in responses between positive and negative feedback (Figure 3). Negative feedback example parameters: kap = 0.1706, kad = 9.786, kpt = 0.2028, ktp = 0.1368, kb = 2.314, kd = 0.9237, kb1 = 6.261, kd1 = 0.001825, kRRPdm = 9.794, kRRPmd = 0.1411, ktxn = 2.115×10−5, kSKtsn = 0.04708, tsn mult = 5.616, ktxnbasal = 2.245×10−6, KmDS = 0.01224, Km = 0.004298, kmRNAdeg = 0.001383, kexp = 0.02668, Kmexp = 0.1361, kexd = 4.218×10−5, Kmexd = 1.388. Positive feedback example parameters: kap = 8.408, kad = 0.004832, kpt = 1.392, ktp = 0.05183, kb = 0.07871, kd = 0.001172, kb1 = 2.395, kd1 = 0.002223, kRRPdm = 2.665, kRRPmd = 7.421, ktxn = 0.0006604, kSKtsn = 0.005240, tsn mult = 6.190, ktxnbasal = 3.413×10−5, KmDS = 0.06276, Km = 0.003958, kmRNAdeg = 0.007446, kexp = 3.452×10−6, Kmexp = 0.0004101, kexd = 5.017×10−6, Kmexd = 0.0001661.
We used Monte Carlo sampling of parameter values to numerically determine the range of realizable dynamic behaviors in the autoregulated TCSs focusing on the sign of autoregulation and possibility of overshoot behavior. Each parameter set was used to simulate activation kinetics after a signal at t = 0 s by changing phosphatase activity (kph) from a high, resting state (10 s−1) to a low level (1 s−1). To determine the effect of feedback, we used an open-loop version of the model in which an exogenous regulator R0 rather than (RRP)2 controls SK/RRP production. The effective sign of feedback is determined by the sign of the gain defined as the derivative . If this gain is positive () the transcriptional feedback is positive; if the gain is negative (), the feedback is negative. A subset of cases predict RRP2 overshoot without mRNA overshoot. This phenomenon is not dependent on transcriptional feedback and is not of interest here. Therefore we do not analyze these parameter sets.
Induction range (Figure 5) is defined as the difference between RRP2 at high and low signal limits, normalized to the case without feedback: .
To determine the range of behaviors attainable by the model, we evolved examples conforming to specific criteria using a simple genetic algorithm. Using a set of seed parameter sets, the algorithm randomly perturbs parameters and selects sets with the highest value for a fitness function. This function depends on the desired criteria.
To select a parameter set that conforms to known important characteristics of Salmonella PhoPQ after a signal, the desired characteristics include at least 2-fold induction of (1) [RRP2] over the resting level (f1); (2) RRtot over the resting level (f2); and (3) peak RRP2 concentration over activated steady state RRP2 (f3). The following fitness function selects for these criteria, with each criterion noted: where .
|
10.1371/journal.pgen.1001221 | Essential Functions of the Histone Demethylase Lid | Drosophila Little imaginal discs (Lid) is a recently described member of the JmjC domain class of histone demethylases that specifically targets trimethylated histone H3 lysine 4 (H3K4me3). To understand its biological function, we have utilized a series of Lid deletions and point mutations to assess the role that each domain plays in histone demethylation, in animal viability, and in cell growth mediated by the transcription factor dMyc. Strikingly, we find that lid mutants are rescued to adulthood by either wildtype or enzymatically inactive Lid expressed under the control of its endogenous promoter, demonstrating that Lid's demethylase activity is not essential for development. In contrast, ubiquitous expression of UAS-Lid transgenes lacking its JmjN, C-terminal PHD domain, and C5HC2 zinc finger were unable to rescue lid homozygous mutants, indicating that these domains carry out Lid's essential developmental functions. Although Lid-dependent demethylase activity is not essential, dynamic removal of H3K4me3 may still be an important component of development, as we have observed a genetic interaction between lid and another H3K4me3 demethylase, dKDM2. We also show that Lid's essential C-terminal PHD finger binds specifically to di- and trimethylated H3K4 and that this activity is required for Lid to function in dMyc-induced cell growth. Taken together, our findings highlight the importance of Lid function in the regulated removal and recognition of H3K4me3 during development.
| Correct spatial and temporal control of gene expression is essential for development. One of the many ways that gene expression is regulated is by the addition, recognition, and removal of methyl groups from the histone proteins around which DNA is wrapped within the nucleus. Here we describe a systematic analysis of Little imaginal discs (Lid), a protein that regulates transcription via a number of different mechanisms that involve regulated removal and recognition of histone methylation. We show that while Lid's histone demethylase activity is not essential for development, numerous other conserved domains of this protein are. Furthermore, we find a genetic interaction between lid and another histone demethylase, dKDM2, that suggests this enzyme can compensate for the loss of Lid's enzymatic activity. These findings have significance for our insight into how gene expression is normally regulated and have implications for our understanding of how this goes awry during disease progression.
| The Drosophila lid gene is essential for development and encodes a protein with multiple domains implicated in chromatin-mediated regulation of transcription, including the recently described lysine demethylase domain, Jumonji C (JmjC). Six lysine residues of histones H3 and H4 can be mono, di or trimethylated, and each modification is found in a stereotypical pattern with respect to the coding region of a gene and correlates with a different transcriptional outcome [1]–[4]. As a general rule, methylation of H3K4, K36 or K79 is found at active genes whereas H3K9, K27 and H4K20 methylation is associated with those that are repressed. We and others have shown that overexpression of Lid reduces H3K4me3 levels and that this chromatin mark is elevated in lid mutants, establishing Lid as a JmjC domain-dependent H3K4me3 demethylase [5]–[7]. The four conserved mammalian orthologs of Lid, KDM5a-d, also demethylate H3K4me3 although these proteins show broader substrate specificity than their Drosophila counterpart, also targeting H3K4me2 [8]–[10].
While there is limited data regarding the biological role of the KDM5 family of proteins in mammals, the findings that KDM5b is overexpressed in breast, bladder and prostate cancers [11]–[13] and that mutations in KDM5c are found in patients with X-linked mental retardation [14] suggest that they play important developmental roles. However, a confounding factor to the analysis of the four mammalian KDM5 paralogs is their functional redundancy, as the mouse KDM5a knock out is viable, fertile and displays no change in global H3K4me2/3 levels. In contrast, Lid is the sole KDM5 protein in Drosophila and it is essential for viability [15], providing an ideal system in which to investigate the function of this family of proteins.
Although the KDM5 family of proteins are named based on the function of their catalytic JmjC domain, metazoan KDM5 proteins have several other conserved motifs: a JmjN domain of unknown function that is present in a subset of JmjC proteins, an ARID (A/T rich interaction domain [16]) implicated in binding both A/T and G/C rich DNA sequences [17], [18], a single C5HC2 zinc finger, and two or three PHD fingers (plant homeobox domain [19]) involved in mediating protein-protein interactions [20]. Importantly, while the JmjC domain-dependent demethylase function is well defined in vitro for KDM5 family proteins, the in vivo relevance of this and other domains remains unclear.
We have previously shown that Lid is rate-limiting for cell growth induced by the Drosophila homolog of the c-Myc oncoprotein, dMyc [7]. Specifically, Lid binds directly to dMyc and is required for dMyc-dependent activation of one of its growth regulatory target genes, Nop60B. While we have demonstrated that this occurs independently of Lid's lysine demethylase activity, the molecular mechanism by which Lid functions in Myc-mediated growth is yet to be determined. Here we present an investigation of the function of Lid's domains and demonstrate that its demethylase activity is dispensable for development, however its JmjN, PHD3 and C5HC2 domains are all essential. While our observation that Lid's demethylase activity is not essential suggests that regulated removal of H3K4me3 serves primarily to modulate gene expression levels, a genetic interaction between lid and the JmjC domain-containing protein dKDM2 is consistent with these two demethylases acting redundantly on H3K4me3. We also show that the essential C-terminal PHD finger of Lid binds di- and trimethylated H3K4 and that this domain is required for lid to genetically interact with dMyc. Based on these data, we propose that Lid-dependent recognition of H3K4me2/3 facilitates dMyc binding to promoters rich in this active chromatin mark.
To assess the contribution of each individual domain of Lid to its demethylase activity and animal development, we generated a series of deletions and point mutations that disrupt each domain of Lid to complement our previously characterized demethylase inactive version of Lid (Lid-JmjC*) that harbors two point mutations in the JmjC domain and prevents Fe2+ binding (H637 and E639 to Alanine) [7]. To enable these analyses, flies carrying UAS transgenes that specifically delete Lid's JmjN, ARID, C5HC2 zinc finger and three PHD fingers were generated to allow conditional Gal4-mediated expression in vivo (see Materials and Methods for details).
To assess the ability of our Lid mutants to demethylate, we generated clones of cells overexpressing each protein in larval fat body and examined the levels of Lid and H3K4me3 (Figure 1). Based on the intensity of the immunofluorescence signal and Western analysis, all of our transgenes expressed Lid at similar levels (data not shown; Figure 1), with the exception of LidΔPHD2, for which we were unable to detect Lid overexpression even after combining multiple transgenes (data not shown). To examine the role of Lid's second PHD finger, we created a point mutant in the first cysteine of this C4HC3 zinc finger (UAS-LidC1296A) and found that overexpression could be detected after combining two transgenes (Figure 1K). Mutating the second or deleting the third PHD domain of Lid did not affect its ability to demethylate H3K4me3 (Figure 1K–1N). In contrast, Lid's JmjN, PHD1 or C5HC2 domains were essential for enzymatic activity as overexpression of these deletion mutants resulted in no change in global levels of H3K4me3 (Figure 1D, 1H, 1J). While the role of Lid's PHD1 and C5HC2 domains in demethylation remains to be investigated, our finding that Lid's JmjN domain is required for demethylase activity is not surprising based on structural analysis of the demethylase KDM4a which shows its JmjN domain making extensive contacts within the catalytic core of its immediately adjacent JmjC domain [21]. Unlike other deletions that prevented Lid's enzymatic function, expression of LidΔARID resulted in a variable increase in H3K4me3 levels, indicating that this mutant protein can behave as a dominant negative in fat body cells (Figure 1E, 1F). We do not yet understand the mechanism by which LidΔARID increases H3K4me3 levels, but have observed a similar effect upon overexpression of Lid-JmjC* [7]. The ARID of KDM5a, b and Lid are required for demethylase activity in transient transfection assays, however a dominant interfering effect has not been reported [17], [22]. Our finding that deletion of Lid's ARID can increase H3K4 trimethylation raises the possibility that in addition or concomitant with its ability to bind DNA, this domain may cooperate with Lid's JmjC domain.
To determine the importance of Lid's conserved domains in vivo, we ubiquitously expressed our UAS-Lid transgenes in animals lacking endogenous zygotic lid expression. Because Lid is normally expressed ubiquitously throughout development [7](data not shown), we expressed our UAS-Lid (and Lid mutant) transgenes at low uniform levels in lid10424 homozygous mutants using actin-Gal4 (Figure 2E). This approximately two-fold overexpression of Lid is not sufficient to cause any change to global levels of H3K4me3 in wing discs (Figure 2E). As a control, we crossed our Lid transgenes to actin-Gal4 in a wildtype background to ensure that expression of these Lid mutants did not have any deleterious effects. In all cases, expression of our UAS transgenes in a wildtype background gave viable adults, however UAS-LidΔARID, Lid-JmjC* or LidΔPHD1 expressing adult females failed to lay eggs, so were sterile. Surprisingly, ovaries from females expressing these three mutant forms of Lid were phenotypically normal as assessed by dapi and phalloidin staining (data not shown), so the basis for their dominant interference with oviposition is not clear. This effect on egg laying is likely to be due to expression these Lid mutant transgenes in somatic cells of the ovary since germline specific expression using Nanos-Gal4 does not result in sterility (data not shown).
As expected, actin-Gal4 driven expression of wildtype Lid rescued lid10424 mutant animals at the expected Mendelian frequency (Table 1). In contrast, expression of UAS-Lid harboring deletions of its JmjN, ARID, or PHD3 domains fail to rescue lid mutants, suggesting that these domains are essential for development. Expression of LidΔC5HC2 resulted in a small percentage (29%) of lid mutant flies eclosing, all of which died within several days indicating that this domain is essential in adults. In contrast to the third PHD, we found that the first and second PHD fingers of Lid are dispensable for development. While both sexes rescued by LidC1296A were fertile, LidΔPHD1-rescued females were sterile and, like overexpression of this transgene in a wildtype background, LidΔPHD1-rescued flies had phenotypically normal ovaries. We also tested our previously generated JmjC domain point mutant that abolishes demethylase activity for rescue of lid-associated lethality. Actin-Gal4-mediated expression of Lid-JmjC* failed to rescue lid mutants (Table 1), initially suggesting that Lid's demethylase activity is essential for development. However, since overexpression of Lid-JmjC* behaves as a dominant negative in a tissue specific manner, most notably in larval fat body cells [7], it may interfere with maternally deposited wildtype Lid in the rescue experiments described above.
To address the function of Lid's demethylase activity during development, we therefore generated genomic rescue transgenes that fused 3.9 kb of Lid's upstream regulatory region to either a wildtype or JmjC* mutant form of the lid coding region (gLid-WT and gLid-JmjC* respectively; Figure 2A). gLid-WT and gLid-JmjC* transgenes were then crossed into lid10424 and lidk6801 mutant backgrounds, the levels of transgene expression confirmed, and the number of homozygous lid mutant flies scored (Figure 2B, 2C; data not shown). Strikingly, lid mutant animals carrying one or two copies of a gLid-WT or gLid-JmjC* transgene produced phenotypically normal and fertile adult flies at the predicted frequency (Figure 2C; data not shown). Lid's demethylase activity is therefore not essential for Drosophila development. Based on the rescue of lid mutants by enzymatically inactive Lid expressed at endogenous levels, it is likely that this mutant form of Lid failed to rescue in our actin-Gal4 based rescue experiments because its overexpression interferes with maternally deposited wildtype Lid. It is therefore possible that LidΔARID also fails to rescue lid mutants due to its dominant interference with endogenous Lid, thus further examination of this domain will require generation of transgenes using lid's endogenous promoter.
A majority of homozygous lid mutant animals die during pupal development and have increased global levels of H3K4me3 [5], [7], [22], [23]. While Lid can remove di and trimethylated histone H3K4 peptides in vitro, we and others have shown that it only targets H3K4me3 in vivo as only this methyl mark is altered upon Lid overexpression, in lid mutants, and in response to Lid RNAi [5], [7], [22]. Because expression of the demethylase inactive form of Lid is able to rescue lid mutants, we asked whether these animals also have increased global levels of H3K4me3. To examine this, we dissected wing discs from wildtype and lid10424 homozygous mutant larvae and compared the levels of H3K4me3 to lid mutants carrying two copies of gLid-WT or gLid-JmjC* by Western blot. As seen in Figure 2B and 2D, gLid-JmjC* animals show increased H3K4me3 indistinguishable from that observed in lid mutants, demonstrating that the increased level of H3K4me3 observed in lid mutant animals is not the cause of their lethality.
Mutants and RNAi-mediated knock-down of the C. elegans Lid ortholog RBR-2 result in elevated levels of H3K4me3 and a 15–25% reduction in lifespan [24]. To determine whether this is a conserved, demethylase-specific phenotype, we assessed the lifespan of our demethylase inactive Lid flies. We find that lid mutant males rescued by gLid-JmjC* have a significantly shorter lifespan (mean of 37 days) than their wildtype (45 days) or gLid-rescued (46 days) flies (Figure 3A; data not shown). Interestingly, this effect is not observed in females, with the average lifespan of gLid-JmjC*-rescued flies not being significantly different to wildtype (48.1 and 45.1, respectively; Figure 3B). Lacking H3K4me3 demethylase activity therefore has adverse effects on processes required during male adulthood and suggests lifespan phenotypes observed in C. elegans hermaphrodites are likely to be specific to modulation of H3K4me3 levels rather than other RBR-2-dependent processes.
Animals with reduced life expectancies also often show sensitivity to oxidative stressed induced by paraquat. We therefore treated wildtype and demethylase inactive flies with paraquat and found a sex-specific effect of this inducer of oxidative damage. In a similar manner to our lifespan studies in which males were more dramatically affected than females, we find that males are sensitive to paraquat whereas females are not (Figure 3C, 3D). Male Drosophila are therefore more sensitive to the loss of Lid-dependent H3K4me3 demethylation than females, although the molecular basis for this remains unclear.
One explanation for our finding that the loss of Lid's enzymatic activity does not adversely affect development is that its H3K4me3 demethylase activity is compensated for by another demethylase. To date, the JmjC domain-containing protein dKDM2 is the only other Drosophila protein shown to target H3K4me3, although it has also been reported to remove H3K36me2 [25], [26]. To address whether dKDM2 and Lid act in a redundant manner, we tested whether hypomorphic mutations in these two genes genetically interact. lidK6801 homozygotes survive until adulthood at a very low frequency (0.5%), but reach pupal development at 71% of the expected frequency (Table 2). The strongest dKDM2 allele, dKDM2DG18120, is semi-lethal with homozygous adults eclosing at 62% of the expected frequency (Table 2) and these adults are phenotypically normal and fertile. By combining these two mutations, we have found that the phenotype of lid, dKDM2 double mutants is significantly stronger than either single mutant (Table 2), with animals dying during the 1st and 2nd larval instar stages. To demonstrate that Lid's demethylase activity is required for this genetic interaction, we tested whether lid;dKDM2 double mutants could be rescued by our gLid-WT or gLid-JmjC* genomic rescue transgenes. As shown in Table 2, gLid-WT, but not gLid-JmjC* rescued lid;dKDM2 animals, suggesting that Lid and dKDM2 act redundantly in the regulation of H3K4me3.
We originally isolated lid in a genetic screen for regulators and mediators of dMyc-dependent cell growth based on an adult eye phenotype generated by dMyc expression in post-mitotic cells of the developing eye using GMR-Gal4 [7]. Furthermore, we showed that Lid's demethylase activity was not required for its dMyc-dependent functions. To pursue the mechanism by which Lid functions in cell growth induced by dMyc, we crossed our UAS-Lid mutant transgenes to the dMyc overexpressing fly strain and compared their ability to enhance this eye phenotype to that observed in response to expression of wildtype Lid (Figure 4A). Expression of Lid lacking its JmjN, C5HC2 or PHD3 domains failed to enhance the dMyc overexpression eye phenotype while not altering the levels of overexpressed dMyc (Figure 4B–4F; data not shown). As controls, we expressed the Lid deletion transgenes in a wildtype background and found that they resulted in no adult eye phenotype and, unlike fat body cells, Lid-JmjC* and LidΔARID do not have a dominant negative effects in post mitotic cells of the developing eye. We have previously shown that dMyc binds to two regions of Lid: its JmjC domain and its C5HC2 zinc finger [7]. To verify that all Lid deletion proteins retain their ability to bind dMyc, we carried out in vitro binding assays and found that they all bind equivalently (Figure 4G), suggesting that the JmjN, C5HC2 and PHD3 domains of Lid are likely to be required for its Myc-dependent functions in cell growth.
The primary characterized function of Lid is its histone H3 lysine 4 demethylase activity. However, since we have demonstrated that this activity is not Lid's essential function, we chose to further characterize Lid's third PHD finger as this domain is required for it to function with dMyc and is essential for development. Moreover, PHD domains have recently emerged as important interpreters the histone code that act by binding to histone tails that are unmodified, mono-, di- or tri- methylated at specific lysine residues [20]. To address whether Lid's third PHD finger is able to bind methylated histones, we incubated bacterially expressed and purified GST-PHD finger proteins with biotinylated histone peptides mono-, di- or trimethylated at K4, K9 or K27 in vitro and compared this to the binding of Lid's other two PHD fingers and the known H3K4me2/3 binding protein hING2 [27], [28](Figure 5; data not shown).
As seen in Figure 5, Lid's PHD1 finger binds to amino acids 1–21 of histone H3, but not amino acids 21–40 or to histone H4. Lid's PHD1 specifically recognizes unmethylated histone H3 (H3K4me0), as binding is abrogated by mono-, di- or trimethylation of lysine 4, but not methylation of lysine 9. We were unable to detect any in vitro histone binding for Lid's PHD2 finger, however PHD3 bound to both H3K4me2 and H3K4me3, showing a consistent preference for the trimethylated form. The function of both of these PHD fingers is likely to be a highly conserved function of KDM5 proteins, as identical binding specificities have recently been reported for KDM5a [29]. Consistent with the binding of PHD3 to H3K4me2/3 being physiologically relevant, a correlation between KDM5a binding and the presence of this activating chromatin mark has been observed previously using genome-wide arrays, although its physiological relevance has remained elusive [30], [31]. Significantly, the binding of c-Myc also correlates with regions rich in H3K4me2/3 [32]. Based on our findings that Lid's third PHD finger binds H3K4me2/3 and that deleting this domain abolishes its ability to genetically interact with dMyc, we propose that Lid functions to recruit dMyc to regions with high levels of H3K4me2/3 by specifically recognizing this local chromatin context.
Our analyses provide the first investigation of the developmental role of a JmjC domain-dependent demethylase. Five major findings come from this work: (1) Lid's JmjC domain-encoded demethylase activity is dispensable for normal development (2) Loss of Lid's demethylase activity is compensated for by dKDM2 (3) Essential functions of Lid are encoded by its JmjN, C5HC2 and C-terminal PHD zinc finger motifs (4) The N- and C-terminal PHD fingers of Lid bind specific methylated forms of histone tails (5) Lid's C-terminal H3K4me2/3 binding PHD finger is required for it to function in dMyc-mediated cell growth. Taken together, this significantly extends our knowledge of the role of regulated removal and recognition of di- and trimethylated histone H3 lysine 4 during development.
Our finding that Lid's lysine demethylase activity is dispensable for development demonstrates that globally increasing the levels of H3K4me3 is not generally detrimental to development. Similarly, elevating H3K4me1/2 levels by mutating the Drosophila demethylase Lsd1 does not adversely affect development, although these animals show some adult phenotypes and subtle changes to expression of the homeobox genes Ubx and Abd-A [33], [34]. Likewise, Lid's enzymatic activity may serve to fine-tune some gene expression patterns. To date, three genes, E(spl)m4, m7 and m8, have been described as direct Lid targets in Drosophila cultured S2 cells, and these show a 4-fold derepression in response to lid RNAi and a concomitant increase in promoter-proximal H3K4me3 levels [35]. Furthermore, a genetic interaction has been observed during wing development between lid and the E(spl) gene upstream regulator Notch, suggesting that this regulation is biologically important [35], [36]. In mammalian cells, MFN2 and Deltex expression are repressed upon KDM5a overexpresion, derepressed when KDM5 is knocked-down, and show changes in H3K4me3 levels in their promoters [31], [36]. We examined the levels of E(spl)m4, m7 and m8, Marf1 (the Drosophila ortholog of MFN2) and Deltex, but found that their levels were unaltered in RNA extracts from whole larvae or dissected wing imaginal discs from wildtype, lid mutant or lid mutants rescued by gLid or gLid-JmjC* (JS, unpublished). These genes may therefore be regulated by Lid in a small subset of cells in vivo, so cannot be detected using whole wing disc extracts. Effects on gene expression may also be sex-specific since male flies lacking Lid-dependent demethylase activity have a shortened lifespan and are sensitive to paraquat, whereas females are not.
While removing Lid's demethylase activity does not result in lethality, removing this function in combination with another JmjC domain-containing protein, dKDM2, does. This suggests that in the absence of Lid's demethylase activity, dKDM2 can carry out its essential functions and vice versa. RNAi-mediated knock down of dKDM2 has been found to increase H3K36me2 levels in S2 Drosophila tissue culture cells and H3K4me3 levels in adult flies [25], [26]. Surprisingly, we find that global levels of H3K4me3 and H3K36me2 are both unchanged in dKDM2DG12810, dKDM2KG04325 or dKDM2EY01336 homozygous mutant wing discs (CG and JS, unpublished). The reason for the disparity between our results obtained with dKDM2 mutants and previously published data are not clear, but may be due to the difference between the acute loss of dKDM2 mediated by RNAi and the chronic loss in dKDM2 mutants, or to off target effects of the RNAi. The most characterized function of dKDM2 and its mammalian orthologs (KDM2A, KDM2B) is its regulation of rRNA expression [25], [37]. Interestingly, repression of rRNA transcription by KDM5A correlates with changes to H3K63me2 levels, whereas H3K4me3 is unaltered [37]. Based on our genetic interaction between lid and dKDM2, this may be because Lid/KDM5a compensates for the loss of dKDM2's H3K4me3 demethylase activity. Conversely, it is likely that dKDM2 also functions outside the nucleolus and that H3K4me3 regulation by Lid and dKDM2 is essential for development. It is important to note, however, that while Lid's demethylase activity is required for the genetic interaction between lid and dKDM2, we cannot rule out the possibility that dKDM2 requires its H3K36me2 demethylase enzymatic activity not its H3K4me3 activity. Both H3K4me3 and H3K36me2 are chromatin marks associated with active transcription, and it is possible that Lid's H3K4me3 demethylase activity is functionally linked to dKDM2-mediated H3K36me2 demethylation.
Among the JmjC domain-containing proteins, Lid is most structurally similar to the founding member of this class of demethylases, Jumonji (JARID2), having a JmjN, ARID and C5HC2 zinc finger in addition to a JmjC domain. In both mammals and Drosophila, Jumonji is enzymatically inactive because it lacks key residues within its JmjC domain required for Fe2+ and α-ketoglutarate binding [38]. Indeed, while Jumonji has been implicated as a regulator of transcription [30], [38]–[40], the molecular function of the JmjC domain has remained elusive. Taken in conjunction with our finding that Lid's demethylase activity is not essential for development, this raises the exciting possibility that the JmjC domain has important demethylase-independent functions. Consistent with this hypothesis, we find that a genomic rescue transgene with a deletion of the JmjC domain fails to rescue lid mutants (CG and JS, unpublished). Because more than half the known JmjC domain-containing proteins in mammals and Drosophila do not have an ascribed enzymatic activity, a demethylase-independent functions of this domain may be a common feature of this class of protein.
Lid has three PHD fingers and we have demonstrated that its N- and C-terminal PHDs bind specific methylated forms of the histone H3 tail. While Lid's N-terminal H3K4me0-binding PHD finger was not required for development, its third PHD finger, which binds to H3K4me2/3, is essential for viability and is required for Lid to function in dMyc-mediated cell growth. One long-standing question regarding many transcription factors is the mechanism by which they find their appropriate binding site within the genome, as many transcription factors recognize short DNA sequences that are similar or identical. This suggests that binding site specificity may additionally involve the recognition of non-DNA elements such as local chromatin environments. In mammalian cells, c-Myc shows a clear binding preference for E boxes located within a chromatin context containing highly di- and trimethylated nucleosomal histone H3K4 [32]. However, the mechanism by which Myc recognizes this chromatin landscape is unclear. We propose that Lid utilizes its H3K4me2/3 binding C-terminal PHD finger to tether Myc to its preferred chromatin context, thereby permitting selection of biologically important E boxes. Further experiments to more precisely define the role of Lid's PHD finger in Myc-mediated cell growth are ongoing.
In summary, we have demonstrated that Lid's JmjC domain-encoded demethylase activity, its histone H3K4me0-binding N-terminal PHD finger and its PHD2 of unknown function, are dispensable for development. In contrast, all other domains of Lid tested were required to rescue lid homozygous mutants, including its C-terminal, H3K4me2/3 binding, PHD finger that functions in dMyc-mediated cell growth. These findings highlight the importance of characterizing the function of individual domains of transcriptional regulators such as Lid in order to understand the mechanisms by which they regulate gene expression in a developmental context.
UAS-lid and UAS-lidJmjC* have been described previously [7]. All other Drosophila strains were obtained from the Bloomington stock center. Deletions within Lid were made in the pUASp vector by site directed mutagenesis and delete the following amino acids: LidΔJmjN (AA160–206), LidΔARID (AA223–314), LidΔPHD1 (AA450–499), LidΔC5HC2 (AA830–883), LidΔPHD2 (AA1296–1354), LidΔPHD3 (1749–1838 by introducing a stop codon). LidC1296A mutates the first cysteine of Lid's second PHD finger. lid genomic rescue transgenes were generated by fusing a 4.5 kb PCR-generated Xho I fragment containing the lid upstream region and a 4.8 kb Xho I/Not I fragment containing the remainder of the lid coding sequence (either wildtype or JmjC*) into the vector pCasper4. All transgenic flies were generated by The Best Gene (thebestgene.com). Lifespan studies were carried out as described by [41].
To test the ability of UAS-Lid (wildtype and deletion) transgenes to rescue the lid mutant phenotype, a UAS-Lid (or deletion) transgene was recombined onto the lid10424 chromosome. At least 2 independent P element insertions were tested for each to minimize chromosomal position effects. This lid10424, UAS-Lid (or deletion) recombinant chromosome, balanced over CyO, was then mated to the lid10424/CyO; Actin-Gal4/TM6B strain. Rescue was assessed by scoring the presence of straight winged, non-TM6B, progeny. Somatic clones overexpressing UAS transgenes marked by the co-expression of GFP were generated as described in [42]. Longevity studies were carried out as described in [41] and paraquat assays as described in [43].
Histone binding assays: 1 µg of biotinylated histone peptides (Fisher) were incubated with 5 µg purified GST-PHD finger in 1 ml of binding buffer (50 mM Tris pH 7.5, 200 mM NaCl, 2 mM dithiothreitol, 0.5% Nonidet P-40 (v/v), 1 µM ZnSO4, 1% BSA) at 4°C overnight. Complexes were then immobilized using 10 µl Streptavidin-agarose beads (Invitrogen) for 1 hr at 4°C. Immobilized complexes were then washed three times with 1 ml of binding buffer, boiled and loaded on a 4–12% gel. Gels were stained with coomassie blue to visualize bound GST-PHD protein. GST-protein binding assays: 1 µg of purified GST-dMycC [7] was incubated with S35-labeled Lid or Lid deletion proteins made using rabbit reticulocyte lysate (Invitrogen) in 1xPBS, 1% BSA and 0.5%NP-40, washed in 1xPBS, 0.5% NP-40, boiled and loaded onto a 4–12% protein gel. GST-dMycC was visualized using coomassie brilliant blue and S35 detected via standard procedures.
The Lid rabbit and dMyc antibodies have been described previously [7], [44]. Anti-trimethylated H3K4 and H3K36me2 were obtained from Active Motif, and γ-tubulin from Sigma. Western analysis was carried out using standard protocols, infrared conjugated secondary antibodies (LiCOR) and Odyssey scanner and software. Immunofluorescence was carried out as described in [7]. Quantitation of Western blots was carried out using LiCOR odyssey v3.0 software.
|
10.1371/journal.ppat.1004314 | Strain-Specific Properties and T Cells Regulate the Susceptibility to Papilloma Induction by Mus musculus Papillomavirus 1 | The immunocytes that regulate papillomavirus infection and lesion development in humans and animals remain largely undefined. We found that immunocompetent mice with varying H-2 haplotypes displayed asymptomatic skin infection that produced L1 when challenged with 6×1010 MusPV1 virions, the recently identified domestic mouse papillomavirus (also designated “MmuPV1”), but were uniformly resistant to MusPV1-induced papillomatosis. Broad immunosuppression with cyclosporin A resulted in variable induction of papillomas after experimental infection with a similar dose, from robust in Cr:ORL SENCAR to none in C57BL/6 mice, with lesional outgrowth correlating with early viral gene expression and partly with reported strain-specific susceptibility to chemical carcinogens, but not with H-2 haplotype. Challenge with 1×1012 virions in the absence of immunosuppression induced small transient papillomas in Cr:ORL SENCAR but not in C57BL/6 mice. Antibody-induced depletion of CD3+ T cells permitted efficient virus replication and papilloma formation in both strains, providing experimental proof for the crucial role of T cells in controlling papillomavirus infection and associated disease. In Cr:ORL SENCAR mice, immunodepletion of either CD4+ or CD8+ T cells was sufficient for efficient infection and papillomatosis, although deletion of one subset did not inhibit the recruitment of the other subset to the infected epithelium. Thus, the functional cooperation of CD4+ and CD8+ T cells is required to protect this strain. In contrast, C57BL/6 mice required depletion of both CD4+ and CD8+ T cells for infection and papillomatosis, and separate CD4 knock-out and CD8 knock-out C57BL/6 were also resistant. Thus, in C57BL/6 mice, either CD4+ or CD8+ T cell-independent mechanisms exist that can protect this particular strain from MusPV1-associated disease. These findings may help to explain the diversity of pathological outcomes in immunocompetent humans after infection with a specific human papillomavirus genotype.
| Infection with papillomaviruses can cause benign warts (papillomas) on skin and mucosae of humans and animals but also malignancies, especially anogenital carcinomas and, in genetically predisposed or immunocompromised individuals, cutaneous squamous cell cancers. Control and clearance of these viruses are thought to be mediated by the cellular immune system, however experimental determination for the necessary cellular effector(s) is lacking. The recently identified mouse papillomavirus (MusPV1, also designated “MmuPV1”) is known to induce papilloma formation on skin of immunodeficient mice. However, its pathogenesis in immunocompetent mice is unclear. Our study shows that in an immunocompetent setting, MusPV1 generally causes asymptomatic skin infections, but no lesion outgrowth. Visible papillomas were consistently observed after profound immunosuppression in some, but not other, strains of mice. By selective removal of distinct cellular immune populations and employing genetically modified mice, we could show that T cells are pivotal for controlling MusPV1 infection and disease. We further show that surprising differences in the T cell subsets are required for protection in different strains of immunocompetent mice. This implies that unanticipated effector mechanisms can control virus infection and pathogenesis in specific genetic backgrounds. The findings may help to explain the wide of range of pathologic outcomes of infection by a specific human papillomavirus type in immunocompetent people.
| Papillomaviruses (PV) are DNA tumor viruses that infect stratified squamous epithelia of the skin and mucous membranes of humans and many other vertebrate species [1]. PV infections are species-restricted and region-restricted, in that only part of the skin and mucous membranes of the host species of a given PV is permissive for productive infection [2]. More than 150 human PV (HPV) genotypes (types) have been identified. These viruses can induce long-term infection that, depending on the virus type and its human host, may not cause lesions, may induce benign lesions (warts or papillomas), or may lead to the development of anogenital carcinomas, most notably cervical and oropharyngeal cancers. Certain cutaneous HPV types have also been implicated in the pathogenesis of some epidermal squamous cell cancers in genetically predisposed or immunocompromised individuals [3].
Although neutralizing antibodies against the viral capsid proteins are sufficient to prevent PV infections, cell-mediated immunity is generally thought to control the infections once they become established. For instance, individuals with underlying T cell deficiencies, but not B cell deficiencies, often have difficulties in controlling and clearing HPV-induced neoplasia [4]. However, it has been difficult to provide experimental support for this concept or to determine which particular subset(s) of immunocytes are responsible for these activities. Although studies of animal PVs, such as bovine PV (BPV), cottontail rabbit PV (CRPV), and canine PV, have contributed to our understanding of PV biology [reviewed in 5], the limited immunological reagents available for these species have hampered critical investigations of immune regulation of PV infection with these systems.
The recent identification of MusPV1 (also designated “MmuPV1”), which is the first domestic mouse PV, provides an excellent opportunity to investigate the immune mechanisms that control PV infection in a mammalian species whose immunology has been well characterized. MusPV1 was found initially in an inbred NMRI-Foxn1nu/Foxn1nu nude laboratory mouse colony [6]. These immunodeficient mice spontaneously developed papillomas at cutaneous surfaces near the mucocutaneous junctions of nose and mouth, from which the virus was isolated. Subsequent studies reported that papillomas could be induced by experimental infection with MusPV1 in immunodeficient B6.Cg-Foxn1nu/Foxn1nu, Foxn1nu/Foxn1nu and SCID SHO mice [7], [8]. We also determined that athymic NCr nu/nu (nude) mice could be infected with in vitro synthesized full-length genomic DNA of MusPV1, giving rise to non-regressing cutaneous papillomas from which high titers of authentic, infectious MusPV1 virions were isolated and serially passaged [9].
In contrast to these reports of MusPV1 in immunodeficient mice, the study of MusPV1 in immunocompetent mice has been limited. The initial report of MusPV1 noted that when cell-free extracts from papillomas in the immunodeficient mice were used to inoculate the dorsal skin of S/RV/Cri-ba/ba mice, which have an unknown mutation that results in thin short hair, they induced small papules at most injection sites. These lesions, which were not characterized further, took at least three weeks to develop, and regressed spontaneously by 8 weeks post-inoculation [6]. Subsequent efforts to induce lesions in C57BL/6J mice were unsuccessful [7]. Thus establishment and further characterization of MusPV1 infection and disease in an immunocompetent setting were warranted.
In this study we aimed at characterizing MusPV1 infection in different immunocompetent murine strains and sought to determine the key immunologic players primarily responsible for control of cutaneous PV infection and papilloma induction. Our results revealed asymptomatic MusPV1 infection in these immunocompetents and demonstrated that profound immunosuppression can render these strains that had various H-2 haplotypes susceptible to MusPV1-induced papilloma formation of the skin. This is reminiscent of infection with HPV of genus beta that also predominantly induces asymptomatic skin infections in immunocompetent individuals but can induce visible lesions after immunosuppression. The observed differences in the efficiency of papilloma outgrowth between the individual murine strains corresponded partially to their previously reported susceptibility to chemical-induced skin papillomatosis [summarized in 10], suggesting that there may be similarities between their relative susceptibility to papilloma formation induced by MusPV1 and to chemical carcinogens. We further provide clear experimental proof that T cell functions are required for control of PV infection and disease and reveal striking differences in the protective capacities of T cell subsets among murine strains, including CD4-mediated effector mechanisms.
Several inbred immunocompetent strains of mice, namely FVB/NCr, BALB/cAnNCr, DBA/2NCr, A/JCr, C57BL/6NCr (C57BL/6), 129S6/SvEv, C3H/HeJCr, and as well the outbred Cr:ORL SENCAR (SENCAR) were evaluated for their susceptibility to papilloma induction by MusPV1 (Table 1). The inbred strains have a range of H-2 haplotypes and a range of reported susceptibility to chemical carcinogen-induced skin papillomas (Table 1) [10]–[12]. DBA/2NCr and BALB/cAnNCr are both H2d, but vary in their sensitivity to chemical carcinogenesis, as is also true for the two H2b strains, C57BL/6 and 129S6/SvEv. The SENCAR strain was included because it was selectively bred for high susceptibility to skin tumor induction by chemical carcinogens. All strains were infected with 6×1010 MusPV1 virions per animal on pre-scarified tail skin, using a previously optimized procedure for PV infection of mouse skin [9], [13], and followed for 4 months. During this period, no papilloma outgrowth was observed in any strain, although the same preparation and dose of MusPV1 virions consistently induced large papillomas in immunodeficient athymic NCr nude mice within one month of inoculation (data not shown).
To determine whether immunosuppression renders these mouse strains susceptible to MusPV1-induced papilloma formation, animals were treated systemically with the immunosuppressant CsA, which can inhibit T cell activation and lymphokine production and is routinely used in humans for prevention of transplant rejection [14], [15]. CsA treatment was initiated one week prior to infection with 4.2×1010 MusPV1 virions and continued for additional 4 weeks, during which the mice were evaluated for papilloma formation. The results uncovered a hierarchy of susceptibility to MusPV1-induced papillomas among the strains (Table 1 and Figure 1A–H). SENCAR and FVB/NCr mice were highly susceptible. Most (6/8; 75%) of the SENCAR mice (Figure 1A) developed large and raised papillomas, with a mean length of 8.2 mm. All of the FVB/NCr mice (Figure 1B) inoculated with MusPV1 developed papillomas (8/8; 100%), which were moderately elevated, with a mean length of 7.6 mm. The BALB/cAnNCr, A/JCr, and 129S6/SvEv strains displayed intermediate susceptibility for papilloma development. For BALB/cAnNCr (Figure 1C), 5/8 animals developed papillomas, which were smaller than those in SENCAR and FVB/NCr, with a mean length of 4.6 mm. The comparable numbers for A/JCr mice (Figure 1D) were 4/8 and 2.9 mm, respectively, while those for 129S6/SvEv (Figure S1A) were 2/4 and 2.5 mm, respectively. By contrast, CsA-treated C57BL/6, DBA/2NCr, and C3H/HeJCr mice were comparatively resistant to MusPV1-induced papillomas. None of the virally inoculated C57BL/6 (Figure 1E) and DBA/2NCr mice (Figure S1B) developed papillomas (0/8 and 0/4, respectively, both 0%), and only one C3H/HeJCr mouse (Figure 1F) developed a lesion (1/8; 12.5%), and its length was only 2 mm.
The papillomas that did develop in the tested strains were attributable to the combination of CsA treatment and MusPV1 infection, as littermates inoculated with the same amount of MusPV1 virions in the absence of CsA or mock inoculation with CsA treatment did not result in papilloma formation (Figure 1G, 1H; SENCAR shown as representative strain). The lesions that developed were histologically verified to be papillomas with numerous koilocytes in the epithelium (Figure 1I; SENCAR shown as representative), and extracts of the lesions contained infectious MusPV1 virions that induced papillomas, in athymic NCr nude mice, whose morphology was identical to those previously reported for MusPV1 in this immunologically impaired strain (Figure 1J) [9].
Papillomas in the CsA-treated/MusPV1-inoculated immunocompetent strains grew progressively and did not regress during the period of CsA administration. However, after cessation of CsA administration, the lesions completely regressed within several weeks (Figure 1K and 1L; same SENCAR mouse at a 6-week interval). The length of time to regression depended upon the size of the lesion, with smaller ones tending to regress sooner than larger ones (Figure S1C). Therefore, the immunosuppressive phenotype resulting from CsA treatment was required for maintenance of the papillomas, in addition to their induction.
In a rabbit model, persistent PV genome expression was reported at mucosal sites of infection following lesion-regression [16]. Subsequent immunosuppression led to an increase in viral copy numbers and even to reappearance of small lesions, consistent with reactivation of latent infection [17]. To investigate latency of MusPV1, all mouse strains, except for DBA/2NCr and 129S6/SvEv, that had been inoculated with 6×1010 virions per animal, were subjected to CsA administration 4 months later for a period of 4 weeks (corresponding to 5 months post-infection). After this period, papilloma outgrowth was not observed in any of the animals (data not shown). MusPV1 E1∧E4 spliced transcripts, a marker for PV infection [9], [18], [19], and viral genomes were undetectable in the skin tissues taken from the inoculation sites (Figure S2), suggesting that MusPV1 infection in cutaneous tissues was efficiently cleared by a fully functioning murine immune system prior to immunosuppression and therefore do not persist long-term in a latent state in immunocompetent mice.
When susceptibility to MusPV1-induced papillomas was considered in the context of the H-2 haplotype of the mice, surprisingly, there was little correlation (Table 1). For example, although C57BL/6 and 129S6/SvEv are both H2b, the former strain was resistant to MusPV1-induced papilloma formation, while the latter strain had intermediate susceptibility. In addition, DBA/2NCr was resistant, while BALB/cAnNCr had intermediate sensitivity, although both strains are H2d. By contrast, we noted some correlation between the observed susceptibility to MusPV1-induced papillomas and their previously reported susceptibility to chemical carcinogenesis (Table 1) [10], suggesting that common mechanisms may, in part, control the relative susceptibility to MusPV1-induced papilloma formation and to chemical carcinogens. However, this correlation is based on different published studies on the strain-specific susceptibility to chemical carcinogenesis [summarized in 10] and is not complete, as DBA/2NCr is reported to be susceptible to chemical carcinogenesis, but MusPV1 inoculation did not produce papillomas, and the sensitivity of 129S6/SvEv to MusPV1 may be less than its reported sensitivity to chemical carcinogenesis. Further experimentation is required to validate the partial correlation between strain-specific susceptibility to papillomatosis and chemical carcinogens.
Next, we investigated the impact of varying doses of MusPV1 virions on representatives of the most sensitive and the most resistant strains, SENCAR and C57BL/6, respectively. Serial titrations of the inocula revealed that 2–3 weeks after infection with very high doses of 1×1012 MusPV1 virions per animal, the majority of the SENCAR mice developed small lesions at the site of infection (covering about 1 cm of the length of the tails) (Table S1), even in the absence of immunosuppression. Sporadic papilloma outgrowth was observed after infection with 1×1011 MusPV1 virions. All lesions spontaneously regressed within 1–2 weeks after formation (corresponding to 3–5 weeks post-infection). Consistently, lesions did not arise with lower amounts of inocula ranging from 1×1010 to 1×108 virions. The transient papillomas in the immunocompetent SENCAR mice obtained after inoculation with 1×1012 MusPV1 virions were morphologically and histologically similar to papillomas that formed under CsA immunosuppression and to those previously reported [9] (Figure S3A and S3B) and contained low, but detectable MusPV1 E1∧E4 spliced transcripts, a marker for infection (Figure S3C, lane designated M). In the epithelium of these papillomas, expression of the major capsid protein L1 was demonstrated by immunofluorescent microscopy (Figure S3D, E). Cell extracts derived from these papillomas contained infectious MusPV1 virions that were able to induce papilloma formation on the tails of athymic nude NCr mice after experimental transmission (Figure S3F), indicating that MusPV1 could be propagated in immunocompetent SENCAR mice. For C57BL/6 mice no tumor formation was observed, even at very high doses of 1×1012 MusPV1 virions, further supporting the higher resistance of this particular strain to MusPV1 pathogenesis (Figure S3G).
To investigate whether qualitative and/or quantitative differences in infection parameters were responsible for the observed variation in papilloma formation between strains, the relative level of MusPV1 E1∧E4 spliced transcripts were determined in skin tissues taken 4 weeks after infection with 4.2×1010 MusPV1 virions (corresponding to 5 weeks of CsA administration) from the virally inoculated sites of all strains, except for DBA/2NCr and 129S6/SvEv (Figure 1M and Table 1). Skin necropsies were taken from all animals within the remaining six tested strains, including those mice that lacked visible lesions. To ensure comparability, specimens of approximately the same size were processed, and the levels of the viral transcripts were adjusted to level of endogenous beta-actin transcripts from the same sample (Figure 1M, the lanes designated CM). The resulting ratios corresponded well to the macroscopic appearance of the lesions. High levels of MusPV1 E1∧E4 copies relative to beta-actin levels were detected in tissues of CsA-treated/MusPV1-infected SENCAR and FVB/NCr mice. In CsA-treated/MusPV1-infected BALB/cAnNCr and A/JCr mice, the lower levels of E1∧E4/beta-actin reflected the less pronounced and smaller lesions observed in these intermediately susceptible strains. Consistent with the lack of visible papillomas, even lower levels of MusPV1 E1∧E4 spliced transcripts were detected in infected C57BL/6 and C3H/HeJCr mice treated with CsA. In tissues taken from MusPV1-infected littermates that had not received CsA, the levels of E1∧E4 spliced transcripts per copy beta-actin were low, but detectable, in all strains (Figure 1M, lanes designated M), but were not detected in mock-infected mice (Figure 1M, lanes designated 0). Thus MusPV1 established persistent asymptomatic infections in all of the immunocompetent mouse strains.
Immunofluorescent microscopy was used to examine the relative expression of the major capsid protein L1 in these tissues. SENCAR was used as the high susceptibility strain, BALB/cAnNCr as the intermediate one, and C57BL/6 as the resistant one. Given the correlation between the susceptibility of these strains to papilloma formation and their relative level of E1∧E4 spliced transcripts, it was anticipated that a similar correlation would be seen for L1 expression. However, there was abundant L1 staining in the epithelium of all CsA-treated/MusPV1-infected animals of all three strains (Figure 1N–P), including the resistant C57BL/6 mice, which did not develop papillomas. Similar to previous reports in athymic NCr nude mice [9], punctate L1 expression was found in the cytoplasm of keratinocytes in the basal and lower spinous layers of the epithelium, while nuclear L1 expression was confined to the upper spinous and granular layers, suggesting active virion production in these more differentiated epithelial layers. Additionally, some L1 positive squames that presumably enclose matured virions prior to shedding were found in these tissues. L1 expression was restricted to the epithelial compartment, although seemingly positive staining could occasionally be observed below the basement membrane due to the (trans-)sectioning of the papilloma's deregulated architecture. L1 staining was not detected at sites of lesion regression following cessation of CsA treatment (data not shown) or in the skin of untreated MusPV1-infected mice at this time point, regardless of the strain (Figure 1Q–S). It remains to be determined whether CsA treatment deregulates L1 protein expression to a greater degree than that of the E1∧E4 spliced transcripts, or if a difference in sensitivities between the two assays might explain the disparate results. Interestingly, L1 was readily detected in infected sites on day 14 after inoculation in untreated mice of all three strains, raising the possibility that adaptive immune responses arising between two and four weeks may regulate late gene expression and thereby virion production (data not shown).
The recruitment of T cells to the infected tissue was evaluated in the SENCAR mice, which developed large papillomas with CsA treatment. Tissue taken 4 weeks after inoculation with 4.2×1010 MusPV1 virions (corresponding to 5 weeks of CsA administration) was analyzed for the presence of CD4+ and CD8+ T cells. By immunofluorescent staining (IFS), higher numbers of both T cell subsets were present in the MusPV1-infected tissue, independent of whether the mice had been treated with CsA, although L1 expression was only detected in the infected CsA-treated mice (Figure 2A and 2B). Many CD4+ T cells infiltrated both the epithelium and the underlying dermis in the papillomatous tissue of the CsA-treated mice and the non-papillomatous tissue of the untreated mice (Figure 2A and 2C), in contrast to the limited number of CD4+ T cells in the mock-infected control skin (Figure 2E). Similarly, numerous CD8+ lymphocytes were found in the papillomas of CsA-treated/MusPV1-infected mice and in the macroscopically unchanged MusPV1-infected skin of untreated animals (Figure 2B and 2D). However, the CD8+ T cells were localized predominantly in the epithelium and were sparse in the dermis. Only a few widely spaced intraepithelial CD8+ T cells were found in the mock-infected control skin (Figure 2F).
Consistent with these findings, quantification of CD4+ and CD8+ T cell numbers in the infected skin tissues (Figure 2G) by flow cytometry confirmed that MusPV1 induced strong recruitment of the T cells even in CsA-treated mice. Viral infection without CsA resulted in a 15.3- and 12.8-fold increase in CD4+ and CD8+ T cell numbers, respectively, compared to mock-infected littermates. Similarly, viral infection with CsA treatment resulted in an increase in CD4+ and CD8+ T cells that was, respectively, 9.3- and 7.8-fold higher when compared with mock-infected mice not treated with CsA, or was, respectively, 31.1- and 15-fold higher, compared to mock-infected/CsA-treated controls. Thus, MusPV1 infection efficiently recruits both T cell subsets to the site of infection, even in the presence of CsA.
However, when the same mice were analyzed for their CD4+ and CD8+ T cell levels in blood (Figure 2H), spleen (Figure 2I), and draining lymph nodes (Figure 2J), the impact of MusPV1 infection was much less pronounced. Although some MusPV1-dependent differences seen in the CD4+ or CD8+ T cells in these systemic compartments were statistically significant, there was less than a two-fold difference for each comparison.
To investigate the role of T cells as key effectors responsible for controlling MusPV1 infection and/or papilloma formation, we studied the consequences of MusPV1 infection following antibody-induced depletion of specific T cell populations in two mouse strains: SENCAR, which had been found to be highly sensitive to papilloma formation following CsA treatment, and C57BL/6, which had been found to be resistant. The SENCAR results are presented in this section and the C57BL/6 results in the next section.
In the first experiment, the CD3+ T cell population, which includes most T cell subsets, was specifically removed from the SENCAR mice by systemic administration of a monoclonal antibody (mAb) recognizing murine CD3. Depletion was started at various time points, from one week prior to viral infection (day −7) to 7 weeks (day +49) after infection, and it was maintained for seven weeks for all groups. The mice were infected on day 0 with 7.3×1010 MusPV1 virions, and the number and size of lesions were determined for each group after 7 weeks of immunodepletion (Figure 3A–H). Depletion starting 1 week prior to infection (day −7; Figure 3A) resulted in the development of large papillomas (mean length = 13 mm) in all 5 mice in this group. When depletion was started on the day of infection (day 0; Figure 3B) or 1 week after infection (day +7; Figure 3C), most mice developed papillomas (4/5 in both groups), but their size was somewhat smaller (mean length = 10.5 mm for both groups). The efficiency of papilloma formation was markedly reduced (2/5) when depletion was started 1 month post-infection (day +28; mean length = 4 mm; Figure 3D). No papillomas were seen when depletion was started 7 weeks after infection (day +49; 0/5; Figure 3E), suggesting that MusPV1 infection is effectively cleared prior to this time. MusPV1 inoculation after administration of an isotype control (Figure 3F) or without mAb addition (Figure 3G) did not induce papilloma formation. The papillomas in the CD3-depleted mice were histologically verified (data not shown), and virions extracted from these lesions were able to cause papillomas in athymic NCr nude mice (Figure 3I), thus demonstrating that depletion of CD3+ T cells enables the complete viral lifecycle.
Throughout the papillomatous tissues taken after 7 weeks of depletion (corresponding to 6 weeks post-infection) from the CD3-depleted mice (Figure 3J and 3K; representative of group day −7 shown), exceptionally large amounts of MusPV1 L1 protein were detected by IFS, and CD4+ (Figure 3J) and CD8+ (Figure 3K) T cells were absent, as expected, since these T cells are CD3+. In contrast, skin tissues taken from isotype-administered controls (Figure 3L and 3M) lacked MusPV1 L1 protein, and CD4+ (Figure 3L) and CD8+ (Figure 3M) T lymphocytes were readily detectable. No L1 protein and only isolated T cells were observed in mock-infected controls (data not shown).
Taken together, these findings indicate that, in SENCAR mice, T cells are obligatory for controlling MusPV1 infection and associated papillomatosis. T cell-mediated responses either clear MusPV1 infection within 7 weeks post-inoculation or control it by a mechanism that does not permit reactivation after their removal.
We next determined whether depletion of either the CD4+ or the CD8+ T cell population alone, by administration of anti-CD4 or anti-CD8 mAbs, respectively, would be sufficient to induce sensitivity to MusPV1-induced papillomas in the SENCAR mice. The immunodepletion was started one week prior to infection with 5.1×109 MusPV1 virions, and the depleted state was monitored in the animals' blood prior to infection (day −1) and every 2 weeks by flow cytometric analysis (Figure S4). At six weeks post-infection, the majority of the CD4-depleted SENCAR mice had developed papillomas (7/9; 78%) (Figure 4A). Similarly, there were papillomas in 7/9 (78%) of the CD8-depleted mice (Figure 4B) at this time point. MusPV1-infected controls with (Figure 4C) and without (Figure 4D) isotype depletion did not develop lesions. The mean length of the lesions was similar in the CD4-depleted and CD8-depleted groups, being 10.7 mm vs. 12.7 mm, respectively (Figure 4E). The fact that removal of either subset allowed papilloma outgrowth suggests that cooperation between CD4+ and CD8+ T cells is required for effective control of infection and lesion development in SENCAR mice.
As expected, MusPV1 L1 protein was present in skin tissues taken at this time point from both CD4- (Figure 4F and 4G) and CD8-depleted (Figure 4H and 4I) animals. The L1 staining pattern and the intensity were similar to the pattern observed in CsA-treated/MusPV1-infected and athymic NCr nude mice [9], with punctate, cytoplasmic L1 expression in the lower epithelium and nuclear L1 positivity in the upper epithelial layers. Compared to the results obtained after CD3+ T cell depletion, L1 expression seemed less pronounced, but this difference may be due to the lower amount of virus used for the initial inoculation.
Depletion of the targeted T cell subset at the site of infection was associated with the apparent absence of the depleted subset but with no loss in the infiltration by the non-depleted subset, indicating the neither subset was needed for recruitment of the other (Figure 4F–I). After 5 weeks of depletion (corresponding to 4 weeks post-infection), when the growth of papillomas was clearly visible, lymphocytes from each group (4 mice/group) were isolated from the site of infection (Figure 4J), and the blood (Figure 4K), spleen (Figure 4L) and draining lymph nodes (Figure 4M) for analysis by flow cytometry. The results in the skin confirmed the validity of the microscopy results, and demonstrated the efficient and specific depletion of the targeted subpopulation in each compartment at this time point.
T cell depletion experiments analogous to those described for the SENCAR mice were performed in C57BL/6 mice. MAb-induced depletion of CD3+ T cells was initiated one week prior to infection with 5.3×1010 MusPV1 virions, and the depleted state was monitored in the animals' blood one day prior to infection and every second week (Figure S5). In parallel, mAb-induced depletion was initiated, for the same length of time, for the CD4+ T cells, the CD8+ T cells, and both the CD4+ and the CD8+ T cells (CD4+8 depletion), and the depletion monitored similarly (Figure S5). After 7 weeks of depletion (corresponding to 6 weeks post-infection), all of the CD3-depleted animals (14/14) inoculated with MusPV1 had developed papillomas (100%) (Figure 5A), similar to the results observed in the SENCAR mice. This finding was somewhat unexpected, given that continuous CsA treatment of the C57BL/6 mice had not led to the development of MusPV1-induced papillomas (Figure 1E). On the other hand, the mice with depletion of only CD4+ T cells (0/5) (Figure 5B) or of only CD8+ T cells (0/5) (Figure 5C) did not develop papillomas, which is in contrast to the results with the SENCAR mice under the same conditions. However, combined CD4+8 depletion in the C57BL/6 mice did lead to MusPV1-induced papillomas (10/10) (Figure 5D).
When crude skin extracts taken from the mice 6 weeks after infection were evaluated by western blotting for the presence of L1 protein, it was only detected in the two groups that developed papillomas, CD3 depletion and the combined CD4+8 depletion (Figure 5E).
Given that the positive results with C57BL/6 mice were seen with CD3 depletion or combined CD4+8 depletion induced by mAb, further analysis was focused on depletion induced by these mAb. At 6 weeks after virus inoculation, high levels of MusPV1 E1∧E4 spliced transcripts, as a measure of persistent infection, were found in infected skin tissues of CD3- and combined CD4+8-depleted animals (Figure 5F), the two depletion conditions that resulted in papillomas. In contrast, no E1∧E4 spliced transcripts were detected in CD4- or in CD8-depleted mice, demonstrating the effective control of MusPV1 infection in these animals.
When the expression of L1 protein at the inoculated skin sites was examined by IFS, the results corroborated the results observed with the E1∧E4 spliced transcripts. L1 was abundantly expressed in the cytoplasm and nuclei of infected keratinocytes in the characteristic pattern of MusPV1 in mice with CD3 depletion and CD4+8 depletion, but not with CD4 depletion or CD8 depletion (Figure 5G–5N). As expected, no CD4+ cells were seen in skin sites from CD3-depleted, CD4+8-depleted, or CD4-depleted mice, although they were detected in CD8-depleted mice (Figure 5G, 5M, 5I and 5K). Conversely, no CD8+ cells were seen in skin sites from CD3-depleted, CD4+8-depleted, or CD8-depleted mice, although they were detected in CD4-depleted mice (Figure 5H, 5N, 5L and 5J). Quantification of T cell infiltrates in the infected skin sites (Figure 5O) and the blood (Figure 5P), spleen (Figure 5Q), and draining lymph nodes (Figure 5R) by flow cytometric analyses further confirmed the efficiency and specificity of the T cell depletion.
To address whether C57BL/6 mice genetically engineered to be knock-outs (KO) for CD4+ or CD8+ T cells might be susceptible to papilloma induction by MusPV1, these KO mice were inoculated with 9.4×1010 MusPV1 virions. However, as had been true of the mice with mAb-induced depletion of these individual T cell subsets, the CD4 KO mice (0/5) (Figure 6A) and the CD8 KO mice (0/5) (Figure 6B) were resistant to papilloma formation when observed for 3 months, as were wild-type (wt) C57BL/6 controls (0/5) (Figure 6C), and did not produce detectable levels of L1 protein in crude skin extracts at this time point (Figure 6D).
However, analysis of MusPV1 E1∧E4 spliced transcripts, as a measure of infection, at earlier time points (Figure 6E) revealed substantially higher numbers of MusPV1 E1∧E4 spliced transcripts relative to beta-actin in CD4 KO mice in the first three weeks post-infection, compared to wt C57BL/6 and CD8 KO mice. The relative MusPV1 E1∧E4 levels in the CD4 KO mice peaked at week 2 post-infection, showing 36-fold and 27-fold higher values than in wt and CD8-deficient C57BL/6 mice, respectively, gradually decreased thereafter and became undetectable at week 6 post infection. Very low numbers of MusPV1 E1∧E4 transcripts could be detected within the first 2 week after infection in wt C57BL/6 mice and were undetectable 3 weeks after infection. In contrast the levels of MusPV1 E1∧E4 transcripts in the CD8 KO mice remained uniformly very low, but detectable, throughout the experiment, suggesting that CD4+ T cells are specifically involved in early, presumably innate, immune responses that initially control viral gene expression.
CD1d-deficient C57BL/6 mice, which selectively lack the natural killer (NK)-T cell population, also failed to form papillomas after inoculation with 6.8×1010 MusPV1 virions and a 3.5 month observation period (Figure 6F), indicating that ablation of this subset by itself is not sufficient to permit papilloma formation, in contrast to the crucial role of T cells.
In this study, we have determined that a variety of immunocompetent mouse strains are resistant to papilloma induction by MusPV1, although a limited degree of virus expression can be detected at 4 weeks after infection. However, immunosuppression induced by CsA uncovered a strain-dependent hierarchy in the degree of susceptibility to papilloma formation and virus production. Papilloma formation correlated more closely with expression of E1∧E4 spliced transcripts than with expression of the major L1 capsid protein, as clinically normal virally inoculated skin sites in the resistant C57BL/6 mouse had relatively high levels of L1, but low E1∧E4 levels. Inoculation with what we assume are super-physiological amounts of MusPV1 virions (1×1012 virions) seems to overcome early control of infection in the highly susceptible SENCAR strain, allowing for transient papilloma formation. In the more resistant C57BL/6 strain, papilloma outgrowth was not observed after this stringent challenge, supporting the conclusion that this strain is better able to control infection than SENCAR mice.
To date, strong circumstantial evidence supports the role of cell-mediated immunity, especially T cells, in controlling and eliminating established PV infection and neoplasia. Evidence for the pivotal role of T cells has emerged from studies of humans infected with the human immunodeficiency virus [20]. In these individuals, higher prevalence of HPV infection, especially of the anogenital tract, viral persistence, and very often the presence of multiple types, has been observed. An important role for T cells is also supported by observations that iatrogenic immunosuppressed transplant recipients have high rates of extensive viral warts, HPV-associated anogenital cancers, and non-melanoma skin cancer [3]. The current study demonstrated that treatment with CsA, whose predominant activity is against T cells and which is routinely used in humans, led to papilloma formation and persistent MusPV1 infection in most, but not all, of the murine strains tested. However, the finding that depletion of CD3+ T cells rendered the mice susceptible to papilloma formation and exuberant viral infection, even in C57BL/6 mice that were resistant after CsA treatment, provides direct experimental evidence for the critical importance of T cell function in the control of MusPV1 infection.
The available data with other PV systems have not defined the T cell subpopulation(s) that is the key player responsible for virus control and papilloma regression. Observational studies in humans have found that regression of anogenital warts is accompanied by a massive infiltration of CD4+ lymphocytes, both within the papilloma's stroma and the epithelium [21]. However, intraepithelial CD8+ T cells have also been associated with regression of cervical intraepithelial lesions [22], [23]. T cell infiltrates, consisting of predominantly CD4+ T cells but also containing CD8+ T cells, have been described in regressing mucosal lesions caused by BPV type 4 [24], canine oral PV [25], and rabbit oral PV [26]. In cutaneous papillomas caused by CRPV, a mostly CD8+ T cell infiltration of the epithelium, with very few accompanying CD4+ T cells, was demonstrated [27].
In our murine system, MusPV1 infection of the sensitive SENCAR mouse induced a dense CD4+ T cell infiltrate in the dermal and the epithelial compartment, as well as an intraepithelial one composed of CD8+ T cells, whether the animals developed papillomas as a result of the immunosuppressive CsA treatment or had clinically normal skin because they had not been immunosuppressed. Thus, the infiltrate was similar, independent of whether the mouse developed lesions, indicating that functional properties of the T cells, rather than epithelial trafficking, were primarily affected by CsA treatment.
The availability of many immunological reagents for the domestic mouse made it possible to critically evaluate individual T cell subsets in the control of PV infection and associated neoplastic disease. In SENCAR, depletion of either CD4+ or CD8+ T cells allowed papillomatosis. By contrast, combined depletion of CD4+ and CD8+ T cells was necessary for papilloma development in C57BL/6. These discrepant results were confirmed using genetically modified C57BL/6 mice deficient in either CD4+ or CD8+ T cells.
Clearance of virally-infected epithelial cells and regression of epithelial neoplasia have most often been attributed to CD8+ T cells, especially cytotoxic ones. However, in many well-characterized systems, CD4+ T help is necessary both for induction of primary CD8+ T cell responses and for their proliferation, activation, and differentiation into effector cytotoxic T lymphocytes (CTL) [28]–[30]. In the absence of CD4+ T help, CD8+ T cells often fail to acquire antiviral effector functions, including the ability to produce antiviral cytokines, such as interferon (IFN)-γ and tumor necrosis factor (TNF)-α, and cytotoxic molecules, such as perforin and granzymes. In this scenario, CD4+ T cells would contribute to protection indirectly, rather than directly providing a critical T cell effector function. However, the T cell subset depletion and knock-out data in C57BL/6 data make it clear that protective antiviral CD8+ T cell responses to a PV infection can develop in the absence of CD4+ T help, and that CD4+ T cell-dependent effector functions can control PV infection in the absence of CD8+ T cell functions.
There is an increasing body of evidence that CD4+ T helper-independent CD8+ CTL responses can be elicited by some pathogens, such as ectromelia virus [31], influenza virus [32], lymphocytic choriomeningitis virus, dengue virus [33], and Listeria monocytogenes [34]. However, CD4+ T cells were required for efficient local recruitment of herpes simplex virus-specific CD8+ T cells to the vaginal epithelium in a murine model of herpes virus infection [35]. It was therefore unexpected that, in both C57BL/6 and SENCAR mice, CD8+ T cells infiltrated the skin sites inoculated with MusPV1 whether or not the mice contained CD4+ T cells, and vice versa. The discrepant conclusions between our study and the murine herpes virus infection report [35] may be attributable to substantial differences in the two experimental systems, most notably the adoptive transfer of transgenic CD8+ T cells in the herpes virus study and the de novo generation of the T cells by in situ virus infection in the current study.
Various mechanisms for CD8+ T cell activation in the absence of CD4+ T cell help have been proposed, such as direct signaling through CD40 present on antigen-presenting cells [36], [37], up-regulation of CD40L on dendritic cells to enhance CD8+ T cell responses via direct engagement of CD40 on activated CD8+ T cells [38], and NK cell-derived IFN-γ [39]. These bypass mechanisms may also be active in C57BL/6 mice and explain the observed CD4-independent suppression of MusPV1-induced papillomatosis.
CD4+ T cells could independently control PV infection by either a cytokine-mediated mechanism or by direct cytotoxicity [40]. Several recent studies have reported that cytolytic CD4+ T lymphocytes, which may represent a new CD4+ lineage, can contribute to the control of certain viral infections [40]–[42]. In C57BL/6 mice, lymphochoriomeningitis virus-specific CD4+ T cells are capable of in vivo cytolytic killing of peptide pulsed MHC II-positive lymphocytes [43]. However, none of these studies have demonstrated complete protection from productive infection and disease in the absence of CD8+ T cells.
Cytolysis by CD4+ T cells would likely require a direct interaction of the T cells with MHC class II molecules on the infected keratinocytes (the only cell type that is normally infected by PVs). We did detect CD4+ T cells within the epithelium at sites of infection, in addition to being present in the dermis, and keratinocytes can express class II molecules under certain inflammatory conditions, e.g. in response to (IFN)-γ [44]. However, we failed to detect MHC class II expression on keratinocytes at sites of infection, both when the infection was being controlled by the immune system and when it was not (unpublished data). Therefore we currently favor the hypotheses that CD4+ T cells are activated by cross-presentation of viral antigens by professional antigen presenting cells, and PV infection is controlled via soluble factors produced by the CD4+ T cells.
Although the details of immune recognition of MusPV1 remain to be determined, it seems likely that T cells control MusPV1 infection by both innate and adaptive immune mechanisms. The timing of papilloma regression after suspending CsA treatment and of spontaneous regression after high dose challenge in the SENCAR mice is consistent with induction of an antigen-specific adaptive response. However, the early control viral gene transcription in C57BL/6 mice at week 1–2, that is maintained in the CD8 KO mice but lost in the CD4 KO mice (Figure 6E), strongly suggests that CD4-mediated innate immune responses also plays a role in controlling infection.
Our findings raise several additional interesting issues. One is what may account for the different requirements for protection in C57BL/6 and SENCAR by CD4+ and CD8+ T cells. One possibility is that both the CD4+ and the CD8+ T cells in C57BL/6 are more potent in their ability to confer resistance than in SENCAR, making the presence of one of the T cell subsets sufficient to keep C57BL/6 resistant. This situation could arise if there were a single factor common to both T cell subsets that is more potent in C57BL/6 than in SENCAR, or if there were separate CD4+-specific and CD8+-specific factors. Alternatively, there could be a factor that, although it is extrinsic to CD4+ and CD8+ T cells, can cooperate with either subset in making a mouse resistant to papillomatosis. The observed strain-dependent difference could then be explained if the activity of this putative non-CD4/non-CD8 factor(s) is sufficiently more potent in C57BL/6 than in SENCAR that it can cooperate with CD4+ or CD8+ T cells to confer resistance in C57BL/6 but not in SENCAR.
A second issue is the partial correlation between the reported strain-dependent susceptibility to chemical-induced papillomatosis and the robustness of persistent infection and papillomatosis after CsA treatment found in this study. The observed partial correlations may be due to yet undetermined strain-specific genetic factors allowing for control of skin tumor susceptibility/resistance. The effects of immunosuppression on 7,12-dimethylbenz[a]anthracene (DMBA)/phorbol ester 12-O-tetradecanoylphorbol 13-acetate (TPA) tumorigenesis have been studied to a limited degree. In C3H/HeN, compared with wt mice, CD8 KO mice had an increased number of tumors, but numbers were decreased in CD4 KO mice [45]. In contrast to C3H/HeN, in FVB/N mice, CD8 KO mice had a decreased number of tumors [46]. It might be informative to examine the role of CD4+ and CD8+ cells in DMBA/TPA tumorigenesis in C57BL/6.
A third issue is whether the characteristics of MusPV1 infection documented herein make it an attractive model of any HPVs. At present MusPV1 infection appears to most closely resemble that of genus beta HPVs. These cutaneous HPVs have been found in immunocompetent individuals in clinically normal skin and plucked hairs where they predominantly induce asymptomatic skin infections [47], [48], similar to MusPV1 infection described herein. However, after immunosuppression, beta HPVs could be more frequently detected, tended to display higher viral loads, and most importantly induced visible lesions after immunosuppression [49]–[51]. The presence of beta-HPVs has been implicated as a causal factor in the increased risk for development of non-melanoma skin cancers in immunosuppressed transplant recipients. Thus, MusPV1 infection may provide a model to study the impact of a cutaneous PV type on the pathogenesis of non-melanoma skin cancer in an immunosuppressed and in an immunocompetent setting.
The following mice (H-2 haplotypes are given in parentheses) were obtained from the National Institutes of Health or The Jackson Laboratories (Bar Harbor, MN): immunocompetent outbred Cr:ORL SENCAR; immunocompetent inbred strains FVB/NCr (H2q), BALB/cAnNCr (H2d), DBA/2NCr (H2d), A/JCr (H2a), C57BL/6NCr (C57BL/6) (H2b), 129S6/SvEv (H2b), C3H/HeJCr (H2k); immunodeficient athymic NCr nu/nu; CD4-, CD8-, CD1d-deficient C57BL/6.
The mice, all females aged 6–10 weeks, were housed and handled in strict accordance to the National Institutes of Health guidelines for the use and care of live animals. Experimental protocols were approved by the National Cancer Institute's Animal Care and Use Committee (Permit Number LCO 027).
Crude extracts of papillomatous tissues were prepared as previously described [9], and MusPV1 virions purified from the extracts by Optiprep gradient centrifugation as detailed on the laboratory website (http://home.ccr.cancer.gov/Lco) [52]. In the purified preparations, MusPV1 viral copy numbers were quantified by real-time PCR [9] after liberation of encapsidated DNA from the viral capsids with proteinase K. The presence of the MusPV1 major capsid protein L1 in the purified fractions or in the crude tissue extracts was determined by Western blot using a polyclonal rabbit immune serum raised against MusPV1 L1 virus-like particles, at a dilution of 1∶1000 [9].
In vivo infection with purified MusPV1 virions was performed on pre-scarified skin of the animals' tails as previously published [9], [13]. The tail was chosen, as it represents a location that is highly permissive for MusPV1 infection [9]. The tail also has the advantage over the equally permissive muzzle skin, in that extensive lesional growth does not cause obvious distress to the animals. The skin on the animals' backs was not tested due to its minimal susceptibility to MusPV1 infection in athymic NCr nude mice using the same technique of inoculation [9].
For systemic immunosuppression, CsA (Sandimmune Inject., Novartis) was diluted under sterile conditions with PBS and administered subcutaneously to the animals five times per week at a dose of 75 mg/kg body weight in a volume of 0.1 ml. Treatment was started one week prior to infection and maintained for additional 4 weeks post-infection for a total of 5 weeks.
In vivo depletion of T cells was achieved by intraperitoneal administration of mAbs (all BioXCell): anti-mouse CD4 (clone GK1.5), anti-mouse CD8a (clone 53-6.72), anti-mouse CD3 (clone 17A2) in a dose of 0.5 mg per mouse in a volume of 0.1 ml. Rat IgG2b (clone LTF-2) or rat IgG2a (clone 2A3) mAbs were used as appropriate isotype controls. Depletion was performed on three consecutive days starting at indicated time points and the depleted state maintained by administration of mAbs twice a week for a period of 7 weeks. Depletion was verified by flow cytometry analyses as described below.
The CsA experiments (n = 4–5 animals per experimental group) were repeated twice for FVB/NCr, A/JCr and C3H/HeJCr mice, three times for BALB/cAnNCr and C57BL/6 mice, and 17 times for Cr:ORL SENCAR. The experiments using DBA/2NCr and 129S6/SvEv mice were tested only once as these haplotypes (H2d and H2b) were already represented by BALB/cAnNCr and C57BL/6 mice, respectively. The depletion experiments in Cr:ORL SENCAR mice (n = 9 per group) and in C57BL/6 mice (n = 5 per group) were repeated twice. The experiments employing C57BL/6 KO mice (n = 5 per group) were repeated twice.
For detection of MusPV1 E1∧E4 spliced transcripts as a measure for initial infection, total RNA was isolated from tail skin necropsies using TRI Reagent (Molecular Research Center Inc.), treated with DNAse I (Qiagen), and reverse-transcribed into cDNA using the SuperScript III First-Strand Synthesis System (Invitrogen), following the manufacturers' instructions. Real-time PCR using primers and probe specific for MusPV1 E1∧E4 spliced transcripts was performed in an ABI PRISM 7900HT Sequence Detection System, as previously reported [9]. The results were correlated to the endogenous control, beta-actin (LifeTechnologies), in the same samples. Quantification of MusPV1 genome copy numbers were performed using MusPV1-forward and MusPV1-reverse primers and compared to defined amounts of re-ligated MusPV1 genome as standards, as described previously [9].
Skin necropsies were snap frozen in Tissue-Tek OCT Compound freezing medium (Sakura Finetek USA Inc.) and HE- and IFS performed on ethanol-fixed tissue sections of 6 µm thickness [9], [13]. For detection of MusPV1 L1 protein, sections were stained with a rabbit polyclonal immune serum directed against MusPV1 L1 at a dilution of 1∶4000 and detected with either an Alexa Fluor 488 or an Alexa Fluor 594-conjugated donkey anti-rabbit secondary antibody (both Life Technologies), as indicated. Co-stainings with either directly conjugated Alexa Fluor 488-anti-mouse CD4 or Alexa Fluor 488-anti-mouse CD8a antibodies (both Biolegend; dilution 1∶100) were performed to detect CD4+ or CD8+ T cells, respectively. To determine localization of MusPV1 L1 in relation to basal keratinocytes, sections were co-stained with a phycoerythrin-conjugated anti-CD49f antibody (integrin alpha 6, BD Biosciences). Nuclei were visualized by mounting sections with ProLong Gold antifade reagent containing 4′,6-diamidino-2-phenylindole (DAPI) (LifeTechnologies). All microscopy analyses were performed on a Zeiss LSM 510 UV system and color levels of images were processed equally in Adobe Photoshop across experiments.
Spleens, iliac and inguinal lymph nodes, and tail skin were harvested from CO2 euthanized mice. To obtain single cell suspensions, spleens and lymph nodes were enzymatically digested with DNAse I (0.2 mg/ml; Roche) with and without Collagenase A (0.5 mg/ml; Roche), respectively, and processed as described previously [53]. Skin tissues were cut into fine pieces and incubated with collagenase IV (2 mg/ml; Worthington Biochemical Corp.) plus DNAse I (0.1 mg/ml) in RPMI 1640, supplemented with 10% fetal bovine serum and 1% penicilin/streptomycin, for 1 hr at 37°C. All cell suspensions were passed through 70 µM nylon mesh filters (BD Falcon) prior to lysis of erythrocytes with Ammonium-Chloride-Potassium Buffer (ACK; Lonza).
Blood was collected in heparinized tubes and erythrocytes removed by ACK lysis.
After blocking of Fc receptors by incubation with anti-mouse CD16/CD32 antibody (BD Pharmingen), surface-stainings were performed on single cell suspensions from tissues and blood using anti-mouse CD4-PerCP/Cy5.5 (clone RM 4-5; BD Pharmingen) and anti-mouse CD8a-FITC labeled antibodies (clone YTS; Abcam). Cells were fixed with Cytofix/Cytoperm (BD Biosciences) and acquisitions of flow cytometric data performed on a FACSCanto with FACSDiva software (BD Biosciences) [53]. The FlowJo software was used for analyses.
Statistical analyses (Mann Whitney tests) were performed using the GraphPad Prism Software 6.00 for Windows.
|
10.1371/journal.pcbi.1000254 | Game Theory of Mind | This paper introduces a model of ‘theory of mind’, namely, how we represent the intentions and goals of others to optimise our mutual interactions. We draw on ideas from optimum control and game theory to provide a ‘game theory of mind’. First, we consider the representations of goals in terms of value functions that are prescribed by utility or rewards. Critically, the joint value functions and ensuing behaviour are optimised recursively, under the assumption that I represent your value function, your representation of mine, your representation of my representation of yours, and so on ad infinitum. However, if we assume that the degree of recursion is bounded, then players need to estimate the opponent's degree of recursion (i.e., sophistication) to respond optimally. This induces a problem of inferring the opponent's sophistication, given behavioural exchanges. We show it is possible to deduce whether players make inferences about each other and quantify their sophistication on the basis of choices in sequential games. This rests on comparing generative models of choices with, and without, inference. Model comparison is demonstrated using simulated and real data from a ‘stag-hunt’. Finally, we note that exactly the same sophisticated behaviour can be achieved by optimising the utility function itself (through prosocial utility), producing unsophisticated but apparently altruistic agents. This may be relevant ethologically in hierarchal game theory and coevolution.
| The ability to work out what other people are thinking is essential for effective social interactions, be they cooperative or competitive. A widely used example is cooperative hunting: large prey is difficult to catch alone, but we can circumvent this by cooperating with others. However, hunting can pit private goals to catch smaller prey that can be caught alone against mutually beneficial goals that require cooperation. Understanding how we work out optimal strategies that balance cooperation and competition has remained a central puzzle in game theory. Exploiting insights from computer science and behavioural economics, we suggest a model of ‘theory of mind’ using ‘recursive sophistication’ in which my model of your goals includes a model of your model of my goals, and so on ad infinitum. By studying experimental data in which people played a computer-based group hunting game, we show that the model offers a good account of individual decisions in this context, suggesting that such a formal ‘theory of mind’ model can cast light on how people build internal representations of other people in social interactions.
| This paper is concerned with modelling the intentions and goals of others in the context of social interactions; in other words, how do we represent the behaviour of others in order to optimise our own behaviour? Its aim is to elaborate a simple model of ‘theory of mind’ [1],[2] that can be inverted to make inferences about the likely strategies subjects adopt in cooperative games. Critically, as these strategies entail inference about other players, this means the model itself has to embed inference about others. The model tries to reduce the problem of representing the goals of others to its bare essentials by drawing from optimum control theory and game theory.
We consider ‘theory of mind’ at two levels. The first concerns how the goals and intentions of another agent or player are represented. We use optimum control theory to reduce the problem to representing value-functions of the states that players can be in. These value-functions prescribe optimal behaviours and are specified by the utility, payoff or reward associated with navigating these states. However, the value-function of one player depends on the behaviour of another and, implicitly, their value-function. This induces a second level of theory of mind; namely the problem of inference on another's value-function. The particular problem that arises here is that inferring on another player who is inferring your value-function leads to an infinite regress. We resolve this dilemma by invoking the idea of ‘bounded rationality’ [3],[4] to constrain inference through priors. This subverts the pitfall of infinite regress and enables tractable inference about the ‘type’ of player one is playing with.
Our paper comprises three sections. The first deals with a theoretical formulation of ‘theory of mind’. This section describes the basics of representing goals in terms of high-order value-functions and policies; it then considers inferring the unknown order of an opponent's value-function (i.e., sophistication or type) and introduces priors on their sophistication that finesse this inference. In the second section, we apply the model to empirical behavioural data, obtained while subjects played a sequential game, namely a ‘stag-hunt’. We compare different models of behaviour to quantify the likelihood that players are making inferences about each other and their degree of sophistication. In the final section, we revisit optimisation of behaviour under inferential theory of mind and note that one can get exactly the same equilibrium behaviour without inference, if the utility or payoff functions are themselves optimised. The ensuing utility functions have interesting properties that speak to a principled emergence of ‘inequality aversion’ [5] and ‘types’ in social game theory. We discuss the implications of this in the context of evolution and hierarchical game theory.
Here, we describe the optimal value-function from control theory, its evaluation in the context of one agent and then generalise the model for interacting agents. This furnishes models that can be compared using observed actions in sequential games. These models differ in the degree of recursion used to construct one agent's value-function, as a function of another's. This degree or order is bounded by the sophistication of agents, which determines their optimum strategy; i.e., the optimum policy given the policy of the opponent. Note that we will refer to the policy on the space of policies as a strategy and reserve policy for transitions on the space of states. Effectively, we are dealing with a policy hierarchy where we call a second-level policy a strategy. We then address inference on the policy another agent is using and optimisation under the implicit unobservable states. We explore these schemes using a stag-hunt, a game with two Nash equilibria, one that is risk-dominant and another that is payoff-dominant. This is important because we show that the transition from one to the other rests on sophisticated, high-order representations of an opponent's value-function.
Let the admissible states of an agent be the set , where the state at any time or trial is . We consider environments under Markov assumptions, where is the probability of going from state j to state i. This transition probability defines the agent's policy as a function of value . We can summarise this policy in terms of a matrix , with elements . In what follows, will use to denote a probability transition matrix that depends on and for a probability on . The value of a state is defined as utility or payoff, expected under iterations of the policy and can be defined recursively as(1)The notion of value assumes the existence of a state-dependent quantity that the agent optimises by moving from one state to another. In Markov environments with n = |S| states, the value over states, encoded in the row vector v∈ℜ1×n, is simply the payoff at the current state ℓ∈ℜ1×n plus the payoff expected on the next move, ℓP, the subsequent move ℓP2 and so on. In short, value is the reward expected in the future and satisfies the Bellman equation [6] from optimal control theory; this is the standard equation of dynamic programming(2)We will assume a policy is fully specified by value and takes the form(3a)Under this assumption, value plays the role of an energy function, where λ is an inverse temperature or precision; assumed to take a value of one in the simulations below. Using the formalism of Todorov [7], the matrix P(0) encodes autonomous (uncontrolled) transitions that would occur when, . These probabilities define admissible transitions and the nature of the state-space the agent operates in, where inadmissible transitions are encoded with P(0)ij = 0. The uncontrolled transition probability matrix P(0) plays an important role in the general setting of Markov decision processes (MDP). This is because certain transitions may not be allowed (e.g., going though a wall). Furthermore, there may be transitions, even in the absence of control, which the agent is obliged to make (e.g., getting older). These constraints and obligatory transitions are encoded in P(0). The reader is encouraged to read Ref. [7] for a useful treatment of optimal control problems and related approximation strategies.
Equation 3a is intuitive, in that admissible states with relatively high value will be visited with greater probability. Under some fairly sensible assumptions about the utility function (i.e., assuming a control cost based on the divergence between controlled and uncontrolled transition probabilities), Equation 3 is the optimum policy.
This policy connects our generative model of action to economics and behavioural game theory [8], where the softmax or logit function (Equation 3) is a ubiquitous model of transitions under value or attraction; for example, a logit response rule is used to map attractions, to transition probabilities:(3b)In this context, λ is known as response sensitivity; see Camerer [8] for details. Furthermore, a logit mapping is also consistent with stochastic perturbations of value, which leads to quantal response equilibria (QRE). QRE are a game-theoretical formulation [9], which converges to the Nash equilibrium when λ goes to infinity. In most applications, it is assumed that perturbations are drawn from an extreme value distribution, yielding the familiar and convenient logit choice probabilities in Equation 3 (see [10] for details). Here, λ relates to precision of random fluctuations on value.
Critically, Equation 3 prescribes a probabilistic policy that is necessary to define the likelihood of observed behaviour for model comparison. Under this fixed-form policy, the problem reduces to optimising the value-function (i.e., solving the nonlinear self-consistent Bellman equations). These are solved simply and quickly by using a Robbins-Monro or stochastic iteration algorithm [11](4)At convergence, becomes the optimal value-function, which is an analytic function of payoff; . From now on, we will assume is the solution to the relevant Bellman equation. This provides an optimum value-function for any state-space and associated payoff, encoded in a ‘game’.
Clearly, this is not the only way to model behaviour. However, the Todorov formalism greatly simplifies the learning problem and provides closed-form solutions for optimum value: In treatments based on Markov decision processes, in which the state transition matrix depends on an action, both the value-function and policy are optimised iteratively. However, by assuming that value effectively prescribes the transition probabilities (Equation 3), we do not have to define ‘action’ and avoid having to optimise the policy per se. Furthermore, as the optimal value is well-defined we do not have to worry about learning the value-function. In other words, because the value-function can be derived analytically from the loss-function (irrespective of the value-learning scheme employed by the agent), we do not need to model how the agent comes to acquire it; provided it learns the veridical value-function (which in many games is reasonably straightforward). This learning could use dynamic programming [12], or Q-learning [13], or any biologically plausible scheme.
The example in Figure 1 illustrates the nature and role of the quantities described above. We used a one-dimensional state-space with n = 16 states, where an agent can move only to adjacent states (Figure 1A). This restriction is encoded in the uncontrolled transition probabilities. We assumed the agent is equally likely to move, or not move, when uncontrolled; i.e., the probability of remaining in a state is equal to the sum of transitions to other states (Figure 1B). To make things interesting, we considered a payoff function that has two maxima; a local maximum at state four and the global maximum at state twelve (Figure 1C). In effect, this means the optimum policy has to escape the local maximum to reach the global maximum. Figure 1D shows the successive value-function approximations as Equation 4 is iterated from τ = 1 to 32. Initially, the local maximum captures state-trajectories but as the value-function converges to the optimal value-function, it draws paths through the local maximum, toward the global maximum. Instead of showing example trajectories under the optimal value-function, we shows the density of an ensemble of agents, ρ(s,t), as a function of time, starting with a uniform distribution on state-space, ρ(s,0) = 1/n (Figure 1E). The ensemble density dynamics are given simply by . It can be seen that nearly all agents have found their goal by about t = 18 ‘moves’.
In summary, we can compute an optimal value-function for any game, G(ℓ,P(0)) specified in terms of payoffs and constraints. This function specifies the conditional transition probabilities that define an agent's policy, in terms of the probability of emitting a sequence of moves or state-transitions. In the next section, we examine how value-functions are elaborated when several agents play the same game.
When dealing with two agents the state-space becomes the Cartesian product of the admissible states of both agents, S = S1×S2 (Note that all that follows can be extended easily to over m agents.). This means that the payoff and value are defined on a joint-space for each agent k. The payoff for the first agent ℓ1(i, j) occurs when it is in state i and the second is in state j. This can induce cooperation or competition, unless the payoff for one agent does not depend on the state of the other: i.e., ∀j,k : ℓ1(i, j) = ℓ1(i, k). Furthermore, the uncontrolled probabilities for one agent now become a function of the other agent's value, because one agent cannot control the other. This presents an interesting issue of how one agent represents the policy of the other.
In what follows, we consider policies that are specified by an order: first-order policies discount the policies of other agents (i.e., I will ignore your goals). Second-order policies are optimised under the assumption that you are using a first-order policy (i.e., you are ignoring my goals). Third-order policies pertain when I assume that you assume I am using a first-order policy and so on. This construction is interesting, because it leads to an infinite regress: I model your value-function but your value-function models mine, which includes my model of yours, which includes my model of your model of mine and so on ad infinitum. We will denote the i-th order value-function for the k-th agent by . We now consider how to compute these value-functions.
In a sequential game, each agent takes a turn in a fixed order. Let player one move first. Here, the transition probabilities now cover the Cartesian product of the states of both agents and the joint transition-matrix factorises into agent-specific terms. These are given by(5)where Πk(0) specifies uncontrolled transitions in the joint-space, given the uncontrolled transitions Pk(0) in the space of the k-th agent. Their construction using the Kronecker tensor product ⊗ ensures that the transition of one agent does not change the state of the other. Furthermore, it assumes that the uncontrolled transitions of one agent do not depend on the state of the other; they depend only on the uncontrolled transitions Pk(0) among the k-th agent's states. The row vectors are the vectorised versions of the two dimensional value-functions for the k-th agent, covering the joint states. We will use a similar notation for the payoffs, . Critically, both agents have a value-function on every joint-state but can only change their own state. These value-functions can now be evaluated through recursive solutions of the Bellman equations(6)This provides a simple way to evaluate the optimal value-functions for both agents, to any arbitrary order. The optimal value-function for the first agent, when the second is using is . Similarly, the optimal value under for the second is . It can be seen that under an optimum strategy (i.e., a second-level policy) each agent should increase its order over the other until a QRE obtains when for both agents. However, it is interesting to consider equilibria under non-optimal strategies, when both agents use low-order policies in the mistaken belief that the other agent is using an even lower order. It is easy to construct examples where low-order strategies result in risk-dominant policies, which turn into payoff-dominant policies as high-order strategies are employed; as illustrated next.
In this example, we used a simple two-player stag-hunt game where two hunters can either jointly hunt a stag or pursue a rabbit independently [14]. Table 1 provides the respective payoffs for this game as a normal form representation. If an agent hunts a stag, he must have the cooperation of his partner in order to succeed. An agent can catch a rabbit by himself, but a rabbit is worth less than a stag. This furnishes two pure-strategy equilibria: one is risk-dominant with low-payoff states that can be attained without cooperation (i.e., catching a rabbit) and the other is payoff dominant; high-payoff states that require cooperation (i.e., catching a stag). We assumed the state-space of each agent is one-dimensional with n1 = n2 = 16 possible states. This allows us to depict the value-functions on the joint space as two-dimensional images. The dimensionality of the state-space is not really important; however, a low-dimensional space imposes sparsity on the transition matrices, because only a small number of neighbouring states can be visited from any given state. These constraints reduce the computational load considerably. The ‘rabbit’ and ‘stag’ do not move; the rabbit is at state four and the stag at state twelve. The key difference is that the payoff for the ‘stag’ is accessed only when both players occupy that state (or nearby), whereas the payoff for the ‘rabbit’ does not depend on the other agent's state. Figure 2A shows the characteristic payoff functions for both agents. The ensuing value-functions for the order i = 1,…,4 from Equation 6 are shown in Figure 2B. It can be seen that first-order strategies defined by regard the ‘stag’ as valuable, but only when the other agent is positioned appropriately. Conversely, high-order strategies focus exclusively on the stag. As one might intuit, the equilibrium densities of an ensemble of agents acting under first or high-order strategies have qualitatively different forms. Low-order strategies result in both agents hunting the ‘rabbit’ and high-order schemes lead to a cooperative focus on the ‘stag’. Figure 2C shows the joint and marginal equilibrium ensemble densities for t = 128 (i.e., after 128 moves) and a uniform starting distribution; for matched strategies, i = 1,…,4.
In contrast to single-player games, polices in multi-player games have an order, where selecting the optimal order depends on the opponent. This means we have to consider how players evaluate the probability that an opponent is using a particular policy or how we, as experimenters, make inferences about the policies players use during sequential games. This can be done using the evidence for a particular policy, given the choices made. In the course of a game, the trajectory of choices or states y = s1,s2,…,sT is observed directly such that, under Markov assumptions(7)Where m∈M represents a model of the agents and entails the quantities needed to specify their policies. The probability of a particular model, under flat priors on the models, is simply(8)To illustrate inference on strategy, consider the situation in which the strategy (i.e., the policy order k1) of the first agent is known. This could be me and I might be trying to infer your policy, to optimise mine; or the first agent could be a computer and the second a subject, whose policy we are trying to infer experimentally. In this context, the choices are the sequence of joint-states over trials, y∈S, where there are n1×n2 possible states; note that each joint state subsumes both ‘moves’ of each agent. From Equation 8 we can evaluate the probability of the second agent's strategy, under the assumption it entails a fixed and ‘pure’ policy of order k2(9)Here, the model is specified by the unknown policy order, m = k2 of the second agent. Equation 9 uses the joint transition probabilities on the moves of all players; however, one gets exactly the same result using just the moves and transition matrix from the player in question. This is because, the contributions of the other players cancel, when the evidence is normalised. We use the redundant form in Equation 9 so that it can be related more easily to inference on the joint strategies of all agents in Equation 8. An example of this inference is provided in Figure 3. In Figure 3A and 3B, we used unmatched and matched strategies to generate samples using the probability transition matrices and ; starting in the first state (i.e., both agents in state 1) respectively. These simulated games comprised four consecutive 32-move trials of the stag-hunt game specified in Figure 2. The ensuing state trajectories are shown in the left panels. We then inverted the sequence using Equation 9 and a model-space of . The results for T = 1,…,128 are shown in the right panels. For both simulations, the correct strategy discloses itself after about sixty moves, in terms of conditional inference on the second agent's policy. It takes this number of trials because, initially, the path in joint state-space is ambiguous; as it moves towards both the rabbit and stag.
We have seen how an N-player game is specified completely by a set of utility functions and a set of constraints on state-transitions. These two quantities define, recursively, optimal value-functions, of increasing order and their implicit policies. Given these policies, one can infer the strategies employed by agents, in terms of which policies they are using, given a sequence of transitions. In two-player games, when the opponent uses policy k, the optimum strategy is to use policy k+1. This formulation accounts for the representation of another's goals and optimising both policies and strategies. However, it induces a problem; to optimise ones own strategy, one has to know the opponent's policy. Under rationality assumptions, this is not really a problem because rational players will, by induction, use policies of sufficiently high order to ensure . This is because each player will use a policy with an order that is greater than the opponent and knows a rational opponent will do the same. The interesting issues arise when we consider bounds or constraints on the strategies available to each player and their prior expectations about these constraints.
Here, we deal with optimisation under bounded rationality [4] that obliges players to make inferences about each other. We consider bounds, or constraints, that lead to inference on the opponent's strategy. As intimated above, it is these bounds that lead to interesting interactions between players and properly accommodate the fact that real players do not have unbounded computing resources to attain a QRE by using . These constraints are formulated in terms of the policy ki of the i-th player, which specifies the corresponding value-function and policy . The constraints we consider are:
These assumptions have a number of important implications. First, because qi(kj) is a point mass at the mode , each player will assume every other player is using a pure strategy, as opposed to a strategy based on a mixture of value-functions. Second, under this assumption, each player will respond optimally with another pure strategy, . Third, because there is an upper bound on imposed by an agent's priors, they will never call upon strategies more sophisticated than ki = Ki+1. In this way, Ki bounds both the prior assumptions about other players and the sophistication of the player per se. This defines a ‘type’ of player [15] and is the central feature of the bounded rationality under which this model is developed. Critically, type is an attribute of a player's prior assumptions about others. The nature of this bound means that any player cannot represent the goals or intentions of another player who is more sophisticated; in other words, it precludes any player ‘knowing the mind of God’ [16].
Under flat priors on the bounded support of the priors pi(kj), the mode can be updated with each move using Equation 9. Here, player one would approximate the conditional density on the opponent's strategy with the mode(10)And optimise its strategy accordingly, by using . This scheme assumes the opponent uses a fixed strategy and consequently accumulates evidence for each strategy over the duration of the game. Figure 4 illustrates the conditional dependencies of the choices and strategies; it tries to highlight the role of the upper bounds in precluding recursive escalation of ki(t). Note, that although each player assumes the other is using a stationary strategy, the players own policy is updated after every move.
Figure 5A shows a realization of a simulated stag-hunt using two types of player with asymmetric bounds K1 = 4 and K2 = 3 (both starting with ki(1) = 1). Both players strive for an optimum strategy using Equation 10. We generated four consecutive 32-move trials; 128 trials in total, starting in the first state with both agents in state one. After 20 moves, the first, more sophisticated, player has properly inferred the upper bound of the second and plays at one level above it. The second player has also optimised its strategy, which is sufficiently sophisticated to support cooperative play. The lower panels show the implicit density on the opponent's strategy, p(k2|y,k1); similarly for the second player. The mode of this density is in Equation 10.
We conclude this section by asking if we, as experimenters, can infer post hoc on the ‘type’ of players, given just their choice behaviours. This is relatively simple and entails accumulating evidence for different models in exactly the same way that the players do. We will consider fixed-strategy models in which both players use a fixed ki or theory of mind models, in which players infer on each other, to optimise ki(t) after each move. The motivation for considering fixed models is that they provide a reference model, under which the policy is not updated and therefore there is no need to infer the opponent's policy. Fixed models also relate to an alternative [prosocial] scheme for optimising behaviour, reviewed in the discussion. The evidence for fixed models is(11)Whereas the evidence for theory of mind models is(12)where are inferred under the appropriate priors specified by Ki. The key difference between these models is that the policy changes adaptively in the theory of mind model, in contrast to the fixed model.
Under flat model priors, the posterior, p(mi|y) (Equation 8) can be used for inference on model-space. We computed the posterior probabilities of fifty models, using Equation 11 and 12. Half of these models were fixed models using k1,k2 = 1,…,5 and the remaining were theory of mind models with K1,K2 = 0,…,4. Figure 5B shows the results of this model comparison using the simulated data shown in Figure 5A. We evaluated the posterior probability of theory of mind by marginalising over the bi-partition of fixed and theory of mind models, and it can be seen that the likelihood of the theory of mind model is substantially higher than the fixed model. Furthermore, the model with types K1 = 4 and K2 = 3 supervenes, yielding a 94.5% confidence that this is the correct model. The implicit densities used by the players on each others strategy p(k2|y,k1) and p(k1|y,k2) (see Equation 11) are exactly the same as in Figure 5A because the veridical model was selected.
Because we assumed the model is stationary over trials, the conditional confidence level increases with the number of trials; although this increase depends on the information afforded by the particular sequence. On the other hand, the posterior distribution over models tends to be flatter as the model-space expands because the difference between successive value-functions, and becomes smaller with increasing order. For the stag-hunt game in Figure 2, value-functions with k≥4 are nearly identical. This means that we could only infer with confidence that, Ki≥5 (see Figure S1).
In this section, we apply the modelling and inference procedures of the preceding section to behavioural data obtained while real subjects played a stag-hunt game with a computer. In this experiment, subjects navigated a grid maze to catch stags or rabbits. When successful, subjects accrued points that were converted into money at the end of the experiment. First, we inferred the model used by subjects, under the known policies of their computer opponents. This allowed us to establish whether they were using theory of mind or fixed models and, under theory of mind models, how sophisticated the subjects were. Using Equation 10 we then computed the subjects' conditional densities on the opponent's strategies, under their maximum a posteriori sophistication.
The subject's goal was to negotiate a two-dimensional grid maze in order to catch a stag or rabbit (Figure 6). There was one stag and two rabbits. The rabbits remained at the same grid location and consequently were easy to catch without help from the opponent. If one hunter moved to the same location as a rabbit, he/she caught the rabbit and received ten points. In contrast, the stag could move to escape the hunters. The stag could only be caught if both hunters moved to the locations adjacent to the stag (in a co-operative pincer movement), after which they both received twenty points. Note that as the stag could escape optimally, it was impossible for a hunter to catch the stag alone. The subjects played the game with one of two types of computer agents; A and B. Agent A adopted a lower-order (competitive) strategy and tried to catch a rabbit by itself, provided both hunters were not close to the stag. On the other hand, agent B used a higher-order (cooperative) strategy and chased the stag even if it was close to a rabbit. At each trial, both hunters and the stag moved one grid location sequentially; the stag moved first, the subject moved next, and the computer moved last. The subjects chose to move to one of four adjacent grid locations (up, down, left, or right) by pressing a button; after which they moved to the selected grid. Each move lasted two seconds and if the subjects did not press a key within this period, they remained at the same location until the next trial.
Subjects lost one point on each trial (even if they did not move). Therefore, to maximise the total number of points, it was worth trying to catch a prey as quickly as possible. The round finished when either of the hunters caught a prey or when a certain number of trials (15±5) had expired. To prevent subjects changing their behaviour, depending on the inferred number of moves remaining, the maximum number of moves was randomised for each round. In practice, this manipulation was probably unnecessary because the minimum number of moves required to catch a stag was at most nine (from any initial state). Furthermore, the number of ‘time out’ rounds was only four out of a total 240 rounds (1.7%). At the beginning of each round the subjects were given fifteen points, which decreased by one point per trial, continuing below zero beyond fifteen trials. For example, if the subject caught a rabbit on trial five, he/she got the ten points for catching the rabbit, plus the remaining time points: 10 = 15−5 points, giving 20 points in total, whereas the other player received only their remaining time points; i.e., 10 points. If the hunters caught a stag at trial eight, both received the remaining 7 = 15−8 time points plus 20 points for catching the stag, giving 27 points in total. The remaining time points for both hunters were displayed on each trial and the total number of points accrued was displayed at the end of each round.
We studied six (normal young) subjects (three males) and each played four blocks with both types of computer agent in alternation. Each block comprised ten rounds; so that they played forty rounds in total. The start positions of all agents; the hunters and the stag, were randomised on every round, under the constraint that the initial distances between each hunter and the stag were more than four grids points.
We applied our theory of mind model to compute the optimal value-functions for the hunters and stag. As hunters should optimise their strategies based not only on the other hunter's behaviour but also the stag's, we modelled the hunt as a game with three agents; two hunters and a stag. Here state-space became the Cartesian product of the admissible states of all agents, and the payoff was defined on a joint space for each agent; i.e., on a |S1|×|S2|×|S3| array. The payoff for the stag was minus one when both hunters were at the same location as the stag and zero for the other states. For the hunters, the payoff of catching a stag was one and accessed only when both the hunters' states were next to the stag. The payoff for catching a rabbit was one half and did not depend on the other hunter's state. For the uncontrolled transition probabilities, we assumed that all agents would choose allowable actions (including no-move) with equal probability and allowed co-occupied locations; i.e., two or more agents could be in the same state. Allowable moves were constrained by obstacles in the maze (see Figure 6).
We will refer to the stag, subject, and computer as the 1st, 2nd, and 3rd agent, respectively. The transition probability at each trial is . The i-th order value-function for the j-th agent, , was evaluated through recursive solutions of the Bellman equations by generalising Equation 6 to three players(13)Notice that the first agent's (stag's) value-function is fixed at first-order. This is because we assumed that the hunters believed, correctly, that the stag was not sophisticated. We used a convergence criterion of to calculate the optimal value-functions, using Equation 4. For simplicity, we assumed the sensitivity λ of each player was one. A maximum likelihood estimation of the subjects' sensitivities, using the observed choices from all subjects together, showed that the optimal value was λ = 1.6. Critically, the dependency of the likelihood on strategy did not change much with sensitivity, which means our inferences about strategy are fairly robust to deviations from λ = 1 (see Figure S2). When estimated individually for each subject, the range was 1.5≤λ≤1.8, suggesting our approximation was reasonable and enabled us to specify the policy for each value-function and solve Equation 13 recursively.
The ensuing optimal value-functions of the subject, , for i = 1,…,4 are shown in Figure 7. To depict the three-dimensional value-functions of one agent in two-dimensional state-space, we fixed the positions of the other two agents for each value-function. Here, we show the value-functions of the subject for three different positions of the computer and the stag (three examples of four value-functions of increasing order). The locations of the computer and stag are displayed as a red circle and square respectively. One can interpret these functions as encoding the average direction the subject would choose from any location. This direction is the one that increases value (lighter grey in the figures). It can be seen that the subject's policy (whether to chase a stag or a rabbit) depends on the order of value-functions and the positions of the other agents. The first-order policy regards the rabbits as valuable because it assumes that other agents move around the maze in an uncontrolled fashion, without any strategies, and are unlikely to help catch the stag. Conversely, if subjects account for the opponent's value-functions (i.e., using the second or higher order policies), they behave cooperatively (to catch a stag), provided the opponent is sufficiently close to the stag. Furthermore, with the highest order value-function, even if the other hunter is far away from the stag, the subject still tries to catch the stag (top right panel in Figure 7). For all orders of value-functions, the stag's value becomes higher than the rabbits', when the other hunter is sufficiently close to the stag (the middle row). However, interestingly, the policies here are clearly different; in the first-order function, value is higher for the states which are closer to the stag and the two states next to the stag have about the same value. Thus, if the subject was in the middle of the maze, he/she would move downward to minimize the distance to the stag. In contrast, in the second and higher-order functions, the states leading to the right of the stag are higher than the left, where the other hunter is. This is not because that the right side states are closer to another payoff, such as a rabbit. In fact, even when the other hunter is on the right side of the stag and very close to the rabbit, the states leading to the other (left) side are higher in the fourth-order function (bottom right panel). These results suggest that sophisticated subjects will anticipate the behaviour of other agents and use this theory of mind to compute effective ways to catch the stag, even if this involves circuitous or paradoxical behaviour.
Using these optimal value-functions, we applied the model comparison procedures above to infer the types of the subjects. We calculated the evidence for each subject acting under a fixed or theory of mind model using k2 = ksub = 1,…,8 and K2 = Ksub = 1,…,8 and data pooled from all their sessions. We used the true order of the other players' policies for the model comparison; i.e., k1 = kstag = 1 for the stag, k3 = kcom = 1 for the agent A and kcom = 5 for the agent B (Figure S3). Although, as mentioned above, these values do not affect inference on the subject's model. This entailed optimising ksub and Ksub with respect to the evidence, for fixed models(14a)and theory of mind models(14b)Figure 8A shows the normalized posterior probabilities over the sixteen models. It can be immediately seen that the theory of mind model has a higher likelihood than the fixed model. Under theory of mind models, we inferred the most likely sophistication level of the subjects was Ksub = 5. This is reasonable, because the subjects did not have to use policies higher than ksub = 6, given the computer agent policies never exceeded five. Among the fixed models, even though the likelihood was significantly lower, the optimal model, ksub = 6, was inferred.
Using the inferred sophistication of the subjects, Ksub = 5, we then examined the implicit conditional density on their opponent's policy using Equation 11. Figure 8B show a typical example from one subject. The upper panels show the actual policies used when playing agent A (the left panel) and agent B (the right panel) and the lower panels show the subject's densities on the opponent's strategies. For both computer agents, the subject has properly inferred the strategy of the agent and plays at a level above it; i.e., the subject behaved rationally. This is a pleasing result, in that we can quantify our confidence that subjects employ theory of mind to optimise their choices and, furthermore, we can be very confident that they do so with a high level of sophistication. In what follows, we relate our game theory of mind to related treatments in behavioural economics and consider the mechanisms that may underpin sophisticated behaviour.
Games with iterated or repeated play can differ greatly from one-shot games, in the sense that they engender a range of equilibria and can induce the notion of ‘reputation’, when there is uncertainty about opponents [17]. These games address important issues concerning how people learn to play optimally given recurrent encounters with their opponents. It has been shown that reputation formation can be formulated as a Bayesian updating of types to explain choices in repeated games with simultaneous moves [18],[19] and non-simultaneous moves [20]. An alternative approach to reputation formation is teaching [21]. In repeated games, sophisticated players often have an incentive to ‘teach’ their opponents by choosing strategies with poor short-run payoffs that will change what the opponents do; in a way that benefits the sophisticated player in the long run. Indeed, Camerer et al [22] showed that strategic teaching in their EWA model could select one of many repeated-game equilibria and give rise to reputation formation without updating of types. The crucial difference between these approaches is that in the type-based model, reputation is the attribute of a particular player, while in the teaching model, a strategy attains a reputation. In our approach, types are described in terms of bounds on strategy; the sophistication level. This contrasts with treatments that define types in terms of unobserved payoff functions, which model strategic differences using an attribute of the agent; e.g., normal or honest type.
Recursive or hierarchical approaches to multi-player games have been adopted in behavioural economics [23],[24] and artificial intelligence [25], in which individual decision policies systematically exploit embedded levels of inference. For instance, some studies have assumed that subject's decisions follow one of a small set of a priori plausible types, which include non-strategic and strategic forms. Under these assumptions, inference based on decisions in one-shot (non-iterated) games suggests that while policies may be heterogeneous, the level of sophistication may be equivalent to an approximate value of k; two or three. Camerer and colleagues [26] have suggested a ‘cognitive hierarchy’ model, in which subjects generate a form of cognitive hierarchy over each other's level of reciprocal thinking. In this model ‘k’ corresponds to the depth of tree-search, and when estimated over a collection of games such as the p-beauty game, yields values of around one and a half to two. Note that ‘steps of strategic thinking’ are not the same as the levels of sophistication in this paper. The sophistication addressed here pertains to the recursive representation of an opponent's goals, and can be applied to any iterated extensive form game. Despite this, studies in behavioural economics suggest lower levels of sophistication than ours. One reason for this may be that most games employed in previous studies have been one-shot games, which place less emphasis on planning for future interactions that rest on accurate models of an opponent's strategy.
In the current treatment, we are not suggesting that players actually compute their optimal strategy explicitly; or indeed are aware of any implicit inference on the opponent's policy. Our model is phenomenological and is designed to allow model comparison and predictions (under any particular model) of brain states that may encode the quantities necessary to optimize behaviour. It may be that the mechanisms of this optimization are at a very low level (e.g., at the level of synaptic plasticity) and have been shaped by evolutionary pressure. In other words, we do not suppose that subjects engage in explicit cognitive operations but are sufficiently tuned to interactions with con-specifics that their choice behaviour is sophisticated. We now pursue this perspective from the point of view of evolutionary optimization of the policies themselves.
Here, we revisit the emergence of cooperative equilibria and ask whether sophisticated strategies are really necessary. Hitherto, we have assumed that the utility functions ℓi are fixed for any game. This is fine in an experimental setting but in an evolutionary setting, ℓi may be optimised themselves. In this case, there is a fundamental equivalence between different types of agents, in terms of their choices. This is because exactly the same equilibrium behaviour can result from interaction between sophisticated agents with empathy (i.e., theory of mind) and unsophisticated agents with altruistic utility-functions. In what follows, we show why this is the case:
The recursive solutions for high-order value-functions in Equation 6 can be regarded as a Robbins-Monro scheme for optimising the joint value-functions over N players. One could regard this as optimising the behaviour of the group of players collectively, as opposed to optimising the behaviour of any single player. Once the joint value-functions have been optimized, such that , they satisfy the Bellman equations(15)However, these value-functions also satisfy(16)This rearrangement is quite fundamental because we can interpret as optimal utility-functions, under the assumption that neither player represents the goals of the other. In other words, if two unsophisticated players were endowed with optimal utility-functions, one would observe exactly the same value-functions and behaviour exhibited by two very sophisticated players at equilibrium. These optimal are trivial to compute, given the optimal value-functions from Equation 6; although this inverse reinforcement learning is not trivial in all situations (e.g., [27]). It is immediately obvious that the optimal utility from Equation 16 has a much richer structure than the payoff ℓi (Figure S4). Critically, states that afford payoff to the opponent now become attractive, as if ‘what is good for you is good for me’. This ‘altruism’ [28] arises because has become context-sensitive, and depends on the other player's payoff. An interesting example is when the optimised utility of state with a local payoff is greater when the opponent occupies states close to their payoff (see Figure S4). In other words, a payoff that does not depend on the opponent has less utility, when the opponent's payoff is low (c.f., guilt).
This sort of phenomenon has been associated with ‘inequity aversion’. Inequity aversion is the preference for fairness [29] or resistance to inequitable outcomes; and has been formulated in terms of context-sensitive utility functions. For example, Fehr and Schmidt [5] postulate that people make decisions, which minimize inequity and consider N individuals who receive payoffs ℓi. They then model the utility to the j-th player as(17)where α parameterises distaste for disadvantageous inequality and β parameterises the distaste for advantageous inequality. Although a compelling heuristic, this utility function is an ad hoc nonlinear mixture of payoffs and has been critiqued for its rhetorical nature [30]. An optimal nonlinear mixture is given by substituting Equation 15 into Equation 16 to give(18)These equalities express the optimal utility functions in terms of payoff and a ‘prosocial’ utility (the second terms), which allow unsophisticated agents to optimise their social exchanges. The prosocial utility of any state is simply the difference in value expected after the next move with a sophisticated, relative to an unsophisticated, opponent. Equation 15 might provide a principled and quantitative account of inequity aversion, which holds under rationality assumptions.
One might ask, what is the relevance of an optimised utility function for game theory? The answer lies in the hierarchal co-evolution of agents (e.g., [15],[31]), where the prosocial part of may be subject to selective pressure. In this context, the unit of selection is not the player but the group of payers involved in a game (e.g., a mother and offspring). In this context, optimising over a group of unsophisticated players can achieve exactly the same result (in terms of equilibrium behaviour) as evolving highly sophisticated agents with theory of mind (c.f., [32]). For example, in ethological terms, it is more likely that the nurturing behaviour of birds is accounted for by selective pressure on ℓ• than invoking birds with theory of mind. This speaks to ‘survival of the nicest’ and related notions of prosocial behaviour (e.g., [33],[34]). Selective pressure on prosocial utility simply means, for example, that the innate reward associated with consummatory behaviour is supplemented with rewards associated with nursing behaviour. We have exploited the interaction between innate and acquired value previously in an attempt to model the neurobiology of reinforcement learning [35].
In summary, exactly the same equilibrium behaviour can emerge from sophisticated players with theory of mind, who act entirely out of self-interest and from unsophisticated players who have prosocial altruism, furnished by hierarchical optimisation of their joint-utility function. It is possible that prosocial utility might produce apparently irrational behaviour, in an experimental setting, if it is ignored: Gintis [33] reviews the evidence for empirically identifiable forms of prosocial behaviour in humans, (strong reciprocity), that may in part explain human sociality. “A strong reciprocator is predisposed to cooperate with others and punish non co-operators, even when this behaviour cannot be justified in terms of extended kinship or reciprocal altruism”. In line with this perspective, provisional fMRI evidence suggests that altruism may not be a cognitive faculty that engages theory of mind but is hard-wired and inherently pleasurable, activating subgenual cortex and septal regions; structures intimately related to social attachment and bonding in other species [36]. In short, bounds on the sophistication of agents can be circumvented by endowing utility with prosocial components, in the context of hierarchical optimisation.
Critically, the equivalence between prosocial and sophisticated behaviour is only at equilibrium. This means that prosocially altruistic agents will adapt the same strategy throughout an iterated game; however, sophisticated agents will optimise their strategy on the basis of the opponent's behaviour, until equilibrium is attained. These strategic changes make it possible to differentiate between the two sorts of agents empirically, using observed responses. To disambiguate between theory of mind dependent optimisation and prosocial utility it is sufficient to establish that players infer on each other. This is why we included fixed models without such inference in our model comparisons of the preceding sections. In the context of the stag-hunt game examined here, we can be fairly confident that subjects employed inference and theory of mind.
Finally, it should be noted that, although a duality in prosocial and sophisticated equilibria may exist for games with strong cooperative equilibria, there may be other games in which this is less clearly the case; where sophisticated agents and unsophisticated altruistic agents diverge in their behaviour. For example, in some competitive games (such as Cournot duopolys and Stackelberg games), a (selfish) understanding the other players response to payoff (empathy) produces a very different policy than one in which that payoff is inherently (altruistically) valued.
This paper has introduced a model of ‘theory of mind’ (ToM) based on optimum control and game theory to provide a ‘game theory of mind’. We have considered the representations of goals in terms of value-functions that are prescribed by utility or rewards. We have shown it is possible to deduce whether players make inferences about each other and quantify their sophistication using choices in sequential games. This rests on comparing generative models of choices with and without inference. Model comparison was demonstrated using simulated and real data from a ‘stag-hunt’. Finally, we noted that exactly the same sophisticated equilibrium behaviour can be achieved by optimising the utility-function itself, producing unsophisticated but altruistic agents. This may be relevant ethologically in hierarchal game theory and co-evolution.
In this paper, we focus on the essentials of the model and its inversion using behavioural data, such as subject choices in a stag-hunt. Future work will try to establish the predictive validity of the model by showing a subject's type or sophistication is fairly stable across different games. Furthermore, the same model will be used to generate predictions about neuronal responses, as measured with brain imaging, so that we can characterise the functional anatomy of these implicit processes. In the present model, although players infer the opponent's level of sophistication, they assume the opponents are rational and that their strategies are pure and fixed. However, the opponent's strategy could be inferred under the assumption the opponent was employing ToM to optimise their strategy. It would be possible to relax the assumption that the opponent uses a fixed and pure strategy and test the ensuing model against the current model. However, this relaxation entails a considerable computational expense (which the brain may not be in a position to pay). This is because modeling the opponent's inference induces an infinite recursion; that we resolved by specifying the bounds on rationality. Having said this, to model things like deception, it will be necessary to model hierarchical representations of not just the goals of another (as in this paper) but the optimization schemes used to attain those goals by assuming agent's represent the opponent's optimization of a changing and possibly mixed strategy. This would entail specifying different bounds to finesse the ensuing infinite recursion. Finally, although QRE have become the dominant approach to modelling human behaviour in, e.g., auctions, it remains to be established that convergence is always guaranteed (c.f., the negative results on convergence of fictitious play to Nash equilibria).
Recent interest in the computational basis of ToM has motivated neuroimaging experiments that test the hypothesis that putative subcomponents of mentalizing might correlate with cortical brain activity, particularly in regions implicated in ToM by psychological studies [37],[38]. In particular, Hampton and colleagues [39] report compelling data that suggest decision values and update signals are indeed in represented in putative ToM regions. These parameters were derived from a model based on ‘fictitious play’, which is a simple, non-hierarchical learning model of two-player inference. This model provided a better account of choice behaviour, relative to error-based reinforcement learning alone; providing support for the notion that apparent ToM behaviour arises from more than prosocial preferences alone. Clearly, neuroimaging offers a useful method for future exploration of whether key subcomponents of formal ToM models predict brain activity in ToM regions and may allow one to adjudicate between competing accounts.
|
10.1371/journal.pntd.0006494 | Molecular, immunological and neurophysiological evaluations for early diagnosis of neural impairment in seropositive leprosy household contacts | Household contacts constitute the highest risk group for leprosy development, and despite significant progress in the disease control, early diagnosis remains the primary goals for leprosy management programs.
We have recruited 175 seropositive and 35 seronegative household contacts from 2014 to 2016, who were subjected to an extensive protocol that included clinical, molecular (peripheral blood qPCR, slit-skin smear qPCR, skin biopsy qPCR) and electroneuromyographic evaluations.
The positivity of peripheral blood qPCR of seropositive contacts was 40.6% (71/175) whereas only 8.6% (3/35) were qPCR positive in seronegative contacts (p = 0.0003). For the slit-skin smear, only 4% (7/175) of seropositive contacts presented positive bacilloscopy, whereas the qPCR detected 47.4% (83/175) positivity in this group compared with only 17.1% (6/35) in seronegative contacts (p = 0.0009). In the ENMG evaluation of contacts, 31.4% (55/175) of seropositives presented some neural impairment, and 13.3% (4/35) in seronegatives (p = 0.0163). The presence of neural thickening conferred a 2.94-fold higher chance of ENMG abnormality (p = 0.0031). Seropositive contacts presented a 4.04-fold higher chance of neural impairment (p = 0.0206). The peripheral blood qPCR positivity presented odds 2.08-fold higher towards neural impairment (OR, 2.08; p = 0.028). Contrarily, the presence of at least one BCG vaccine scar demonstrated 2.44-fold greater protection against neural impairment (OR = 0.41; p = 0.044).
ELISA anti-PGL-I is the most important test in determining the increased chance of neural impairment in asymptomatic leprosy household contacts. The combination of the two assays (ELISA anti-PGL-I and peripheral blood qPCR) and the presence of BCG scar may identify individuals with higher chances of developing leprosy neuropathy, corroborating with the early diagnosis and treatment.
| Despite the apparent progress observed in recent years in leprosy control, early identification of cases remains one of the primary objectives of control programs. In addition, the failure of the current therapeutic scheme on the incidence of leprosy demonstrates that the disease elimination as a public health program promoted by the World Health Organization (WHO) depends on an incisive action to interrupt its transmission chain. The long incubation period of leprosy, its insidious symptoms and signs may difficult its diagnosis. Several studies have recently demonstrated that IgM anti-PGL-I seropositive contacts present higher chances to become ill than seronegative ones. Therefore, our question was: do seropositive contacts at greater risk of becoming sick already present subclinical neural damage? Therefore, our approach was to analyse anti-PGL-I seropositive contacts through electroneuromyography. The development and implementation of more specific and sensitive methods for the detection of M. leprae and its neural impairment, using immunological, molecular and neurophysiological tools are mandatory to increase the knowledge of leprosy epidemiology, to break its chain of transmission, thereby enabling effective control of this disease. This report demonstrated that seropositive contacts is the population group with higher chances of neural impairment.
| Leprosy is a chronic infectious disease caused by Mycobacterium leprae (M. leprae), an obligate intracellular parasite with a predilection for infecting peripheral nerves and skin. Leprosy is a current and challenging disease, because it still represents a problem for public health in developing countries such as Brazil, which ranks second worldwide in the number of new cases [1].
The predominance of multibacillary (MB) cases with neural disabilities indicates late diagnosis, reinforcing the ineffective epidemiological control in many countries [2]. In addition, new cases not only with high functional impairment, but also in children, reflect failure of early leprosy detection and indicate ongoing transmission [3,4,5].
Leprosy contacts of MB patients present a risk towards leprosy occurence 5 to 10 times higher than the general population [6,7]. Because of the complex relationships between genetic, immunological and environmental factors, most infected contacts will not develop leprosy, although recent studies have reported that they can be healthy carriers and spread M. leprae to susceptible individuals [8,9,10].
The investigation of the transmission and infectivity of M. leprae through molecular and immunological tools has shown that half of the leprosy contacts are healthy carriers, evidenced by the presence of M. leprae DNA in nasal swabs and, in nasal turbinate biopsies, and/or in the peripheral blood of healthy individuals, while about 18% presented subclinical infection (presence of anti-PGL-I IgM antibodies) with higher risk of illness [10,11,12]. It is important to emphasize that the subclinical neural involvement in this group has not yet been well defined, and its documentation is fundamental. Such elucidation would enable the discussion of chemoprophylaxis and early treatment, as a complementary strategy for leprosy control.
This is a case-control study that aimed to evaluate the clinical and laboratory predictors of subclinical neural impairment in leprosy household contacts.
We recruited leprosy household contacts from the National Reference Center of Sanitary Dermatology and Leprosy in Brazil, under the approval of the Ethics Committee of the Federal University of Uberlandia (CAAE: 48293215.7.0000.5152). A written informed consent was obtained from all participants for research participation. Some participants were minors and their parents provided written consent on behalf of them.
At this center, leprosy contacts are followed up for a period of at least 7 years, annually, when they are evaluated by a multidisciplinary team and submitted to dermatoneurological examination and serological (ELISA anti-phenolic glycolipid I Immunoglobulin M; anti-PGL-I IgM) analyses.
In Brazil, epidemiological investigation of contacts consists of an anamnesis addressed to signs and symptoms of leprosy, dermatoneurological examination and vaccination with BCG for contacts without signs and symptoms of leprosy at the time of evaluation, regardless their index case (PB or MB). Application of the BCG vaccine depends on the vaccination history and/or the presence of a vaccine scar, so contacts with no or only one scar should receive a new dose of BCG [7,8].
From 2014 to 2016, 373 new cases of leprosy and 2125 household contacts were reported, totaling an average of 5.7 contacts per patient. A total of 1902 contacts (90.5%, 1902/2125) were examined and 18% (342/1902) were seropositive. In this study, 175 seropositive and 35 seronegative contacts were recruited.
We excluded those who showed clinical evidence of disease (coprevalence cases) and those who presented other etiologies of peripheral neuropathies, such as: chronic alcoholism, diabetes mellitus, thyroid disease, hormonal dysfunctions, malnutrition, hereditary neuropathy, hepatitis B or C, HIV, autoimmune diseases.
Epidemiological and clinical data were recorded. All patients underwent a rigorous dermatoneurological evaluation by expert professionals. Intradermal sensory neuropathy or superficial leprosy neuropathy was defined by the presence of sensory abnormalities in a region not respecting the anatomical distribution of a specific nerve or spinal root, as terminal branches of several nerves may be involved in an affected area, while truncal neuropathy was defined as sensory and/or motor loss respecting the anatomical distribution of a specific nerve or spinal root.
Bacilloscopy–analyses of bacillary load of intradermal smears from six sites were performed: the two ear lobes, both elbows and knees, as well as from skin and/or nerve biopsy samples. Sample collection was preceded by topical application of cream containing lidocaine (7%) and tetracaine (7%) at all sites.
ELISA anti PGL-I IgM serology–Enzyme-linked immunosorbent assay (ELISA) was performed on all household contacts. Serum anti-PGL-I IgM antibodies were detected by enzyme-linked immunosorbent assay (ELISA) performed against the purified native PGL-I from the Mycobacterium leprae cell wall, according to a methodology previously described elsewhere. The reagent was obtained through BEI Resources, NIAID, NIH: Monoclonal Anti-Mycobacterium leprae PGL-I, Clone CS-48 (produced in vitro), NR-19370 [13].
DNA Extraction and Real-Time Quantitative Polymerase Chain Reaction (Real-Time PCR)–the DNA extraction from blood (500 μL), slit-skin smear, nerve and skin biopsies were performed. The quantitative real-time PCR (qPCR) assay targeting M. leprae DNA was performed by targeting the bacillus-specific genomic region (RLEP) in a real-time PCR system (ABI 7300, Applied Biosystems, Foster City, CA, USA) [10,14,15].
ENMG studies were carried out utilizing the MEB 4200K (NIHON-KODHEN) device. In the sensory conduction study, the median, ulnar, dorsal hand cutaneous, radial, lateral antebrachial cutaneous, median antebrachial cutaneous, sural, fibular superficial, saphenous and medial plantar nerves were examined bilaterally. In the motor conduction study, the median, ulnar, common fibular, and tibial bilaterally nerves were examined, supplemented by techniques for focal impairment identification at compression sites often affected in leprosy neuropathy, such as median nerve at the wrist, ulnar nerve at the elbow, fibular nerve at the fibular head and tibial nerve at the ankle. The parameters used to evaluate each nerve are described separately as a supplementary file.
All of the leprosy contacts selected did not present any skin lesion. For this reason, the biopsy of a small elbow skin fragment (approximately 1 cm) was performed, considering that it is a cold region with possible intradermal impairment, and therefore a site often altered in leprosy neuropathy.
Nerves that underwent biopsy were selected according to the patient’s clinical condition, and included exclusively sensory nerves that presented sensory changes and/or thickening, and also one of the following electrophysiological changes in the sensory conduction analysis: absence of response on both sides; unilateral absence of response; bilaterally decreased amplitude of the sensory action potential (SAP), considering reference values; and over 50% decrease in the amplitude of the SAP, compared with the contralateral side. During the biopsy, the nerve was isolated and completely transected. All patients signed a specific informed consent form referring to this process. During the procedure, a skin biopsy of the area superjacent to the corresponding territory of the nerve also underwent a biopsy procedure. The biopsied nerve and skin were processed and studied according to routine standard procedures. Formalin-fixed paraffin-embedded were cut longitudinally and transversely at 5-μm thickness and stained with hematoxylin and eosin stain. Additionally, special staining with Masson Trichome was performed to assess fibrosis. Fite-Faraco stain was performed for bacilli identification.
The Shapiro Wilk test was used to test data normality within groups. The Wilcoxon-Mann-Whitney U Test was carried out, and the Binomial Test was applied for the Study of Dichotomous Variables, with significance defined as p<0.05. Multiple logistic regression was used to verify the dependence relation between the presence of ENMG abnormality (categorical variable) and the independent variables (ELISA anti-PGL-I IgM, intradermal smear qPCR, skin biopsy qPCR, peripheral blood qPCR and BCG scar). After verifying the dependence between variables, odds ratios (OR) were determined, and the probability of outcomes analyzed. The statistical program used was the software GraphPad Prism version 7.
Comparisons of all epidemiological characteristics between groups did not show any significant difference (Table 1). The mean anti-PGL-I IgM ELISA index was 2.05 in seropositive contacts, and 0.52 in seronegative contacts (p<0.0001). In the analysis of the peripheral blood qPCR from seropositive contacts, 40.6% (71/175) presented positivity, while only 8.6% (3/35) in seronegative contacts (p = 0.0003). In the intradermal smear analysis, only 4% (7/175) of the seropositive contacts presented positive bacilloscopy, whereas the evaluation by the qPCR in this group showed positivity of 47.4% (83/175) and only 17.1% (6 / 35) in the seronegative contacts (p = 0.0009), all with negative bacilloscopy (Table 1).
Regarding the clinical evaluation of seropositive contacts, 18.3% (32/175) presented a pattern of intradermal impairment, compared with 14.3% (5/35) in the seronegative contacts (p = 0.5717), defined as multifocal painful hypoesthesia, especially with a greater involvement in the elbow, knee and earlobe regions, setting a temperature-dependent pattern. Sensorial impairment with a specific territory distribution (truncal pattern) was present in 17.1% (30/175) of seropositive contacts and in 8.6% (3/35) of seronegative contacts (p = 0.2079). The impairment of the deep sensation (vibratory and kinetic postural) and deep osteotendin reflexes was not observed in any case. Only 2.3% (4/175) of seropositive contacts and 2.8% (1/35) of seronegative contacts presented motor manifestation (p = 0.8598). The presence of neural thickening was observed in 21.1% (37/175) of seropositive versus 8.6% (3/35) of seronegative contacts (p = 0.0838). Among contacts with thickening, the ulnar nerve alteration was the most frequently one (72.5%, 29/40). None of the evaluated contacts presented skin lesion.
ENMG evaluation detected some neural impairment in 31.4% (55/175) of seropositive contacts. In seronegative contacts, only 13.3% (4/35) showed changes in this examination (p = 0.0163). (Table 1) Of the 59 contacts with altered ENMG, 81.3% (48/59) were contacts of MB index cases, although this condition did not confer greater chances of alteration in this examination (OR, 0.99; CI95%, 0.45 to 2.15; p = 0.9865).
Only 32.2% (19/59) presented neural thickening in the clinical evaluation. However, the presence of neural thickening conferred a 2.94-fold higher chance of presenting ENMG abnormality (OR = 2.94; CI95%, 1.43 to 6.00; p = 0.0031). The mean number of nerves affected was 1.44 per contact. The nerves most frequently affected are described in Table 2.
In the neurophysiological pattern observed in ENMG, 69.5% (41/59) presented only one altered nerve (mononeuropathy), and 30.5% (18/59) two or more altered nerves (multiple mononeuropathy).
According to clinical data and ENMG results, 50.8% (30/59) of leprosy contacts demonstrated at least one nerve eligible for biopsy, but only 60.0% (18/30) of those were submitted to this process. The most frequent biopsied nerve was the sensory ulnar—dorsal cutaneous of the hand (72.3%; 13/18), followed by superficial fibular (16.7%; 3/18), sural (5.5%; 1/18), and deep fibular (5.5%; 1/18). Only 27.8% (5/18) of the nerves presented some anatomopathological alterations suggestive of leprosy, such as endoneural or epineural infiltrate, presence of fibrosis, perineural thickening or presence of endoneural granuloma. No leprosy contacts presented positive bacilloscopy in the peripheral nerve biopsy. On the other hand, qPCR of nerve biopsies was positive in 61.1% (11/18) of the cases. The qPCR of the suprajacent skin area was positive in 16.7% (3/18) of the nerve biopsies, whereas bacilloscopy was negative in all samples.
In order to further explore the complex interaction among results, a multivariate statistical method was conducted to confirm the dependence relation of variables elucidated above with the chance of occurrence of ENMG abnormalities, demonstrating that ELISA anti-PGL-I positivity confers a 4.04-fold greater chance of neural damage (OR = 4.04; CI95%, 1.24 to 13.21; p = 0.020), while peripheral blood qPCR positivity presents a 2.08-fold higher chance (OR = 2.08; CI95%, 1.08 to 4.02; p = 0.028). The presence of at least one BCG vaccine scar demonstrated 2.44-fold greater protection against neural impairment (OR = 0.41; CI95%, 0.18 to 0.98; p = 0.044). There was no dependence relation with the variables intradermal smear qPCR or skin biopsy qPCR (Table 3).
The combination of unfavorable results for the three assays (no BCG scars, seropositivity of anti-PGL-I IgM, and positive qPCR in peripheral blood) indicated the highest probability (62.6%) of neural impairment in contacts. The presence of BCG scars in combination with other disease predictors led to the reduced probability of neural impairment. The group of contacts with favorable results (presence of BCG scars, negative anti-PGL-I and negative qPCR in peripheral blood) was the one with the lowest probability (7.6%) of neural damage (Table 4)
This is a case-control study in Brazil that measured the chance of occurrence of peripheral neural impairment in asymptomatic leprosy household contacts, through serological, molecular and neurophysiological tests.
The prevalence of abnormalities in the ENMG reinforce the importance of epidemiological surveillance and follow-up of leprosy contacts, allowing early recognition, by a combination of diagnostic tools, of neural impairment in this population.
Previous studies have already documented neural involvement in leprosy contacts, but none has explored how predictors and laboratorial tests are correlated with such pathological occurrence. This is the first study in an endemic country evidencing that subclinical neural impairment may be the first and only clinical manifestation of leprosy, and when appropriately recognized may contribute to early diagnosis and treatment of leprosy, which by definition is primarily neural [16,17,18,19,20].
Some ENMG abnormalities may precede the classic clinical symptoms of leprosy, such as the absence or amplitude reduction of the sensory action potential of some nerves, focal myelinic impairment, which is corroborated by our findings [20–22].
Although asymptomatic, some leprosy contacts already had at least one abnormality detected in the detailed neurologic physical examination, mainly sensory impairment and neural thickening, corroborating the pattern described in the classical forms of leprosy, an asymmetric peripheral neuropathy that is predominantly sensorial. These contacts present a subclinical form in which the ENMG is superior to the thermal, tactile and vibratory sensation evaluation, with capacity for early detection of neural impairment [23,24,25].
Neural thickening, despite being one of the cardinal signs of leprosy and a risk factor for the presence of ENMG abnormalities, as demonstrated in the present study, is a subjective parameter and does not always show agreement with the ENMG, since only one third of the leprosy contacts with ENMG abnormality presented neural thickening [24,25,26].
Leprosy contacts of MB patients did not present higher chances of neural impairment, although this factor is associated with an increase in the disease outcome in several prospective studies [6,9,10,11,12], which only evaluated the natural history of the disease, but without a neurophysiological, serological or molecular intervention for early diagnosis, as shown in our report.
Our results have demonstrated that ELISA anti-PGL-I is the most important test in determining the increased chance of neural impairment in leprosy contacts, corroborating previous studies that also demonstrated its importance as a screening test in the definition of leprosy contacts that present a higher risk of illness. The use of the ELISA anti-PGL-I test is justified due to its high correlation with MB clinical forms, being directly proportional to bacillary load, and also its association with a increased risk of developing leprosy in seropositive contacts [7,10,12,13,27,28,29].
BCG vaccination has been associated with prevention of leprosy in different studies, especially MB forms [7,30]. Based on our results, the presence of one or more BCG scars provided protection against neural damage. Thus, an additional intradermal BCG booster dose should be maintained in leprosy control programs, aiming for protection against leprosy, including neural forms [7].
Concerning the molecular evaluation, studies have shown good prospects regarding the detection of M. leprae in several samples (blood, skin, swabs, smear) of leprosy patients and contacts by qPCR, which have contributed to the definition of the existence of healthy carriers and subclinical infection [10,11,14,15]. We have shown previously that the positivity of peripheral blood qPCR in contacts was 6.7% with a 5.54-fold risk for disease outcome [10]. Our current results reinforce our previous findings, demonstrating an increased chance of neural involvement in contacts with positive peripheral blood qPCR.
Although the qPCR positivity of intradermic smear and skin biopsy did not determine an increased chance of neural damage, these tools may play a role in diagnostic confirmation, even allowing the initiation of treatment of asymptomatic contacts.
Leprosy household contacts constitute a group of individuals at high risk for disease development, so their participation in the dissemination of M. leprae to susceptible individuals in endemic communities cannot be neglected [10]. Despite significant progress in controlling leprosy in recent years, early diagnosis remains the primary goal and challenge of leprosy control.
Therefore, with the prospect of eliminating leprosy as a public health problem, the development and implementation of more specific and sensitive methods for the detection of M. leprae and its neural impairment, using immunological, molecular and neurophysiological tools are mandatory to increase the knowledge of leprosy epidemiology, to break its chain of transmission, thereby enabling effective control of this disease.
Taking into consideration our findings, we propose an algorithm for the follow-up of leprosy household contacts (Fig 1). We suggest annual monitoring through serological (ELISA anti-PGL-I) evaluation for at least 5–7 years, considering the better risk-benefit in relation to neural impairment and development of MB clinical forms [28].
The combination of the three assays in (ELISA anti-PGL-I, peripheral blood qPCR and BCG scars) may identify individuals with higher chances of developing leprosy neuropathy, not only justifying the treatment initiation in those with confirmed diagnosis, but also indicating chemoprophylaxis in contacts with unfavorable predictors.
One of the limitations of the study was that it did not present the follow-up of interventions proposed above, regarding early treatment and chemoprophylaxis, which should be done in future work. In addition, unfortunately, leprosy remains a neglected disease, making it difficult to apply this study to clinical practice in endemic countries.
|
10.1371/journal.ppat.1006500 | An essential EBV latent antigen 3C binds Bcl6 for targeted degradation and cell proliferation | The latent EBV nuclear antigen 3C (EBNA3C) is required for transformation of primary human B lymphocytes. Most mature B-cell malignancies originate from malignant transformation of germinal center (GC) B-cells. The GC reaction appears to have a role in malignant transformation, in which a major player of the GC reaction is Bcl6, a key regulator of this process. We now demonstrate that EBNA3C contributes to B-cell transformation by targeted degradation of Bcl6. We show that EBNA3C can physically associate with Bcl6. Notably, EBNA3C expression leads to reduced Bcl6 protein levels in a ubiquitin-proteasome dependent manner. Further, EBNA3C inhibits the transcriptional activity of the Bcl6 promoter through interaction with the cellular protein IRF4. Bcl6 degradation induced by EBNA3C rescued the functions of the Bcl6-targeted downstream regulatory proteins Bcl2 and CCND1, which resulted in increased proliferation and G1-S transition. These data provide new insights into the function of EBNA3C in B-cell transformation during GC reaction, and raises the possibility of developing new targeted therapies against EBV-associated cancers.
| Epstein-Barr virus (EBV) is the first characterized human tumor virus, which is associated with a broad range of human cancers. One of the latent proteins, EBV encoded nuclear antigen 3C (EBNA3C) plays a critical role in EBV-mediated B-cell transformation. Bcl6 is a master regulator required in mature B-cells during germinal center (GC) reaction. As a transcriptional repressor, Bcl6 can be targeted during malignant transformation and therefore contributes to its function as an oncoprotein during lymphomagenesis. In this study, we demonstrated that EBNA3C interacts with Bcl6 and facilitates its degradation through the ubiquitin-proteasome dependent pathway, and suppresses Bcl6 mRNA expression by inhibiting the transcriptional activity of its promoter. Furthermore, EBNA3C-mediated Bcl6 regulation significantly promotes cell proliferation and cell cycle by targeting Bcl2 and CCND1. Therefore, our findings offer new insights into the functions of EBNA3C during B-cell transformation in GC reaction and B-cell lymphoma development. This increases the possibility of developing new therapies for treating EBV-associated cancers.
| B-cell development through the germinal center (GC) is controlled strictly by sequential activation or repression of crucial transcription factors, executing the pre- and post-GC B-cell differentiation [1]. The deregulation of induced GC reactions during B-cell development is associated with malignant transformation giving rise to different types of lymphoma and leukemia [2]. Most mature B-cell malignancies, including diffuse large B-cell lymphoma (DLBCL), follicular lymphoma (FL) and Burkitt’s lymphoma (BL) are derived from malignant transformation of GC B-cells [2,3]. Furthermore, DLBCL is the most common subtype of non-Hodgkin’s lymphoma (NHL), accounting for approximately 40% of all cases [4]. DLBCL is considered a heterogeneous group of tumors, with some specific clinicopathological variants of DLBCLs being associated with the presence of EBV [5,6].
A major regulator of the GC reaction is represented by B-cell lymphoma 6 (Bcl6), a sequence specific transcriptional repressor [7–9]. Knock-out of Bcl6 in vivo results in lack of GC formation and the maturation of high-affinity antibodies [10,11]. Interestingly, deregulation of Bcl6 expression can be found in BL, FL and DLBCL [12,13]. In addition, Bcl6 is the most frequent oncogene involved in roughly 40% of the cases of DLBCLs, and its locus is frequently rearranged due to chromosomal translocations in DLBCL [14,15]. As a key transcriptional repressor in normal B-cell differentiation, Bcl6 was shown to repress NF-κB and the positive regulatory domain I element (PRDM1) also known as Blimp-1 in DLBCLs [16–18]. Also, Bcl6 is now been investigated as a potential therapeutic target for the treatment of tumors with rationally designed specific Bcl6 inhibitors [19–21].
EBV is a lymphotropic virus that is linked to many kinds of B-cell malignancies, including BL, FL and DLBCL [22,23]. EBV infection transforms primary human B-cells into continuously growing lymphoblastoid cells (LCLs) and different latent types were established in EBV-infected cells [23,24]. During latency III or the growth program, EBV expresses the full complement of oncogenic latent proteins, including EBV nuclear antigens EBNA1, EBNA2, EBNA3A, EBNA3B, EBNA3C and EBNA-LP, as well as latent membrane proteins LMP1, LMP2A and LMP2B in addition to numerous RNAs and miRNAs [25]. Genetic studies using recombinant virus strategies demonstrated that EBNA1, EBNA2, EBNA3A, EBNA3C, EBNA-LP and LMP1 are essential or very important for EBV-mediated transformation of primary B-cells in vitro [26–28]. Specifically, EBNA3C has the ability to function as a transcriptional activator and repressor, and can regulate the transcription of both cellular and viral genes [29,30]. A number of earlier studies have shown that EBNA3C interacts with a wide range of transcription factors and modulators, such as c-Myc [31], IRF4 [32], CtBP [33], p53 [34], E2F1 [35] and E2F6 [36], which leads to dysregulation of their associated cellular functions.
Previous studies have indicated that expression of EBV latent proteins were associated with Bcl6 expression [37–39]. For example, in some B-cell lymphomas, Bcl6 expression was inversely correlated with LMP1 expression, and some data suggested that LMP1 can cause downregulation of Bcl6 [6,37]. However, the link between LMP1 and Bcl6 was not fully explained as Bcl6 expression was also inhibited in LMP1-negative cells [38]. Similar studies have shown that LMP1 through heterologous expression was unable to suppress expression of Bcl6 in DLBCLs [39]. In addition, EBNA2 can interfere with the germinal center phenotype by downregulating Bcl6 in Non-Hodgkin's Lymphoma cells [39]. Furthermore, EBV encoded microRNAs can repress Bcl6 expression in DLBCL [38]. However, the mechanism by which Bcl6 is down-regulated in EBV-infected cells is still not fully understood. Our goal is to determine the role of EBNA3C in regulating expression of Bcl6 oncoprotein in B-cells, and further uncover novel molecular mechanisms by which EBNA3C-mediated regulation of cellular functions can lead to B-cell transformation.
To determine the expression levels of Bcl6 during EBV infection of primary B-cells, 10 million human peripheral blood mononuclear cells (PBMCs) from two donors, respectively, were infected with wild-type BAC-GFP-EBV or EBNA3C-deleted (ΔE3C) BAC-GFP-EBV. The infected cells were harvested at different time points (0, 2, 4, 7 and 15 days post-infection), then mRNA was extracted and Real-time PCR was performed for Bcl6 detection. The results showed that for both donors Bcl6 expression was down-regulated and expressed at a relatively low level after wild-type EBV infection. However, its expression was consistently enhanced with the EBNA3C-deleted EBV infection as much as 20–35 fold over wild-type infection (Fig 1A and 1B). This suggested that EBNA3C can play a key role in Bcl6 expression during EBV infection.
To determine the effect of EBNA3C on Bcl6 expression in Burkitt’s lymphoma cells, western blot analysis was also performed in EBV-negative BL41 and Akata cells, as well as the corresponding EBV-positive BL41/B95.8, Akata-EBV cells. We found a significant downregulation of Bcl6 expression in the presence of EBV-infected BL41 and Akata cell lines of approximately 2–4 fold (Fig 1C). To further investigate whether the differential expression was due to the presence of EBNA3C, EBV-negative Ramos and BJAB cells; EBNA3C expressing stable BJAB cells (BJAB7 and BJAB10); and EBV-transformed lymphoblastoid cell lines (LCL1 and LCL2), were analyzed by western blot. Similarly, Bcl6 protein expression was down-regulated close to 50% in the presence of EBNA3C in Burkitt’s lymphoma cells and were dramatically suppressed in EBV-transformed LCLs (Fig 1D). These results strongly suggested that EBNA3C contributes to inhibition of Bcl6 expression.
To further examine if Bcl6 expression was regulated by EBNA3C specifically, HEK293T and BJAB cells were transfected with a dose-dependent increase of EBNA3C in addition to heterologous expression of Bcl6. The western blot results demonstrated that EBNA3C expression led to a strong reduction in Bcl6 protein expression in human cells, including B-cell lines of about 4–10 fold (Fig 1E and 1F). To further explore the role of EBNA3C in modulating Bcl6 expression levels in EBV transformed LCLs, EBNA3C knocked-down LCL1 cell line was generated with specific EBNA3C short hairpin RNA (sh-E3C) [40]. Compared to the control LCL1 (sh-Ctrl), Bcl6 protein expression was significantly increased by at least 2-fold in sh-E3C LCL1 cells (Fig 1G). The results provide additional supporting evidence demonstrating that down-regulation of Bcl6 expression can be specifically linked to EBNA3C.
Next, we examined whether EBNA3C could interact directly with Bcl6. Two experiments using Co-Immunoprecipitation (Co-IP) assays were performed in different cell types. First, HEK293T cells were transfected with Myc-tagged EBNA3C and HA-tagged Bcl6. The Co-IP results showed that EBNA3C associated in a complex with Bcl6 (Fig 2A). Similarly, an experiment using Saos-2 cells also showed the formation of a complex of EBNA3C and Bcl6 in these cells (Fig 2B). Second, to further determine the interaction in B-cell lines, BJAB, EBNA3C stably expressed BJAB cells (BJAB7 and BJAB10), and EBV-transformed cells (LCL1 and LCL2) were used in our Co-IP experiment. Western blot analysis also validated the above results demonstrating that endogenous EBNA3C can physically associate with Bcl6 in the background of B-cells and more importantly in EBV-transformed lymphoblastoid cell lines (Fig 2C and 2D).
To determine the functional binding domain of EBNA3C that specifically interacts with Bcl6, Myc-tagged full length and truncated regions of EBNA3C (1-365aa, 366-620aa, 621-992aa) were co-transfected into HEK293T cells with HA-tagged Bcl6. The targeted protein was immunoprecipitated with anti-Myc or anti-HA antibody, respectively. The results demonstrated that Bcl6 was associated with EBNA3C (366-620aa) along with the full-length EBNA3C protein (1-992aa) (Fig 2E and 2F). Little or no detectable co-immunoprecipitation was observed with the control vector indicating the specificity of the complex between EBNA3C and Bcl6. These results showed that EBNA3C amino acid residues 366-620aa which includes the acidic domain were responsible for the interaction of EBNA3C and Bcl6 protein (Fig 2G).
Previous studies have shown that EBNA3C binds to Bcl6 specifically in human cells, so it is expected that these two proteins would be localized within the same cellular compartments. To determine the co-localization of EBNA3C and Bcl6, HEK293T cells were transfected with constructs expressing Myc-tagged EBNA3C and HA-tagged Bcl6, and the cellular localization of the expressed proteins was studied using immunofluorescence microscopy. In cells transfected independently with Myc-EBNA3C or HA-Bcl6 alone, both were found to be primarily expressed in the nucleus (Fig 3A). In cells co-transfected with Myc-EBNA3C and HA-Bcl6, the merged yellow fluorescence demonstrated that EBNA3C co-localized with Bcl6 in human cells (Fig 3A). Similar results were also observed in Saos-2 cells (Fig 3B).
To further determine the co-localization of EBNA3C and Bcl6 proteins in more relevant B-cells, immunofluorescence assays were performed using antibodies specific to EBNA3C and Bcl6 proteins in order to examine the endogenous expression in different B-cell lines. The results further confirmed that EBNA3C co-localized with Bcl6 in nuclear compartments of EBV-transformed LCLs (Fig 3C). This was consistent with the results of the above studies which demonstrated that EBNA3C associated with Bcl6 in nuclear complexes as seen in the Co-IP experiments in human cells.
To explore the potential mechanism of EBNA3C-mediated down-regulation of Bcl6, a stability assay was performed to determine whether EBNA3C regulated Bcl6 expression at the protein level. HEK293T cells were transfected with HA-tagged Bcl6 and Myc-tagged EBNA3C or Myc-tagged empty vector. Twenty-four hours post-transfection, cells were incubated with protein synthesis inhibitor cycloheximide (CHX) and monitored for Bcl6 protein levels at 0, 4, 8, 12 hours by western blot analysis. As expected, the results showed that the stability of Bcl6 protein was significantly decreased by greater than 50% in the presence of EBNA3C by 12 hours, while the Bcl6 protein level was more stable in the absence of EBNA3C (compare right and left panels, Fig 4A). To further support these results, BJAB (EBNA3C negative B-cells), and BJAB7 (BJAB stably expressing EBNA3C cells) were treated with CHX for 0, 2, 4, and 6 hours. The following western blot analysis also demonstrated that there was a dramatic reduction in the stability of Bcl6 protein which was directly associated with EBNA3C expression as seen by the significant change in the Bcl6 levels by 2 hours post cycloheximide treatment and greater than 50% by 6 hours (compare right and left panels, Fig 4B).
Bcl6 expression is strictly regulated during GC reaction, and its degradation through the ubiquitin-proteasome pathway is crucial for B-cell development or lymphomagenesis in temporal function. Earlier studies showed that Bcl6 could be degraded by the ubiquitin-mediated proteasome [41,42]. Therefore, it is expected that Bcl6 degradation is likely mediated by EBNA3C utilizing a similar pathway as EBNA3C has been shown to recruit E3 ligases for targeted degradation of cellular substrates [43,44]. To determine whether this is the case, HA-Bcl6 was transfected along with Myc-EBNA3C or control vector. Twenty four hours post-transfection, the cells were treated with the proteasome inhibitor MG132 for 12 hours or vehicle control. The following western blot analysis showed that EBNA3C promoted the degradation of Bcl6 protein, which was similar to the above results. However, Bcl6 protein expression was increased after MG132 incubation, even in the presence of EBNA3C (Fig 4C). These results demonstrated that the stability of the Bcl6 protein is regulated by EBNA3C via the ubiquitin-proteasome pathway.
To further support our hypothesis, ubiquitination assays were performed with different expression plasmids for Myc-E3C, HA-Ub and HA-Bcl6, and incubated for 24 hours followed by MG132 treatment for another 12 hours. This was followed by immunoprecipitation and western blot analysis. The results demonstrated enhanced ubiquitination of Bcl6 when EBNA3C was expressed, when compared with control vector or HA-Ub alone (Fig 4D). This strongly indicated that Bcl6 is likely degraded by expression of EBNA3C through the ubiquitin-proteasome dependent pathway.
Bcl6 gene expression is tightly regulated during mature B-cell development [2,45,46]. Our above studies showed that Bcl6 mRNA expression was down-regulated after EBV infection, and that this was associated with EBNA3C expression. To further define how EBNA3C can regulate Bcl6 expression at the mRNA level, different B-cell lines (BJAB, BJAB7, BJAB10, LCL1 and LCL2) were used to monitor endogenous Bcl6 mRNA expression. Bcl6 mRNA expression was significantly greater (>20 fold) in BJAB cells compared to EBNA3C stably expressed BJAB7 and BJAB10 cell, and also EBV-transformed LCL1 and LCL2 cells (Fig 5A). To verify that the inhibition was related to the presence of EBNA3C, EBNA3C stably knocked-down LCL1 (sh-E3C) and the control LCL1 (sh-Ctrl) were used to detect Bcl6 mRNA expression. The results showed that Bcl6 mRNA expression was upregulated significantly after the knockdown of EBNA3C (Fig 5B). In addition, BJAB10 cells were then transfected with specific EBNA3C short hairpin RNA (sh-E3C) to knock down EBNA3C expression. Expectedly, Bcl6 mRNA expression was increased in the EBNA3C knockdown cell lines (Fig 5C). These findings undoubtedly provide new evidence that EBNA3C can inhibit Bcl6 mRNA expression.
Bcl6 promoter transcriptional activity could not only be controlled by Bcl6 through binding to the upstream regulatory region of its gene [15], but is also inhibited directly by the transcription factor IRF4 via binding to multiple sites within its promoter [47]. To test whether EBNA3C-mediated Bcl6 mRNA down-regulation was related to its transcriptional activity at the Bcl6 promoter, a dual-luciferase reporter system was implemented. The reporter construct containing a wild-type Bcl6 promoter (pLA/B9) and a dose-dependent increase of Myc-EBNA3C were transfected into cells. Meanwhile, the thymidine kinase promoter-Renilla luciferase reporter plasmid (pRL-TK) was additionally transfected and used as a control for transfection efficiency. The luciferase assay results clearly revealed that the Bcl6 promoter activity was inhibited by EBNA3C in a dose-dependent manner (Fig 5D).
Previous experiments showed that EBNA3C did not bind with DNA directly and functions through binding of other cellular transcription proteins to regulate gene expression [48]. Therefore, other transcription proteins mediate the inhibition of viral and cellular genes. EBNA3C interacted with p53, attenuated its function and mediated its degradation [34,44,49]. Furthermore, p53 can activate Bcl6 transcription [50]. It suggests that the transcription activity of the Bcl6 promoter may be inhibited by EBNA3C-induced p53 degradation. However, our results using MEF(p53-/-) cells showed that the regulation of Bcl6 promoter by EBNA3C was independent of the function of p53 protein (S1 Fig). Among several other transcriptional proteins that inhibited Bcl6 promoter, we found that IRF4, a DNA-binding protein, was an important transcription factor for regulating Bcl6 promoter activity [47]. Interestingly, EBNA3C also interacted with IRF4 and contributed to stabilization of IRF4 [32]. One study showed that a high level of IRF4 was expressed in LMP1-KO EBV-induced lymphoma [51]. Thus, we further assessed the possible function of EBNA3C on IRF4-mediated Bcl6 promoter activity. To specifically test the Bcl6 promoter activity, the wild-type and DNA binding domain (DBD)-deleted IRF4 plasmids were used (Fig 5E). The results showed that EBNA3C enhanced the IRF4-mediated inhibition of the Bcl6 promoter activity, and the effect was dependent on the DNA binding domain of IRF4 as the promoter repression was rescued when EBNA3C and IRF4-ΔDBD were co-expressed (Fig 5E). It also suggested that IRF4 is one of the major transcription factors that mediate EBNA3C-regulated inhibition of Bcl6 promoter activity.
To examine the effect of EBNA3C on Bcl6-mediated cell proliferation, Saos-2 cells were transfected with expression constructs of EBNA3C and Bcl6, and selected with G418 for two weeks to monitor colony formation. We observed a significant increase in colony numbers when EBNA3C and Bcl6 were co-transfected in comparison to those transfected with only EBNA3C or Bcl6 (Fig 6A). We further extended these studies by performing cell proliferation assays as determined by cell counting for 10 days in Saos-2 cells (Fig 6B). A similar experiment was also repeated in HEK293 cells (Fig 6C). These results demonstrated that expression of EBNA3C and Bcl6 results in a strong induction in cell proliferation.
The anti-apoptotic proto-oncogene Bcl2 protein is a critical regulator protein and its expression is inhibited by Bcl6 in GC B-cells [52,53]. Therefore, it was reasonable to believe that EBNA3C-mediated Bcl6 down-regulation will lead to up-regulation of Bcl2 expression. Therefore, its anti-apoptotic function will be activated and leads to promotion of cell proliferation. To determine the expression of Bcl2 in B-cells, LCL1 was treated with a Bcl6-specific inhibitor (79–6) to suppress Bcl6 activity [20]. The Bcl6 inhibitor disrupts Bcl6 transcription activity by binding to its BTB/POZ domain [20]. The following western blot results showed that Bcl2 expression was up-regulated after Bcl6 inhibitor incubation in B-cells by approximately 2-fold (Fig 6D). To further support the results, Bcl2 mRNA expression was determined in the stable EBNA3C or Bcl6 knock-down LCL1 cells to verify that EBNA3C promoted Bcl2 up-regulation through Bcl6 down-regulation in LCLs (S2 Fig). This suggests that EBNA3C-mediated Bcl6 inhibition can contribute to cell proliferation through the Bcl2-associated signaling pathway.
The soft agar assay for colony formation measures anchorage-independent in vitro transformation. The oncogene Bcl6 can confer anchorage-independent growth to immortalized cells [54]. To investigate the effects of EBNA3C on Bcl6-related transforming activity, the stable Bcl6 knock-down BJAB and LCL1 cells were generated with lentivirus transduction and puromycin selection (Fig 7A–7C). Soft agar assays were performed using the Bcl6 knock-down BJAB and LCL1 cells. The down-regulation of Bcl6 inhibited the ability of colony formation in BJAB cells (Fig 7D), but oppositely, the ability was enhanced in EBV-transformed LCLs (Fig 7E). The results indicate that EBV promotes transformation and anchorage-independent growth through the inhibition of Bcl6 expression.
Our previous study showed that EBNA3C could only stabilize Cyclin D1 (CCND1) protein, but not promote its transcription activity [40]. This suggests that other cellular factors may regulate CCND1 expression. Interestingly, CCND1 is induced by Bcl6 in human B-cells [55]. To further investigate the function of Bcl6 in cell cycle, we analyzed CCND1 mRNA expression in stable B-cells. The results show that CCND1 expression is also suppressed when Bcl6 is knocked-down in EBV-negative BJAB cells. However, its expression is upregulated in stable Bcl6 knock-down EBV-transformed LCL1 cells (Fig 8A and 8B). These results suggest that Bcl6 plays a critical role in controlling CCND1 mRNA expression in a B-cell background. The following cell cycle experiments also demonstrated that the upregulation of CCND1 through Bcl6 inhibition facilitates G1-S transition in EBV-transformed LCL1 cells, but not in the EBV-negative BJAB cells (Fig 8C and 8D). Similar results were observed with other sh-Bcl6 clones.
Bcl6 is a nuclear phosphoprotein of the BTB/POZ/Zinc Finger (ZF) protein family, and functions as a transcription repressor to repress target genes by binding to specific DNA sequences and recruiting corepressors [8,56], including SMRT, MTA3, N-CoR and HDAC [57–60]. Bcl6 is indispensable for GC formation and somatic hypermutation (SHM) during B-cell development, thus chromosomal translocations and mutations of Bcl6 regulatory region lead to the deregulation of Bcl6 expression in about 40% DLBCL and 5–10% FL [46]. Although Bcl6 expression is associated with EBV latent antigen EBNA2 and LMP1, the reported conflicting results did not provide a reasonable explanation or a detailed mechanism on EBV-mediated Bcl6 degradation in B-cell lymphoma [37–39]. A recent study indicated that EBNA3C had no effects on Bcl6 expression, but a previous paper also showed that Bcl6 expression can be increased more than 10-fold in EBNA3C-deleted EBV infection [61,62]. Here, our data clearly show that Bcl6 expression can be down-regulated by EBNA3C specifically at transcriptional and post-transcriptional levels (Fig 9). This is different from the well-known Bcl6 translocations or mutations identified on the human genome associated with oncogenesis.
First of all, our results demonstrated that EBNA3C was specifically associated with Bcl6, and mediated Bcl6 protein degradation through the ubiquitin-proteasome dependent pathway. A previous study showed that Bcl6 protein can be targeted for degradation by cellular factor FBXO11 in DLBCL [63]. However, the role of EBNA3C on FBXO11-related Bcl6 stability is still unclear. It is possible that FBXO11 is the E3 ligase recruited by EBNA3C for Bcl6 degradation. In addition, the acetylation of Bcl6 within the PEST domain inactivates its function of recruiting co-repressors [54], and the activation of MAPK signaling pathway induces phosphorylation of Bcl6 followed by degradation through the ubiquitin-proteasome pathway [64]. Further studies are warranted to determine whether EBNA3C-mediated Bcl6 degradation is related to Bcl6 acetylation and phosphorylation.
BCL6 activity is also dysregulated by translocation or mutation in a remarkably high proportion of DLBCL and FL [65]. The chromosomal translocations of Bcl6 regulatory region referred to as promoter substitution, and frequent mutations of the 5’ noncoding region of Bcl6 result in its deregulated expression, suggesting a key role for Bcl6 in pathogenesis of B-cell lymphoma [12,13,66,67]. However, our results indicate that Bcl6 mRNA expression is down-regulated through EBNA3C-mediated inhibition of the transcription activity of Bcl6 promoter by recruiting another cellular factor IRF4. This is consistent with a previous study showing that the CD40 receptor signaling pathway leads to NF-κB-mediated IRF4 activation, and furthermore Bcl6 downregulation [47]. Meanwhile, we also showed that EBNA3C could interact with IRF4 and was critical for IRF4 stabilization [32]. This suggests that EBNA3C may mimic the activities of the CD40 ligand to induce NF-κB-IRF4 signaling pathway or enhance the stability of IRF4 protein directly to repress the transcription activity of the Bcl6 promoter. Moreover, EBNA3C was associated with IRF8 and mediated its destabilization and degradation [32]. Interestingly, IRF8 is the only transcriptional activator of Bcl6 to upregulate its mRNA expression in GC reaction [68]. Therefore, it is conceivable that EBNA3C may downregulate Bcl6 expression by activating CD40 signaling pathway as well as regulating IRF4/IRF8 stability.
The importance of Bcl6 function in GC B-cells is reflected in the multiple functional pathways it can regulate in the cell. To date, more than one thousand genes are found to be targeted by Bcl6 through binding on their promoters and further modulating the downstream signaling pathways during GC development, involved in cell apoptosis, cell cycle and cell differentiation [69,70]. Among the targeted proteins, Bcl2 is a critical anti-apoptosis protein and the direct target of Bcl6 that can interact with Miz1 and bind to Bcl2 promoter to inhibit Miz1-induced Bcl2 transcription activity in GC B-cells [52,53]. The dysregulation of Bcl6-mediated Bcl2 expression is often found in DLBCL and FL [2]. Our results show that Bcl2, a Bcl6 target protein, is regulated by EBNA3C and is increased in LCLs treated with a Bcl6 inhibitor. Thus EBNA3C can induce cell proliferation by degrading and inhibiting the expression of Bcl6 and so releasing the suppression of Bcl2, therefore activating the anti-apoptosis pathway for tumorigenesis. Moreover, CCND1, a direct target of Bcl6 in human B-cells, is de-repressed to promote G1-S transition in EBV-transformed LCLs. Whether other cyclin proteins are also under the control of Bcl6 is still unknown. Interestingly, several studies have shown that CCND2 is another target of Bcl6, but its expression is negatively correlated [71–75]. Activation-induced cytidine deaminase (AID) which is responsible for somatic hypermutation and class-switch recombination is also required in GC-derived lymphomas, and its expression is upregulated by EBNA3C in EBV-infected cells [61,76]. Bcl6 could promote AID expression by inhibiting miR-155 and mir-361, so how EBNA3C regulates AID expression without the help of Bcl6 needs to be further explored [2]. A recent study concluded that Bcl6 targeted genes in T follicular helper (Tfh) cells through analysis of its genome-wide occupancy and transcriptional regulatory networks [77]. The current development of Bcl6 small-molecular inhibitor indicates a huge potential for Bcl6 as a therapeutic target to treat human lymphomas [19]. However, the Bcl6-mediated regulatory networks are still unknown in EBV-transformed LCLs. Next, xenografts of LCLs in Bcl6 knock-out mice will further reveal the biological function of Bcl6 in EBV-related lymphomagenesis. However, a more efficient in vivo model will be necessary to uncover the crucial functions of EBNA3C or other latent antigens in GC reaction. In summary, the inhibition of Bcl6 expression by the essential EBV antigen EBNA3C may provide a novel insight into the current understanding of EBV contribution on lymphomagenesis by blocking GC reaction. Importantly, a number of EBV latent proteins are expressed in EBV infected cells, but how these latent proteins cooperate with each other to regulate B-cell development, or lead to B-cell lymphoma still needs further investigation. Nevertheless, our observations have implications for emerging strategies targeted at the EBV-associated cancers.
The University of Pennsylvania Immunology Core (HIC) provided us human peripheral blood mononuclear cells (PBMC) from different unidentified and healthy donors with written, informed consent. All the procedures were approved by the Institutional Review Board (IRB) and conducted according to the declarations of Helsinki protocols [36,78].
Myc-tagged full-length EBNA3C or its truncations such as 1-365aa, 366-620aa, 621-992aa, and Flag-tagged IRF4 plasmids have been described previously [32]. Myc-tagged constructs expressing full length or DNA binding domain (DBD) mutant IRF4, and HA-tagged full length Bcl6, wild-type Bcl6 promoter plasmids [9,54] were kindly provided by Dr. Riccardo Dalla-Favera (Columbia University, New York, USA). HEK293 or HEK293T (human embryonic kidney cell line), Saos-2 (human osteosarcoma cell line), EBV-negative or -positive cells have been described earlier in detail [32,36]. MEF (mouse embryonic fibroblast cell line) was a gift from Xiaolu Yang (University of Pennsylvania) [79]. HEK293, HEK293T, Saos-2, MEF (p53-/-) and MEF (p53+/+) cells were grown in Dulbecco's modified Eagle's medium (DMEM), while B-cell lines were maintained in RPMI 1640 media. All the above-mentioned cells were incubated at 37°C in a humidified 5% CO2 environment.
The Bcl6 inhibitor (79–6) was purchased from EMD Millipore (Billerica, MA, USA). Bcl6 antibody (N-3) and Bcl2 antibody (C-2) were purchased from Santa Cruz biotechnology (Santa Cruz, CA, USA). Bcl6 antibody (ab19011) were purchased from Abcam (Cambridge, UK). Antibodies for IRF4, Ub, GAPDH have been described earlier [32]. Flag antibody (M2) was purchased from Sigma-Aldrich (St. Louis, MO, USA). Other antibodies to mouse anti-Myc (9E10), anti-HA (12CA5), anti-EBNA3C (A10) were prepared from hybridoma cultures and mentioned previously [80].
10 million transfected cells or 50 million B-cells were harvested, washed with ice-cold 1×PBS twice, lysed in 400μl ice-cold RIPA buffer [1% Nonidet P-40 (NP-40), 10 mM Tris (pH8.0), 2 mM EDTA, 150 mM NaCl, supplement with protease inhibitors (1 mM phenylmethylsulphonyl fluoride (PMSF), 1 μg/ml each aprotinin, pepstain and leupeptin]. Lysates were precleared with normal control serum plus 30 μl of a 2:1 mixture of Protein-A/G Sepharose beads (GE Healthcare Biosciences, Pittsburgh, PA) for 1 h at 4°C. Approximately 5% of the lysate was saved as input. About 1 μg of specific antibody was used to capture the protein of interest by overnight rotation at 4°C.
Input and IP samples were boiled in laemmli buffer, resolved on SDS-PAGE gel and transferred to a 0.45 μm nitrocellulose membrane. The membrane was blocked in 1×TBS-Tween with 5% w/v non-fat dry milk probed with appropriate primary antibody, subsequently incubated with corresponding secondary antibody, and visualized on a Licor Odyssey imager (LiCor Inc., Lincoln, NE). Image analysis and quantification measurements were performed using Image Quant application software (LiCor Inc., Lincoln, NE). The relative density (RD) of indicated proteins were shown.
HEK293T or Saos-2 cells plated on coverslips were transfected with expression plasmids or not as indicated. Forty eight hours post-transfection, cells were fixed by 4% paraformaldehyde (PFA) including 0.1% Triton X-100 for 15–20 mins at room temperature [81]. B-cells were air-dried and fixed similar to above. The fixed cells were washed with 1×PBS for three times, and 5% Bovine serum albumin (BSA) was used for blocking. EBNA3C and Bcl6 were detected by mouse anti-EBNA3C (A10) and rabbit anti-Bcl6 antibody, respectively. The slides were examined using an Olympus Fluoview 300 confocal microscope, and Images were analyzed by Fluoview software (Olympus Inc., Melville, NY).
HEK293T cells were co-transfected with pLA/B9 plasmid (Bcl6 promoter, a gift from Dr. Riccardo Dalla-Favera) [47], pRL-TK (Promega, Madison, WI, USA), Myc-tagged EBNA3C, and control or Myc-tagged IRF4/IRF4-ΔDBD plasmids. Forty eight hours post-transfection, cells were harvested and the dual-luciferase reporter assay was performed according to the manufacture’s protocols (Promega, Madison, WI, USA). At the same time, the supernatant was collected and prepared for detection by western blot.
Cells were collected and washed with ice-cold 1×PBS prior to RNA isolation. Then total RNA extraction was performed using Trizol reagent (Invitrogen, Inc., Carlsbad, CA) and treated with Dnase I (Invitrogen, Inc., Carlsbad, CA), then cDNA was prepared with Superscript II reverse transcriptase kit (Invitrogen, Inc., Carlsbad, CA) according to the manufacturer’s protocol. Primers for GAPDH were 5′-TGCACCACCAACTGCTTAG-3′ and 5′-GATGCAGGGATGATGTTC-3′ [40]. Quantitative Real-time PCR analysis was performed by using SYBR green Real-time master mix (MJ Reserch Inc., Waltham, MA). The assays were performed in triplicate.
Transfected HEK293T cells were treated with protein synthesis inhibitor cycloheximide (CalBiochem, Gibbstown, NJ) after 24 hours transfection as 40 μg/ml concentration. Cells were harvested after 16 hours incubation and lyse with RIPA buffer, then protein samples were quantitated and used for western blot analysis. Protein band intensities were quantified using Image Quant 3.0 software.
10 million HEK293 or Saos-2 cells were transfected with control vector, Myc-EBNA3C, Myc-Bcl6 and GFP vector by electroporation and allowed to grow in DMEM supplemented with 1 mg/ml G418 (Sigma-Aldrich, St. Louis, MO, USA). After two weeks selection, GFP fluorescence of every plate was scanned by PhosphorImager (Molecular Dynamics, Piscataway, NJ) and the area of the colonies measured by using Image J software (Adobe Inc., San Jose, CA). Three independent experiments were performed.
The two sense strands of Bcl6 shRNA are 5’-tcgagtgctgttgacagtgagcgaGCCTGTTCTATAGCATCTTTAtagtgaagccacagatgtaTAAAGATGCTATAGAACAGGCgtgcctactgcctcggaa–3’ (sh-Bcl6-1), and 5’- tcgagtgctgttgacagtgagcgaCCACAGTGACAAACCCTACAAtagtgaagccacagatgtaTTGTAGGGTTTGTCACTGTGGgtgcctactgcctcggaa–3’ (sh-Bcl6-2), respectively. The upper-cases designate Bcl6 target sequences, while lower cases specify hairpin and enzyme sequences. These sense stranded oligos were annealed with their respective anti-sense stranded oligos, and then cloned into pGIPZ vector with Xho I and Mlu I restriction sites. Besides, a negative control was set using a sh-Ctrl plasmid including control shRNA sequence 5’-TCTCGCTTGGGCGAGAGTAAG–3’ (Dharmacon Research, Chicago, IL). Lentivirus production and transduction of B-cell lines has been described previously with a slight modification [32]. A pool of two shRNAs that targeted different regions of the Bcl6 mRNA were co-transfected to generated shRNA-expressing lentiviruses.
The BJAB or LCL1 stable cell lines were generated according to the above-mentioned protocols. Approximately, 5 million BJAB or LCL1 stable cells were collected, fixed with 80% ethanol for 2 hours or overnight at -20°C, then washed with 1×PBS and incubated with PI staining buffer (0.5 mg/ml propidium iodide in 1×PBS, 50 μg/ml RNase A) for 30 minutes to 2 hours at room temperature. The indicated cells were washed with 1×PBS once, resuspended in 500 μl 1×PBS, and analyzed on FACS Calibur (Becton Dickinson, San Jose, CA, USA) using FlowJo software (TreeStar, San Carlos, CA, USA).
The soft agar assays were performed using BJAB or LCL1 cells. Briefly, 1 ml of 0.5% agar in supplemented RPMI media was poured into 6-well plate and set aside to solidify. 0.5 ml 0.3% agar/medium containing 2×105 cells was added to the previously plates as the middle layer. Then cells were covered with a top layer of another 1ml 0.5% agar/medium. After two weeks, colonies were stained with 0.005% crystal violet for 1 hour, and scanned using a Licor Odyssey system (LiCor Inc., Lincoln, NE). The number of colonies was counted using ImageJ software.
Data represented here are the mean values with standard deviation (SD). The significance of differences in the mean values was calculated by performing 2-tailed student's t-test. P-value of <0.05 was considered as statistically significant in all our results (*P < 0.05; **P < 0.01; ***P < 0.001; NS, not significant).
Epstein-Barr virus (EBV) genome, strain B95-8-GenBank: V01555.2; EBNA3C (Human herpesvirus 4)-NCBI Reference Sequence: YP_401671.1; Bcl6 (Homo sapiens)-NCBI Reference Sequence: NM_001130845.1; IRF4 (Homo sapiens)-NCBI Reference Sequence: NM_002460.3; Bcl2 (Homo sapiens)-NCBI Reference Sequence: NM_000633.2; CCND1 (Homo sapiens)-NCBI Reference Sequence: NM_053056.2; p53 (Homo sapiens)-NCBI Reference Sequence: NM_000546.5.
|
10.1371/journal.pgen.1001140 | dMyc Functions Downstream of Yorkie to Promote the Supercompetitive Behavior of Hippo Pathway Mutant Cells | Genetic analyses in Drosophila epithelia have suggested that the phenomenon of “cell competition” could participate in organ homeostasis. It has been speculated that competition between different cell populations within a growing organ might play a role as either tumor promoter or tumor suppressor, depending on the cellular context. The evolutionarily conserved Hippo (Hpo) signaling pathway regulates organ size and prevents hyperplastic disease from flies to humans by restricting the activity of the transcriptional cofactor Yorkie (yki). Recent data indicate also that mutations in several Hpo pathway members provide cells with a competitive advantage by unknown mechanisms. Here we provide insight into the mechanism by which the Hpo pathway is linked to cell competition, by identifying dMyc as a target gene of the Hpo pathway, transcriptionally upregulated by the activity of Yki with different binding partners. We show that the cell-autonomous upregulation of dMyc is required for the supercompetitive behavior of Yki-expressing cells and Hpo pathway mutant cells, whereas the relative levels of dMyc between Hpo pathway mutant cells and wild-type neighboring cells are critical for determining whether cell competition promotes a tumor-suppressing or tumor-inducing behavior. All together, these data provide a paradigmatic example of cooperation between tumor suppressor genes and oncogenes in tumorigenesis and suggest a dual role for cell competition during tumor progression depending on the output of the genetic interactions occurring between confronted cells.
| One of the major challenges of developmental biology and cancer research is to get a better understanding of how different signals regulate proper organ growth and prevent tumor formation. Even though there is a strong correlation between tumor progression and Myc family misexpression or Hippo signaling pathway malfunction, the relationship between these organ growth regulators remains unclear. Here, we demonstrate that the Hippo signaling pathway controls the transcription of Drosophila dmyc. Furthermore, we show that the misregulated expression of dMyc in Hippo mutant cells elicits their proliferative expansion at the expense of normal surrounding cells. These findings reveal a molecular mechanism of cooperation between oncogenes and tumor suppressor genes that favors both tumor progression and wild-type tissue elimination. Additionally, our findings indicate a dual role for cell competition during the tumour progression depending on the cellular context.
| Growth regulation requires the fine tuning between the rate of cell death and cell proliferation in developing organs. Studies in Drosophila have revealed that somatic cells within a growing epithelium compete with one another for contribution to the adult organ and this phenomenon, known as “cell competition” [1], is possibly conserved among organisms, for a review [2]. Cell competition was discovered several decades ago comparing the clonal growth parameters of Drosophila wild type cells (+/+) and slow-dividing Minute/+ cells [1]. From those analyses and recent data [3], it has been concluded that the contact between wild type and slow-growing cells, in genetic mosaics, favors the positive selection and clonal expansion of faster cells (winners) at the expense of slow-dividing ones (losers), although eventually the final number of cells in the organs is unaffected [3]. The biological function of cell competition remains unclear but it is thought to contribute to tissue homeostasis by coordinating the rate of cell proliferation and cell death [4], [5]. One of the best examples illustrating cell competition was obtained from the analysis of Drosophila myc [4], [5], opening to the speculation that this phenomenon might play a role in tumorigenesis [2], [6], however the basis of cell competition in tumorous situations has just begun to be investigated [7]. dmyc is an evolutionarily conserved proto-oncogene associated with different cellular processes, including cell cycle progression, cell growth and apoptosis [8]–[11]. The function of dMyc protein is both necessary and sufficient to control rRNA synthesis and ribosome biogenesis [12]. In Drosophila, cells carrying hypomorphic alleles of dmyc are viable in a homotypic context, but they are outcompeted and excluded from the epithelium when surrounded by wild type cells [5]. By contrast, dmyc overexpressing cells become “supercompetitors” able to kill wild type surrounding cells [4], [5]. Remarkably, dMyc upregulation is related with many types of human cancers [13] and it favors the clonal expansion of cells carrying additional oncogenic mutations [14], [15].
During the last years, the Hippo (Hpo) tumor suppressor pathway has emerged as a safeguard system restricting organ growth and preventing hyperplastic disease in metazoans [16], [17]. Mutations in several members of this pathway have been associated with tumor formation both in Drosophila and in humans [18]. It has also been reported that mutations in many members of the Hpo pathway can rescue the viability of heterozygous M/+ cells in genetic mosaics [19], suggesting that these mutant cells behave as “supercompetitors”. Therefore the detailed analysis of Hpo pathway members appears to be an attractive model in which to evaluate the relationship between cell competition and tumor growth, as well as the molecular mechanisms required for this crosstalk. Hpo, Salvador (Sav) and Warts (Wts) constitute the core of the Hpo pathway that regulates by phosphorylation the downstream transcriptional co-activator Yorkie (Yki) [18], [20]. The hyperphosphorylated form of Yki is retained in the cytoplasm [21], [22], thereby preventing the expression of several target genes involved in cell proliferation control (Cyclin E, E2F1, bantam miRNA) [16], [23]–[25], cell death (dIAP1) [16] and cell signaling regulation (dally and dally-like) [26]. It has been demonstrated that Yki regulates its target genes by binding to Scalloped (Sd), a TEAD/TEF family transcription factor [27]–[30]. In addition, recent data indicate that Yki is also able to bind to the homeoprotein Homothorax (Hth) forming a complex which regulates the transcription of bantam in the eye disc [31]. The atypical cadherins Fat (Ft) [26], [32]–[37] and Dachsous (Ds) [20], [26], [33], [38], as well as the FERM-domain proteins Expanded (Ex) and Merlin (Mer) [39], have also been implicated in the pathway as upstream components. Although their biochemical functions are still uncertain, it is assumed that they converge on Wts to regulate Yki activity [40], [41].
Here we provide a detailed analysis of the autonomous and non-autonomous effects on growth of yki-expressing cells and mutations of members of the Hpo pathway. In addition we show that dmyc is a transcriptional target of Yki, able to confer competitive properties to the Hpo pathway mutant cells in the Drosophila wing. Furthermore, dmyc upregulation is essential to sustain the high rate of cell proliferation of Hpo mutant cells and to protect them from being eliminated in a competitive background. Finally, we show that the relative levels of dMyc protein between neighboring cells are critical in order to define the role of cell competition during tumor progression.
In order to analyze the competitive properties of Hpo pathway mutant cells, we used mosaic analysis to compare the size of yki overexpressing clones (hereafter referred to as ykiover) with their wild type twins. While clones and twins showed a comparable size in the wild type control (Figure 1A, 1F, and 1G, and Figure S1A, S1C, S1D), ykiover clones were notably larger than their wild type twins in wing discs dissected either 60h (Figure 1D, 1H, and 1I) or 48h (Figure S1B, S1E, S1F) after heat-shock induction. Furthermore, ykiover wild type twins were almost disappeared from the epithelium at 120h after egg laying (AEL) (Figure 1B and 1C). These differences in size were also prominent when discs were dissected at 96h AEL (Figure 1D and 1E). Interestingly, the clonal expansion of ykiover cells was also correlated with non-autonomous apoptosis, as revealed by active Caspase 3 immunoreactivity of a subset of surrounding wild type cells (Figure 1B–1E). The size advantage of ykiover clones and the induction of apoptosis in wild type cells is consistent with the broadly assumed definition of cell competition, which implies that the clonal expansion of the winner cells occurs at the expense of the juxtaposed losers, that are eliminated by apoptotic death [2], [42], [43]. The pattern of cell death in wild type and ykiover cells (Figure 1B–1E) was not confined to the interface between the two cell types; as can be seen in Figure 1E, cell death extends several cell diameters away and wild type cells tend to die massively when enclosed between nearby mutant clones (Figure 1E, yellow arrowhead). A similar pattern of non-autonomous cell death was observed in wild type cells nearby mutant clones for other members of the Hpo pathway, such as ft and ex (Figure S2A, S2B, S2C). Strikingly, ykiover clones and wts mutant clones grown for a longer period presented autonomous cell death (Figure 1C, see active Caspase 3 staining, and Figure S2D), despite the upregulation of anti-apoptotic molecules such as dIAP1 [16]; this might be possibly due to either developmental constraints compensating for excessive proliferation of the entire organ or toxicity caused by high and constant levels of Yki. Altogether, these results confirm the previously suggested supercompetitive properties of the Hpo pathway mutant clones [19] by revealing their ability to overgrow and eliminate surrounding wild type cells.
It is well documented that the confrontation of different levels of dMyc protein between two populations of cells either in vivo [4], [5] or in cell culture [44] can trigger cell competition, however the molecular mechanism by which this occurs is unknown. In addition, myc family oncogenes are frequently overexpressed in human cancers and it contributes to tumor progression of YAP-expressing cells (mammalian orthologue of yki) [17]. We have previously shown that a transcriptional activation of dmyc occurs in ft mutant tissues and that ft clones fail to grow in a dmyc hypomorphic background [45], indicating a possible regulation of this oncogene by the Hpo pathway. Moreover, the expression pattern of dMyc is complementary to that of Ds in the wing imaginal disc (Figure 2A), suggesting a possible functional interaction. To validate this hypothesis, we analyzed dMyc expression in mutant clones for several members of the Hpo pathway and in ykiover cells by immunofluorescence. Noticeably, we found that dMyc was upregulated in a cell-autonomous manner in ykiover clones throughout the wing disc (Figure 2B and Figure S3), with the weakest activation in the lateral regions, and in a subset of clones mutant for several Hpo pathway members (Figure 2C–2F). These differences in dMyc activation between ykiover clones and clones mutant for other members of the Hpo signaling pathway might be due to additional levels of regulation of the Hpo cascade operating on upstream members. According to our previous observations, we would predict a repression of dMyc upon Hpo pathway hyperactivation. To investigate this hypothesis, we expressed Hpo in the spalt expression domain of the developing wing disc. Since Hpo overexpressing cells die massively by apoptosis during development [25], we coexpressed the anti-apoptotic factor p35. As expected, cells coexpressing Hpo and p35 show reduced levels of dMyc with respect to the control (Figure S4A) in both late (Figure S4B) and early (Figure S4C) wing discs. Thus dMyc levels can be regulated by the Hpo pathway activity.
dmyc was observed upregulated in RT-PCRs performed on ft mutant imaginal discs [45], suggesting that it could be a transcriptional target of the Hpo pathway. In order to investigate this, we first performed an in situ hybridization in Drosophila wing discs expressing yki under the control of the decapentaplegic (dpp) promoter. As expected, dmyc transcript is detectable in the dpp domain both in yki and control dmyc-expressing discs (Figure 3A). No signal within the dpp domain was detected in dpp>GFP control discs (not shown). We were able to reproduce these data using a dmyc>lacZ line [46] which recapitulates accurately the dmyc pattern throughout the wing disc during development [7], [47]. As can be seen in Figure 3B, the ßGal expression is increased in the dpp domain upon yki expression, indicating that Yki acts upon dmyc transcription. This result was supported using clonal analysis, both in ykiover cells, as shown in Figure 3C, and in cells mutant for ft (Figure S5). Altogether, these data demonstrate the ability of the Hpo pathway to regulate dmyc transcription in the imaginal wing disc.
Yki transcriptional activity depends on the formation of tissue-specific complexes with different partners such as Scalloped and Homothorax [27]–[31]. In order to study the contribution of Sd to dmyc upregulation by Yki in the wing disc, we generated ykiover clones coexpressing either a UAS-sd or a UAS-sd-RNAi construct (see Figure S6A for validation). As can be seen in Figure 3D, sdover; ykiover clones overgrew relative to ykiover clones (compare with Figure 2B, 68% increase on average, n = 27, P<0,005) confirming previous data [29], but we were not able to detect significant differences in dMyc protein levels compared to ykiover clones (n = 22, P = 0,43). As expected, control sdover clones did not overgrow and did not deregulate dMyc (Figure 3E), demonstrating that Yki is required for dMyc upregulation. We were not able to recover sd-RNAi; ykiover clones in the wing pouch region, but clones generated in other territories of the wing disc, although large, did not upregulate dMyc (Figure 3F), nor showed the same degree of hyperplasia as Yki expression alone (Figure 1B–1E). sd-RNAi control clones were very small and did not deregulate dMyc (not shown). These data indicate a key role for Sd in vivo in upregulating dMyc in ykiover clones, and in contributing to the ykiover tumorous phenotype.
Interestingly, examination of dmyc locus revealed the existence of several CATTCCA repeats in non-coding regions of the gene, which perfectly match the mammalian [48], [49] and Drosophila [28], [29] TEAD/TEF family transcription factor consensus binding motifs (mammaliam orthologues of Scalloped). In addition, these putative binding motifs for Yki/Sd complexes are evolutionarily conserved in D. simulans (Figure 3G) and relatively close to the insertion point of P elements that recapitulate the endogenous expression of the gene (dmPL35 LacZ [50], [51] and dmBG02383 Gal4 insertions - http://flybase.org/reports/FBti0018138.html).
To test the significance of these sequences in dmyc regulation, we generated a dmyc-firefly reporter containing the putative responsive elements for Yki/Sd complexes (Figure 3H) and performed a transient dual luciferase assay in S2 cells. As can be seen in Figure 3I, the reporter was specifically activated upon Sd and Yki cotransfection but, unexpectedly, the transfection of Yki alone was able to activate the reporter as efficiently as the cotransfection Yki/Sd (Figure 3I). This result suggests that in presence of high levels of Yki alone, additional partners such as Hth [31] could bind it and co-regulate dmyc expression.
Indeed, complementarily to Yki/Sd complexes, Yki/Hth complexes seemed to play the same role in the presumptive thoracic region of the wing disc. Supporting this conclusion, hth-RNAi; ykiover clones down-regulated dMyc in the notum (30% reduction on average, n = 15, P<0,05, Figure S6B, yellow arrows) and did not grow as tumors in that region. By contrast, they were undistinguishable from ykiover clones in the wing pouch region (Figure S6B, white arrowhead), where Hth expression is almost undetectable (Figure S6C). Altogether, these latter results indicate that Sd and Hth play a role in Yki-induced tumorigenesis by regulating dmyc expression in the wing disc, with Sd playing a more critical role in the pouch and Hth acting in the presumptive thorax.
With the aim to investigate the cell-autonomous contribution of dMyc overexpression to ykiover phenotypes, we first compared the size of ykiover clones with that of ykiover; dmyc-RNAi clones (Figure 4, see also Figure S7A, S7A′, and [7] for RNAi construct validation ). As expected, dmyc-RNAi clones showed a reduced number of cells with respect to that observed in wild type clones (21% reduction on average, compare Figure 4B and 4B′ with Figure 4A and 4A′, P<0.01). The reduction in cell number displayed by the ykiover; dmyc-RNAi clones with respect to the ykiover clones was even more evident (43% reduction on average, compare Figure 4D and 4D′ with Figure 4C and 4C′, P<0.01), and this percentage raised up to 65% (n = 87, P<0,001) when these clones were induced earlier in development (42–54h AEL), indicating a strong cell-autonomous requirement of dMyc protein for the expansion of ykiover clones. We also observed that the non-autonomous apoptosis induced by yki overexpression was reduced upon dmyc deprivation (32% on average, n = 28, P<0,01, Figure S7B). These data suggest that dMyc upregulation promotes cell proliferation of ykiover clones in an autonomous manner, and also promotes their competitive behavior.
To further characterize this proliferation-promoting effect of dMyc, we compared the clonal behavior of various mutations in members of the Hpo pathway grown in two different genetic backgrounds: a wild type context and a genetic background overexpressing dmyc under the control of a hedgehog promoter in the posterior (P) compartment of the wing disc. We found that ft, ex and ds mutant clones were consistently larger in those territories expressing uniform levels of dMyc than in the wild-type background (Figure 5 and Figure S8). It is however described that the overexpression of dMyc is able to autonomously increase apoptosis [8]–[11]. In fact, the wild type tissue expressing high amounts of dMyc tends to die and does not overgrow (see active Caspase 3 stainings in Figure 5A and 5D). Noticeably, the apoptosis mediated by dMyc overexpression seems to be extremely reduced inside ft and ex clones (Figure 5A and 5D) with respect to the wild type surrounding territories, likely due to the upregulation of antiapoptotic genes such as dIAP1, a target of the Hpo pthway [20]. In addition, the dying cells in this genetic background might induce morphogens to promote compensatory proliferation [52] that may contribute to the extra-growth of ft- or ex-UAS-dmyc expressing clones. To circumvent this problem, we repeated the same experiment coexpressing dmyc and dIAP1. As can be seen in Figure S9, both ft (Figure S9A) and ex (Figure S9D) mutant clones grown in the P compartment were still consistently larger than those originated in the A compartment, thus confirming a specific cooperation of dmyc and Hpo pathway mutants in clonal expansion.
Hpo mutant cells therefore seem to show the ability to take advantage of the cell mass accumulation boosted by dMyc overexpression to proliferate faster.
To address the non-autonomous relevance of dmyc upregulation in providing ykiover cells with a supercompetitive behavior, we compared the size of ykiover clones generated in a wild type background to that of ykiover clones generated in a background ubiquitously overexpressing dmyc under the control of a tubulin (tub) promoter (cell competition assay, [4], [5]). In this assay cells express the endogenous dmyc gene plus an extra copy of the gene under the control of a tub promoter that ensures two-to-threefold increase of dmyc transcript [5]. This extra copy of dmyc is located in a removable cassette between the tub promoter and a Gal4 cDNA. Upon dmyc cassette excision, the tub promoter drives Gal4 expression in the clones and, as a result, those cells express lower levels of dmyc relative to the background and are rapidly eliminated from the tissue by cell competition. Only few genes have so far been found whose overexpression rescues cell viability in this context [5]. The relative difference in dMyc levels between yki-expressing cells and the surrounding tub>dmyc cells was minimized in a competitive background compared to a wild type context (compare Figure S7C and S7C′ with Figure 2B). In this competitive background, ykiover clones showed a diminished ability to overgrow compared to a wild type background (44% reduction on average, compare Figure 6C and 6C′ and Figure 6B and 6B′; P<0,01). Besides the reduction in size, ykiover clones showed an important reduction in clone number both in discs (Figure 6C) and adult wings (compare Figure S7E to Figure S7D). Moreover, ykiover clones induced earlier in development (42–54h AEL) were never recovered at the end of larval development (not shown). These data indicate that the competitive properties of ykiover cells are extremely reduced when they are surrounded by cells expressing very high amounts of dMyc.
We then performed the same competition assay as before while reducing dmyc activity inside the clones. We used the pupal lethal dmycPL35 allele [49] and, taking advantage of dmyc locus association to chromosome X, we were able to analyze both female (heterozygous condition, the expression of dmyc is halved) and male (hemizygous condition, the expression of dmyc is completely removed) larvae. In dmycPL35/+; tub>dmyc females, ykiover clones were smaller than those described in the previous assay (28% reduction on average, compare Figure 6D and 6D′ to Figure 6C and 6C′, P<0,05), whereas they were completely outcompeted by 48h after the heat shock in males (not shown). Since it has been observed that a dmycPL35 heterozygous condition does not impair cell growth or proliferation rate [49], our results reveal an important role for dmyc-induced cell competition in controlling the clonal expansion of ykiover cells, which may occur via their non-autonomous capabilities to compete with neighboring wild type cells.
yki LOF clones generated in a wild type background are not able to grow [16], [25] and the ectopic expression of the antiapoptotic proteins dIAP1 [25] or p35 (Figure S10A) poorly rescues their viability, whereas a Minute background [53] or bantam overexpression within yki clones has been shown to partially rescue their growth [25]. Since our results have indicated that dmyc participates in tumor growth of the Hpo pathway mutant cells, we therefore analyzed if the expression of dMyc was sufficient to prevent the death of yki mutant cells. The overexpression of dMyc failed to rescue the viability of yki−/− cells (Figure S10B). Since yki mutant cells express low levels of the apoptosis inhibitor dIAP1 (not shown), this result is not surprising, considering the autonomous cell death described for cells overexpressing dMyc [11]. However, yki mutant cells coexpressing dMyc and p35 also failed to grow (Figure S10C). The lack of expression of additional antiapoptotic genes and cell cycle regulators [18] possibly impedes the clonal growth of yki mutant cells even though they overexpress dMyc. This result suggests that dmyc expression is able to enhance the ability of Hpo pathway mutant cells to grow, but it is not sufficient to rescue tissue growth of yki−/− clones.
Cells within a tissue coordinate and execute complex genetic programs in order to succeed in completing a variety of processes during development. In this context, the phenomenon of cell competition may be part of the developmental plan that ensures removal and replacement of defective cells in growing organs, thus keeping their size invariant. In this work, we have evaluated in details the relationships between the phenomenon of cell competition and the clonal expansion of tumorous cells, using for that purpose mutants in components of the evolutionarily conserved Hpo pathway. From our studies we reveal that the Hpo pathway regulates dMyc expression, and show that this is critical for the tissue growth and competitive behavior of Hpo pathway mutant clones.
dmyc upregulation has been demonstrated in many studies to provide cells with supercompetitive properties [4], [5], [7]. The model explaining how dMyc can confer competitive properties to cells is based on the relative levels of this protein in neighboring cell populations, transforming those cells expressing higher levels of dMyc into supercompetitors [4], [5]. dmyc overexpression is nevertheless insufficient to drive tumorous growth; dmycover clones fail to overproliferate and show strong autonomous apoptosis [9]. Interestingly, we found that dMyc protein is overexpressed in Hpo pathway mutant clones, indicating an involvement for this cascade in dmyc regulation (Figure 2). Furthermore, the upregulation of dMyc in Yki-expressing cells correlates with an increase in the amount of mRNA, observed by in situ hybridization (Figure 3A) and using a dmyc>lacZ line (Figure 3B and 3C). Finally, we have identified a regulatory region in the second intron of dmyc that is sensitive to Yki abundance; importantly, this regulatory region includes predicted consensus-binding motifs for Sd (Figure 3H). Clonal experiments in the wing disc indicate that Sd is necessary for Yki function in vivo, since upon Sd downregulation Yki is no longer able to induce tumorous growth and does not upregulate dMyc (Figure 3F). All these findings support the notion that there is a transcriptional regulation of dMyc mediated by Yki/Sd complexes in the wing pouch. Importantly, similar results were observed for dMyc regulation in the notum by Yki/Hth complexes, suggesting that tumor growth and dmyc regulation are tissue-specific.
We found that dMyc upregulation is a common feature of Hpo pathway mutant cells. Since dmyc has been repeatedly associated with tumor progression and cell competition, we analyzed its role in the clonal expansion of Hpo pathway mutant cells. We observed that the reduction of dMyc expression restricts the ability of Hpo pathway mutant cells to proliferate (Figure 4), whereas its uniform overexpression strongly promotes their proliferation (Figure 5). Furhermore, while dMyc-expressing wild type cells surrounding mutant clones are rapidly eliminated by autonomous apoptosis, Hpo pathway mutant cells are able to take advantage of dMyc role in protein biosynthesis and cellular growth to divide rapidly. This is a clear example of functional cooperation between different genes in order to favor tumor progression, but it also indicates a specific role of dMyc in promoting the clonal expansion of Hpo pathway mutant cells. According to these data, we conclude that dMyc behaves as a growth-promoting factor which sustains the hyperplastic phenotype of Hpo pathway mutant cells. Importantly, this specific cooperation might be evolutionarily conserved, since c-myc appears to be upregulated in a murine model of YAP-induced carcinoma [17].
It has been suggested that cell competition may be a mechanism potentially restricting the clonal expansion of tumorous cells [7], but it might also help faster proliferation of transformed cells. Our data indicate that Hpo pathway mutant cells are able to use high levels of dMyc to proliferate rapidly (Figure 5), but in a competitive context, where neighboring cells express high levels of dMyc, clonal expansion of ykiover cells is restrained (Figure 6), therefore suggesting a tumor suppressor role for cell competition. Conversely, dMyc upregulation in ykiover clones grown in a wild type background favors their clonal expansion promoting cell autonomous proliferation and also conferring the ability to outcompete sourrounding cells in a non-autonomous manner. These findings suggest that the phenomenon of cell competition may play a dual role in tumor progression depending on the output of the genetic interactions occurring between adjacent cells.
In summary, we have shown a tumor-braking gene network in Drosophila epithelia which tightly controls cell proliferation, apoptosis and cell competition via the Hpo pathway and dMyc expression. Importantly, YAP deregulation has been reported in several types of human cancers [54]–[56], therefore the mechanism of clonal expansion of Hpo pathway mutant cells in Drosophila might be relevant to understand tumor progression in mammals.
The fly strains used in the present work were obtained by the Bloomington Stock Center and are described at http://flybase.bio.indiana.edu. The following strains were instead obtained by: w; UAS-yki (D Pan); yw, tubFRTdmycFRTGal4 and yw, dmycPL35, actFRTy+FRTGal4 (P Gallant); w, hs-FLP; actFRTy+FRTGal4, UAS-GFP (B Edgar); w; FRT40A, dsD36 (I Rodríguez). The UAS-RNAi constructs for dmyc, sd and hth were obtained from the VDRC.
All experiments were carried out at 25°C unless otherwise indicated.
MARCM UAS-yki twin-spot clones were induced at different stages of development by a 35-minutes heat shock at 37°C and larvae of the following genotype were dissected at either 84-100h AEL or 120h AEL: yw, hs-Flp, tub-Gal4, UAS-GFP; FRT42D, tub-Gal80/FRT42D, Ubi-GFP; UAS-yki/+. Clones of the same genotype were induced 54–66 h AEL and dissected 48h after a 20-minutes heat shock (Figure S1). For FRT-Flp twin analysis, the following hypomorphic or null alleles were used: dsD36, ftG-rv, exE1, wtsX1, ykiB5. Loss-of-function clones of ds, ft, ex and wts in either wild-type or mutant backgrounds overexpressing different transgenes in the posterior compartment were induced at 48–72h AEL by 1 hour heat shock at 37°C. Larvae of the following genotype were dissected at 120h AEL:
yw, hs-Flp; FRT40A, Ubi-GFP/FRT40A, dsD36 or ftG-rv or exE1
yw, hs-Flp; FRT82B, Ubi-GFP/FRT82B, wtsX1
yw, hs-Flp; FRT40A, Ubi-GFP/FRT40A, dsD36 or ftG-rv or exE1; hh-Gal4/UAS-dmyc
yw, hs-Flp; FRT40A, Ubi-GFP/FRT40A, ftG-rv or exE1; hh-Gal4/UAS-dmyc, UAS-dIAP1
The size of non-confluent clones was measured drawing each Z-stack of the confocal images using ImageJ software (http://rsbweb.nih.gov/ij). Afterwards the area of the clones was normalized dividing by the area of the wing pouch, considered as the territory encircled by the first outer folding of the wing. In Figure S1, the narrower window of clonal induction allowed us to compare clonal size without size normalization respect to the wing pouch. Statistical analysis was performed with Microsoft Excel and R (www.r-project.org). Statistical significance was determined by two tailed Student's t test and reported as the associated probability value (P).
Flp-Out clones were induced at 60h AEL by a 8-minutes heat shock at 37°C; imaginal discs of the following genotype were dissected at 120h AEL:
yw, hs-Flp; actFRTy+FRTGal4, UAS-GFP
yw, hs-Flp; UAS-dmycRNAi/+; actFRTy+FRTGal4, UAS-GFP/+
yw, hs-Flp; actFRTy+FRTGal4, UAS-GFP/UAS-yki
yw, hs-Flp; UAS-dmycRNAi/+; actFRTy+FRTGal4, UAS-GFP/UAS-yki.
yw, hs-Flp/w, dmyc>lacZG0354; actFRTy+FRTGal4, UAS-GFP/UAS-yki.
Cell competition assays were performed at 72h AEL inducing a 40-minutes heat shock at 36°C. Larvae of the following genotype were dissected at 120h AEL:
yw, tubFRTy+FRTGal4/hs-Flp; UAS-GFP/+
yw, tubFRTy+FRTGal4/hs-Flp; UAS-GFP/+; UAS-yki/+
yw, tubFRTdmycFRTGal4/hs-Flp; UAS-GFP/+; UAS-yki/+
yw, dmycPL35, hs-Flp, tubFRTdmycFRTGal4/+-Y; UAS-GFP/+; UAS-yki/+.
MARCM yki clones overexpressing p35, dMyc or both were generated at 48–72h AEL by a 45-minutes heat shock at 37°C and larvae were dissected 48h later.
Immunostainings were performed using standard protocols. The following primary antibodies were used: mouse anti-dMyc (1∶5, P Gallant), mouse anti-En (1∶50, DSHB), rabbit anti-active Caspase 3 (1∶100, Cell Signaling Technology), rabbit anti-p35 (1∶1000, Stratagene), rabbit anti-Ds (1∶100, D Strutt), rabbit anti-Hth (1∶400, A Salzberg, [57]), mouse anti-dIAP1 (1∶100, B Hay) and rabbit anti-ßGal (1∶400, F Graziani). Anti-mouse and anti-rabbit Alexa Fluor 555 (1∶200) (Molecular Probes) and anti-mouse Cy5 (1∶200) (Jackson Laboratories) against corresponding primary antibodies were used as secondary antibodies. Imaginal discs were mounted in Vectashield (Vector Laboratories) for confocal imaging. Single Z stacks were acquired with Leica SP2 and SP5 confocal microscopes. Images for Figure 4 and Figure 6 were captured with an epifluorescence Nikon 90i microscope. Entire images were elaborated with Photoshop CS2 (Adobe) and the projections along the Z axis were rebuilt starting from 35–55 Z stacks using the ImageJ public software (NIH). For measurements of dMyc abundance, fluorescence intensity was calculated using the ImageJ public software (NIH) as the average gray value within selectioned portions of confocal Z stacks. For measurement of active Caspase 3 signal outside UAS-dmyc-RNAi; UAS-yki and UAS-yki clones, staged wing discs were chosen containing as few clones as possible and single cells positive to active Caspase 3 observed at a maximum distance of five nuclei (counterstained with DAPI) from the border of the clone were counted on confocal Z stacks. In situ hybridization was performed with a full length dmyc probe [9] on wing imaginal discs of L3 larvae expressing UAS-GFP, UAS-dmyc or UAS-yki under the control of dpp-Gal4. RNA in situ hybridization was carried out using digoxigenin-labeled RNA probes [58].
Drosophila S2 cells were grown at 25°C in Schneider medium (GIBCO) supplemented with 10% heat-inactivated FCS and 100 units of penicillin.
1189 base pairs located in the second intron of the dmyc sequence (Figure 3H) were subcloned into a pGL3-firefly vector (Promega) and co-transfected with Sd and/or Yki-expressing pAc5.1/V5-HisB plasmids [28] using Effectene Qiagen Transfection Kit. The primers used for that purpose were:
5′ CAGCGGTACCAGTTTGCTGTCCTCTGC 3′
5′GCACTCTAGAGCCATGCGGAATTGTGCG 3′.
The PCR product was first cloned in pCR 2.1 TOPO-TA (Sigma) and then subcloned in KpnI/XhoI sites of pGL3 Promoter vector. For luciferase transient expression assays, 2×104 cells were plated in 96-well dishes. Cells were harvested at 48 hours after transfection and luciferase activity was measured using the Dual-Luciferase reporter assay system (Promega). Dual-Luciferase measurements were performed using a FLUOstar Optima luminometer (BMG Labtech) and normalized to the Renilla luciferase activity using pAct5C-seapansy as an internal control. All transient expression data reported in this paper represent the means from three parallel experiments, each performed in triplicate. Average relative luciferase activity was graphed and statistically analyzed by the Student's t-test.
|
10.1371/journal.pntd.0000547 | Molecular Characterization of the Schistosoma mansoni Zinc Finger Protein SmZF1 as a Transcription Factor | During its development, the parasite Schistosoma mansoni is exposed to different environments and undergoes many morphological and physiological transformations as a result of profound changes in gene expression. Characterization of proteins involved in the regulation of these processes is of importance for the understanding of schistosome biology. Proteins containing zinc finger motifs usually participate in regulatory processes and are considered the major class of transcription factors in eukaryotes. It has already been shown, by EMSA (Eletrophoretic Mobility Shift Assay), that SmZF1, a S. mansoni zinc finger (ZF) protein, specifically binds both DNA and RNA oligonucleotides. This suggests that this protein might act as a transcription factor in the parasite.
In this study we extended the characterization of SmZF1 by determining its subcellular localization and by verifying its ability to regulate gene transcription. We performed immunohistochemistry assays using adult male and female worms, cercariae and schistosomula to analyze the distribution pattern of SmZF1 and verified that the protein is mainly detected in the cells nuclei of all tested life cycle stages except for adult female worms. Also, SmZF1 was heterologously expressed in mammalian COS-7 cells to produce the recombinant protein YFP-SmZF1, which was mainly detected in the nucleus of the cells by confocal microscopy and Western blot assays. To evaluate the ability of this protein to regulate gene transcription, cells expressing YFP-SmZF1 were tested in a luciferase reporter system. In this system, the luciferase gene is downstream of a minimal promoter, upstream of which a DNA region containing four copies of the SmZF1 putative best binding site (D1-3DNA) was inserted. SmZF1 increased the reporter gene transcription by two fold (p≤0.003) only when its specific binding site was present.
Taken together, these results strongly support the hypothesis that SmZF1 acts as a transcription factor in S. mansoni.
| Schistosomes are parasites that exhibit a complex life cycle during which they progress through many morphological and physiological transformations. These transformations are likely accompanied by alterations in gene expression, making genetic regulation important for parasite development. Here we describe a Schistosoma mansoni protein (SmZF1) that may act as a parasite transcription factor. These factors are key proteins for gene regulation. We have previously demonstrated that SmZF1 is able to bind DNA and that its mRNA is present at different stages during the parasite life cycle. In this study we aimed to define if this protein can function as a transcription factor in S. mansoni. SmZF1 was detected in the nucleus of adult male worms, cercariae and schistosomula cells. It was not, however, observed in female cells, suggesting it to be gender specific. We used mammalian cells expressing recombinant SmZF1 to analyze if SmZF1 protein is able to activate/repress gene transcription and demonstrated that it increased the expression of a reporter gene by two-fold. The results obtained confirm SmZF1 as a S. mansoni transcription factor.
| Schistosomiasis is a disease caused by trematode worms, mainly Schistosoma mansoni, S. haematobium and S.japonicum. According to World Health Organization, this parasitic disease affects 200 million people throughout the world [1]. Although the level of schistosome-associated morbidity is unclear, some recent studies have demonstrated that the illness is a more serious problem than it was previously thought to be [2],[3]. Therefore, emphasis should be focused on mechanisms that could not only prevent, but also cure schistosomiasis. A useful approach to fight the disease should include infrastructure and educational components, as well as the development of vaccines and new drugs [4]. Luckily we are living a special moment, with the recent publication of both S. mansoni [5] and S. japonicum [6] genomes, which will bring to the scientific community an enormous amount of data to be mined in the search for new therapeutic targets and vaccine development. Lastly, additional effort should also be dedicated to studies regarding the biology and development of the parasite.
During its life cycle, S. mansoni is exposed to different environmental conditions: water, intermediate molluscan host, and a definitive vertebrate host. As a consequence, this parasite suffers many transformations in its morphology and physiology, and, as such, represents an interesting but challenging biological system to investigate gene regulation processes [7]–[9]. A variety of publications have focused on the identification and characterization of S. mansoni stage-, tissue- and sex-specific/abundant proteins and their coding genes [10]–[14], which may uncover hidden aspects of parasite biology and thus provide useful leads for the development of novel intervention strategies [7]. In a primary analysis of the S. mansoni transcriptome, Verjovski-Almeida and colleagues suggested that the number of differentially expressed genes could reach as many as 1000 for each stage [15]. In more recent publications, in which analyses of gene expression were carried out using microarray, SAGE (Serial Analysis of Gene Expression) and proteomic experiments, the authors confirmed a number of sex- and stage-specific, differentially expressed genes [8], [16]–[26].
In order to better understand the transcriptional regulation of S. mansoni genes, it is necessary to identify new transcription factors, coactivators/corepressors and chromatin remodeling factors that control this molecular process, along with regulatory elements in the promoter region of genes [9]. Several efforts to describe new transcription factors in this parasite have been made [27]–[31], but given the complexity of its life cycle there are still many components to be discovered and characterized. Zinc finger motifs are found in several proteins amongst eukaryotic organisms and are key proteins for transcription regulation [32]–[34]. SmZF1 is a S. mansoni 19 kDa protein (GenBank accession number AAG38587) containing three C2H2 type zinc finger motifs. Its cDNA was casually isolated from an immune screening of a S. mansoni adult worm lambda gt11 expression library using an anti-tegumental serum. The transcript coding for SmZF1 was also detected by PCR amplification in egg, cercaria, schistosomulum and adult worm cDNA libraries, suggesting that the protein is essential for metabolism during different stages of the parasite life cycle [35]. In a previous work, we used a recombinant SmZF1 protein in EMSA experiments to investigate its binding capacity/specificity for DNA and RNA oligonucleotides. SmZF1 was found to bind both double and single-stranded DNA, as well as RNA oligonucleotides, but with about 10-fold lower affinity. Although we noticed that SmZF1 recognized DNA and RNA oligonucleotides not containing putative target sites, the protein bound preferentially to the ones containing the sequence 5′-CGAGGGAGT-3′ (oligonucleotide D1-3DNA). Furthermore, unrelated oligonucleotides were not able to abolish this interaction. Taken together, these initial results suggested that SmZF1 may act as a putative transcription factor in S. mansoni [36].
In order to better characterize the biological function of the SmZF1 protein, in this study we proposed to: (i) verify the subcellular localization of SmZF1 in the cells of S. mansoni, as well as in mammalian COS-7 cells expressing a recombinant YFP (Yellow Fluorescent Protein)-SmZF1 protein; (ii) test the ability of SmZF1 to activate or repress gene transcription. The results described herein define SmZF1 as a S. mansoni nuclear protein capable of activating gene transcription.
In order to obtain anti-SmZF1 antibodies, the MBP (Maltose Binding Protein) portion of a MBP-SmZF1 recombinant protein [36] was cleaved using Factor Xa protease (New England Biolabs, Ipswitch, MA, USA). The cleavage reaction was carried out for 48 h at 4°C in a 1∶25 enzyme: protein proportion. After digestion and fractionation by electrophoresis, a Coomassie blue-stained protein band (450 µg), representing the SmZF1 portion of the recombinant protein was excised from a 10% SDS-PAGE, homogenized with PBS (Phosphate Buffered Saline – 130 mM NaCl, 2 mM KCl, 8 mM Na2HPO4, 1 mM KH2PO4), then emulsified with Complete Freund Adjuvant and used for the primary intramuscular injection into a rabbit or with Incomplete Freund Adjuvant for the two subsequent boosts (15 and 30 days after the first immunization). Pre-immune serum was obtained before the first immunization and rabbit serum containing anti-SmZF1 antibodies was collected 15 days after the third immunization.
S. mansoni adult worms used in this study were recovered from perfused mice. Lung-stage schistosomula were prepared according to Harrop and Wilson [37]. Cercariae were obtained from Biomphalaria glabrata by exposing the infected snails to light for 2 h to induce shedding of parasites.
Sections of Omnifix (AnCon Genetics Inc., Melville, NY, USA) fixed, paraffin-embedded adult male or female worms were deparaffinized using xylol, hydrated with an ethanol series, washed in PBS and then incubated in a blocking solution (0.05% Tween 20, 1% w/v BSA (Bovine Serum Albumin) in PBS pH 7.2) overnight at 4°C. Samples were reacted for 1 h with either the anti-SmZF1 or a control, pre-immune rabbit serum, both diluted 1∶30 in 10x diluted blocking solution. Sections were then washed in PBS and reacted for 1 h with a 1∶400 diluted goat anti-rabbit IgG-Cy-5 conjugate (Jackson Immunoresearch Laboratories Inc., West Grove, PA, USA) in 10x diluted blocking solution, which also contained Alexa Fluor 488 phalloidin (Invitrogen, Carlsbad, CA, USA) diluted 1∶100 to stain actin microfilaments (except for adult male worms). Afterwards, samples were washed, incubated for 10 min with 1∶3000 diluted propidium iodide (Sigma-Aldrich, St. Louis, MO, USA) in 10x diluted blocking solution to stain nuclei and then washed with PBS.
For experiments using cercariae and lung-stage schistosomula, a whole-mount protocol was chosen. Omnifix fixed cercariae were treated with a permeabilizing solution (0.1% Triton X-100, 1% w/v BSA and 0.1% w/v sodium azide in PBS pH 7.4) for 3 h at 4°C under constant agitation. Subsequent immunostaining steps used the same solution and condition. Samples were incubated overnight with the anti-SmZF1 antibody diluted 1∶90, washed several times and reacted for 4 h with the goat anti-rabbit IgG-Cy-5 conjugate diluted 1∶1200 in solution containing Alexa Fluor 488 phalloidin (1∶500). The cercariae were then incubated for 20 min with propidium iodide diluted 1∶6000 and washed once more. The schistosomulum immunohistochemistry assays were carried out as with cercaria, with the following modifications: lung stage schistosomula were treated with permeabilizing solution overnight and then incubated with the anti-SmZF1 antibody (1∶90) for 2 h. The secondary antibody was used at a 1∶1000 dilution, and the phalloidin at a 1∶100 dilution for 2 h.
Samples (adult male and female worms, schistosomula and cercariae) were prepared with a mounting solution (90% glycerol, 10% tris-HCl 1 M, pH 8.0) and the fluorescence images were captured with a Carl Zeiss LSM 510 META confocal microscope using a 63x oil-immersion objective lens in the Center of Electron Microscopy (CEMEL-ICB/UFMG). Images were analyzed with Zeiss LSM Image Browser software and edited with Adobe Photoshop CS.
All research protocols involving mice used in the course of this study were reviewed and approved by the local Ethics Committee on Animal Care at Universidade Federal de Minas Gerais (CETEA – UFMG N° 023/05).
Adult worms recovered from perfused mice were manually separated and pooled according to their sex. Total RNA of both male and female worms was extracted using Trizol reagent (Invitrogen) and treated with DNase using Ilustra RNAspin Mini RNA Isolation Kit (GE Healthcare, Waukesha, WI, USA) according to the manufacturer's instructions. RNA was then quantified using a NanoDrop Spectrophotomer ND-1000 (Thermo Scientific, Waltham, MA, USA). cDNA was synthesized using 0.3 to 1.0 µg total RNA and Superscript III First-Strand Synthesis SuperMix for qRT-PCR (Invitrogen) according to the manufacturer's protocol. For q-PCR reactions, the primers SmZF1_real2_forw (5′–ACTTCTCTCAGAAATCCAGCCT–3′) and SmZF1_real2_rev (5′–TGGAGAGGATTATACAATCTGGTT–3′) were used at a 600 nM initial concentration. The S. mansoni glyceraldehyde 3-phosphate dehydrogenase (GAPDH) gene (primers GAPDH_forw 5′–TCGTTGAGTCTACTGGAGTCTTTACG–3′ and GAPDH_rev 5′–AATATGAGCCTGAGCTTTATCAATGG–3′) was used as an endogenous control in order to normalize relative amounts of total RNA. GAPDH primers were used at a 900 nM initial concentration. The amplicon sizes were 96 bp and 65 bp for SmZF1 and GAPDH, respectively. q-PCR reaction mixtures consisting of 2.5 µl of cDNA, 12.5 µl of Power SYBR Green PCR Master Mix (Applied Biosystems, Foster City, CA) and 5 µl of each primer in a total volume of 25 µl were added to 48-Well Optical Reaction Plates for amplification and quantification in a StepOne Real-Time PCR System (Applied Biosystems). Each q-PCR run was performed with two internal controls in order to assess both potential genomic DNA contaminations (i.e., no reverse transcriptase added in the cDNA synthesis) and purity of the reagents used (i.e., no cDNA added). Dissociation curve standard analyses were performed at the end of each assay to certify the specific amplifying of targets. For each set of primers, both male and female conditions (including negative controls) were run in three technical replicates. The experiment was repeated two times (biological replicates) and the delta-delta Ct method [38] was used in order to make a relative quantification comparing male and female transcript levels. Due to the nonparametric distribution of data, statistical analysis of delta-delta Ct values was performed using the Mann-Whitney U-test with significance set at P<0.05.
The SmZF1 cDNA was PCR amplified in a reaction mixture prepared in a 50 µL final volume containing 25 ng of template DNA, 0.2 pmol µL−1 of each primer (SmZF1-start-Xba: 5′–CAGTCTAGAACTTTAACTATGGAATT-3′ and SmZF1-stop-Apa: 5′-CAGGGGCCCCATCCGGAAAGGCTTGAGA-3′, or SmZF1-start-Sac: 5′-CAGGAGCTCACTTTAACTATGGAATT-3′ and SmZF1-stop-Hind: 5′-CAGAAGCTTCATCCGGAAAGGCTTGAGA-3′), 200 mM dNTPs and 5 U of Taq DNA polymerase (Phoneutria, Belo Horizonte, MG, Brazil) in the appropriate buffer (50 mM KCl, 10 mM Tris-HCl pH 8.4, 0.1% Triton X-100, 1.5 mM MgCl2). The fragments obtained were double-digested with XbaI and ApaI or SacI and HindIII restriction enzymes (New England Biolabs) and purified using a Wizard SV Gel and PCR Clean-up System (Promega, Madison, WI, USA) following the manufacturer's instructions. The fragments were then inserted, respectively, into the commercial vectors pCDNA4/TO/myc-His (Invitrogen) or pEYFP-c1 (Clontech, Mountain View, CA, USA), generating the constructions pCDNA4-SmZF1 and pEYFP-SmZF1, which express the recombinant proteins SmZF1-myc tag and YFP-SmZF1, respectively. In addition, the viral thymidine kinase (tk) promoter region was inserted (NheI/BglII) into the commercial vector pGL3-basic (Promega), generating the vector pGL3-tk-luc, with the luciferase (luc) reporter gene under control of the thymidine kinase promoter. Subsequently, an oligonucleotide containing four repetitions of the putative SmZF1 DNA binding site, D1-3DNA [36], was inserted (KpnI/NheI) upstream of the minimal tk promoter, producing the vector pGL3-zf-tk-luc. The oligonucleotide sequence was as follows: 5′-CAGGAAACAGCTATGACCGGCGAGGGAGTGATCGGCGAGGGAGTGATCGGCGAGGGAGTGATCGGCGAGGGAGTGTCGTGACTGGGAAAACCCTGGCG-3′ (specific binding sites D1-3DNA are indicated in bold).
Ligation products were used to transform the E. coli DH5a strain and the rescued plasmids were sequenced using 10 pmol of appropriate primers (for constructions based on pCDNA4/TO/myc-His: CMV-fow 5′-CGCAAATGGGCGGTAGGCGTG–3′ and BGH-rev 5′-TAGAAGGCACAGTCGAGG–3′, for constructions based on pEYFP-c1: YFP-fow 5′-TTTTGCTCACAGGTTCT–3′ and YFP-rev 5′-GCCGTAGGTGGCATCGCC–3′, for constructions based on pGL3-basic: GLprimer2 5′-CTTTATGTTTTTGGCGTCTTCCA-3′ and RVprimer3 5′-CTAGCAAAATAGGCTGTCCC-3′), 4 µL of DYEnamic ET Dye Terminator Kit – MegaBACE (GE Healthcare) and 300 ng of DNA. The sequencing products were analyzed in the MegaBACE 1000 DNA Sequencer (GE Healthcare).
The above plasmid constructs were used either to transfect or co-transfect COS-7 cells using Lipofectamine™ 2000 Transfection Reagent (Invitrogen), according to the manufacturer's protocol. COS-7 cells were maintained at 37°C, 5% CO2 in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 10% fetal bovine serum and 1% glutamine (Invitrogen).
The plasmids pEYFP-c1 (control) or pEYFP-SmZF1 were transfected (as above) into COS-7 cells for transient protein expression studies. Forty-eight hours after transfection the culture medium was carefully removed and cells were fixed (15 min) with 3% paraformaldehyde in PBS, washed and then quenched using PBS plus 10 mM NH4Cl (10 min). Cells were washed three times with PBS and incubated for 7 min with 0.1% Triton X-100. After another wash in PBS, COS-7 cells nuclei were stained (4 min) with 5 µL of 1 mM Hoechst 33342 dye (Sigma-Aldrich). The fluorescence was directly observed using a confocal microscope (Carl Zeiss LSM 510 META, 200x) equipped with a Photometrics Quantix CCD camera controlled by MetaMorph imaging software (MDS Analytical Technologies, Downingtown, PA, USA).
For Western blot assays, COS-7 cells (0.5×106) transfected either with pCDNA4-SmZF1 or pEYFP-SmZF1 and control cells transfected either with pEYFP or pCDNA were washed and resuspended in 200 µL of cold TNE (150 mM NaCl, 50 mM Tris-HCl pH 7.5 and 1 mM ethylenediaminetetraacetic acid (EDTA)). A 50 µL aliquot of cells was centrifuged (700 g, 4 min, 4°C) and the pellet resuspended in 50 µL of 2x SDS gel-loading buffer (100 mM Tris-HCl pH 6.8, 200 mM dithiothreitol, 4% SDS, 0.2% bromophenol blue, 20% glycerol) and boiled for 5 min, generating the total extract. The remaining 150 µL of cells was centrifuged (700 g, 4 min, 4°C) and the pellet resuspended in 40 µL of lysis buffer (10 mM Tris-HCl pH 7.5, 10 mM NaCl, 2 mM MgCl2, 1 mM phenylmethylsulphonylfluoride (PMSF), one dissolved tablet of Complete Protease Inhibitor Cocktail (Roche, Basel, Switzerland), 1 mM Na3VO4 and 1 mM NaF) plus 100 µL of 1% Nonidet P-40 (Sigma-Aldrich) in 50 mM Tris-HCl pH 7.5. Samples were incubated in an ice bath for 10 min and centrifuged (700 g, 4 min, 4°C). Ninety-five microliters of 5x SDS gel-loading buffer was added to the supernatant, which was boiled for 5 min, generating the cytoplasmic fraction. The pellet was washed twice with cold TNE, centrifuged (700 g, 4 min, 4°C), resuspended in 50 µL of 2x SDS gel-loading buffer and boiled for 15 min, generating the nuclear fraction. COS-7 total, cytoplasmic and nuclear extracts, normalized at equal volume percentage, were separated using 10% SDS-PAGE and blotted (2 h, 20 mA) onto nitrocellulose membranes (Whatman GmbH, Dassel, Germany) using a semi-dry blot system (GE healthcare). Antibody reactions were performed as described by Koritschoner and colleagues [39]. Briefly, membranes were blocked overnight in TBS (25 mM Tris-HCl pH 7.4, 137 mM NaCl, 5 mM KCl, 0.6 mM Na2HPO4, 0.7 mM CaCl2, 0.5 mM MgCl2) plus 1 mM EDTA, 1 mM Na3VO4, 0.05% Tween-20 and 3% BSA followed by two washes with 100 mM Tris-HCl pH 8.0, 200 mM NaCl, 0.2% Tween-20 (wash buffer). Samples were reacted with anti-myc, anti-GFP or anti-c-erbB-2 (1∶1000) peroxidase conjugated antibodies (BD Biosciences, Franklin Lakes, NJ, USA) in blocking buffer for 1 h. Subsequently, blots were washed and developed with ECL enhanced chemiluminescence reagents (GE Healthcare) and exposed to X-ray film. The exclusively cytoplasmic protein c-erbB-2 was used as a quality control for extracts.
For the electrophoretic mobility shift assay (EMSA), 20 pmol of the D1-3DNA oligonucleotide (5′-CGAGGGAGT-3′) was incubated with 1 µg of the total extract of COS-7 cells transfected with plasmids pEYFP-c1 (control) or pEYFP-SmZF1. Extracts were produced as follows: cells (0.5×106) were washed in PBS and resuspended in 100 µL of TDGK solution (20 mM Tris-HCl pH 7.5, 2 mM dithiothreitol, 400 mM KCl, 5 µg/ml leupeptin, 5 µg/ml aprotinin, 20% glycerol, 0.5 mM PMSF, 1 mM Na3VO4). Samples were maintained on ice for 30 min, centrifuged (15000 g, 20 min, 4°C) and then the supernatant was collected. Protein concentrations were measured and normalized as previously described [36]. The extract/DNA binding reactions were carried out in a final volume of 15 µL of binding solution (4 mM Tris-HCl pH 8.0, 40 mM NaCl, 1 mM ZnSO4, 4 mM MgCl2, 5% glycerol) for 15 min at 4°C. For supershift reactions, the DNA/extracts mixture was incubated, as above, with 1 µL of anti-GFP or 2 µL of anti-SmZF1 antibodies. After incubations, samples were fractionated in a 4% non-denaturing polyacrylamide gel in TBE buffer (89 mM Tris-borate pH 8.0, 2 mM EDTA), at a constant 25 mA at 4°C, to separate the bound complex from the free oligonucleotides. The resulting gels were stained with VISTRA Green DNA specific dye (GE Healthcare), according to the manufacturer's protocol.
Plasmid DNA co-transfections of COS-7 cells were carried out in 24-well plates (Corning Inc., Corning, NY, USA). The day before transfection, 8×104 COS-7 cells were plated in 0.5 ml of medium/well. For each well, 2 µl of LipofectamineTM 2000 Transfection Reagent were mixed with 1.2 µg of the plasmid DNA of interest and 300 ng of TK-Renilla reporter plasmid in serum-free Opti-MEM (Invitrogen) to allow the formation of DNA-LipofectamineTM 2000 Transfection Reagent complexes. The complexes were added to the respective wells and mixed by gently rocking the plate back and forth. Cells were incubated in a 5% CO2 incubator at 37°C for 48 h and then lysed with 60 µl of reporter lysis buffer (Promega). Luciferase activity (Relative Light Units – RLU) was assayed with 20 µl of lysate and 80 µl of luciferase assay reagent (Promega) in a TD20/20 luminometer (Promega) using a 10 s measurement period. Each transfection was performed in triplicate. Transfection efficiency was normalized to TK-Renilla luciferase reporter plasmid. Statistical analysis of the data was carried out with Minitab Version 1.4 using Student's t test with Welch's correction. Only p values<0.05 were considered as significant.
SmZF1 (GenBank accession AF316828) was initially identified during a screen of an adult worm S. mansoni cDNA library [35]. Although it has also been detected in cDNA libraries of other developmental stages of this parasite (i.e., egg, 3 h schistosomulum and cercaria), the biological function of the protein coded by this gene remains to be elucidated. The SmZF1 protein contains three C2H2-type zinc finger motifs and binds specific DNA oligonucleotides, as do similar nuclear proteins involved in gene transcriptional regulation [35],[36]. Therefore, to investigate whether SmFZ1 is present in the nucleus, where it could act as a transcription factor, we decided to verify its subcellular localization at diverse S. mansoni life stages.
We carried out in situ immunohistochemistry experiments using an anti-SmZF1 antibody on S. mansoni collected at various stages during its life cycle. Western blot assays using the recombinant SmZF1 protein previously separated from its MBP portion, as well as fractionated extracts form adult worms revealed that this polyclonal antibody is specific to SmZF1 (Supporting information, Figure S1). The immunohistochemistry assays showed that SmZF1 protein localizes in the cells nuclei of adult male worms (Figures 1A–D), cercariae (Figures 1K–N) and lung stage schistosomula (Figures 1P–S). Although we have performed three different experiments in which we analyzed various paraffin sections of female adult worms, the protein could not be detected in this stage using this technique (Figures 1F–I and Supporting information, Figure S2). No SmZF1 staining was observed in the negative controls (Figures 1E, J, O, T) in which only the rabbit pre-immune serum was used. These results suggest that SmZF1 is a S. mansoni protein present in the nuclei of cells from diverse developmental stages where it may act as a transcription factor. Plus, SmZF1 expression might be sex-specific since it could not be detected in adult female worms.
We were unable to confirm the results from the immunohistochemistry experiments showing differences in expression of SmZF1 between male and female by Western blot, since nuclear protein extraction from single sex pooled S. mansoni worms did not provide sufficient material necessary for SmZF1 detection. Therefore, we decided to verify gene expression by comparing the transcript levels between adult male and female worms. Total RNA extraction was performed in separate pools of male or female worms and q-PCR analyses were carried out using primers specifically designed for SmZF1 amplification. We detected no difference in SmZF1 expression (p = 0.22) between male and female worms when comparing the amplification profile, indicating that the SmZF1 mRNA is equally present in both genders (data not shown). These results suggest that although the SmZF1 gene is transcribed in female worms, a post-transcriptional regulatory mechanism could be occurring to block SmZF1 protein production in adult female worms.
After demonstrating the nuclear localization of SmZF1 in S. mansoni cells, the next step in the protein characterization was to heterologously express it in a mammalian system to test its ability to activate the transcription of a reporter gene. To accomplish this, we initially transfected COS-7 cells with the pEYFP-SmZF1 construction and forty-eight hours after transfection, we verified the presence of the YFP-SmZF1 recombinant protein mainly in the cells nuclei using fluorescence microscopy. However, a low level of fluorescent staining remained in the cytoplasm. In some cases, the protein was also visualized as fibrous material in the perinuclear region, probably associated with the cytoskeleton or Golgi complex. The YFP protein (negative control) was visualized diffusely distributed throughout the cells area (Figure 2A).
Since part of the fusion protein still remained in the cytoplasm of the cells, a second construction lacking YFP (SmZF1-myc tag) was used to confirm the SmZF1 nuclear localization in mammalian cells. Western blot assays using equal amounts of total, cytoplasmic and nuclear extracts of COS-7 cells expressing the proteins YFP, YFP-SmZF1 or SmZF1-myc tag were performed. Fractions were analyzed using either anti-GFP (which also recognizes YFP) or anti-myc antibodies (Figure 2B). The results corroborated those obtained by fluorescence microscopy (Figure 2A), showing that YFP-SmZF1 is present in both nuclear and cytoplasmic extracts, with a slight enrichment of the protein in the nuclear extract (Figure 2B). However, the recombinant protein SmZF1-myc tag is only present in the nuclear COS-7 extract, suggesting that YFP may be interfering in the transport of the fusion protein to the nucleus. The quality of the fractionation was confirmed by the localization of the cytoplasmic protein c-erbB-2 in the total and cytoplasmic fractions only (Figure 2B).
In previous experiments using purified recombinant SmZF1 protein expressed in bacteria, we demonstrated the nucleic acid binding ability and specificity of SmZF1, its preference for DNA as compared to RNA, and its putative best DNA binding sequence (D1-3DNA) [36].
To verify whether the recombinant protein YFP-SmZF1 expressed in mammalian COS-7 cells was able to interact with D1-3DNA binding site in a manner comparable to its recombinant prokaryotic counterpart, EMSA assays were performed. Total extracts of COS-7 cells transfected with either pEYFP-c1 or pEYFP-SmZF1, expressing YFP or YFP-SmZF1, respectively, were incubated with the D1-3DNA oligonucleotide. To confirm the SmZF1/D1-3DNA interaction, supershift assays using anti-GFP and anti-SmZF1 antibodies were also performed. Extracts of cells expressing the YFP-SmZF1 recombinant protein were able to shift the oligonucleotide migration in the gel (Figure 3, lane 5). Additionally, both anti-GFP and anti-SmZF1 antibodies were able to supershift D1-3DNA migration, confirming that the YFP-SmZF1 protein was responsible for the oligonucleotide binding (Figure 3, lanes 6 and 7). Extracts of cells expressing only the YFP protein (Figure 3, lanes 2–4), as well as anti-GFP and anti-SmZF1 antibodies (Figure 3, lanes 8 and 9), were not able to shift the D1-3DNA migration.
Although new vectors which will allow transfection of schistosome cells are under development [40]–[42], it is still not possible to continuously cultivate schistosome cells lineages in vitro. Accordingly, some authors describe the use of mammalian cells to study aspects of S. mansoni gene regulation processes, such as testing transcription factor activities or mapping promoter regions of genes [28],[43],[44]. Thus, a luciferase system assay in COS-7 mammalian cells expressing YFP-SmZF1 fusion protein was used here to test SmZF1ability to regulate gene transcription.COS-7 cells co-transfected with the expression vector pEYFP-SmZF1 and the construction pGL3-zf-tk-luc, which contains four repetitions of the SmZF1 D1-3DNA binding site and a thymidine kinase minimal promoter upstream of the luciferase coding gene, were able to increase gene transcription by 2-fold (p≤0.003) when compared to negative controls, using the Student's t test (Figure 4). These results suggested that SmZF1 positively affects the transcriptional activity of the minimal thymidine kinase promoter in COS-7 cells.
Schistosomiais is one of 13 neglected tropical diseases that together affect 1 billion people worldwide. The disease is considered the second most socioeconomically devastating parasitic disease, the first being malaria [45]. According to Chirac and Torreele, in the past 30 years the number of drugs which target these neglected diseases is about 1% of all the new chemical entities commercialized by the pharmaceutical industry [46].
S. mansoni presents a variety of interesting biological regulatory processes, such as transcriptional control, which can be used to allow its adaptation to the diverse biotic and abiotic environments [8]. Description of genes expressed in a stage- or sex-specific manner may help to elucidate the events used by the parasite to deal with these potentially adverse conditions. In turn, this information may also help to develop suitable vaccines and chemotherapeutic drugs against this organism [7]. As stated in the recent and high quality review on schistosome genomics by Han and colleagues [47], some potential drug targets should include proteins involved in DNA replication, transcription and repair systems. This suggestion is also corroborated by a chemogenomics screening approach described as part of the up-to-date S. mansoni genomic analysis, in which the authors used a strategy to find significant matches between parasite proteins and proteins known to be targets for drugs in humans and human pathogens. That study revealed 26 putative S. mansoni protein targets and their potential drugs. Of these 26 targets, three proteins are involved in DNA metabolism and two others are involved in chromatin modification (histone deacetylase 1 and 3) [5]. These two examples emphasize the importance of nuclear proteins as potential drug targets.
According to the authors of the S. mansoni transcriptome project [48], 2.4% of the categorized ESTs (Expressed Sequence Tags) under the Molecular Function in Gene Ontology (GO) encode transcriptional regulators. A search for conserved domains using the Pfam database in a subset of those transcripts showed that 5% of them consist of zinc fingers of the C2H2 group [48]. Moreover, most of the 15 Pfam domains found were from proteins involved in either intercellular communication or transcriptional regulation. These findings reinforce the importance of this class of regulatory proteins for S. mansoni biology. In addition, using the SAGE approach, Ojopi and colleagues found that 9.7% of the most abundant genes (genes containing more than 500 tags) from S. mansoni adult worms comprise those from the nucleic acid binding GO functional category [49].
The present study defines the SmZF1 protein as a S. mansoni transcription factor. SmZF1 is a C2H2 zinc finger protein able to specifically bind to RNA and DNA, but with higher affinity for DNA molecules. Its transcript was identified in the cercaria, egg, schistosomulum and adult worm stages, suggesting its importance as a regulatory protein [35],[36].
To define SmZF1 activity as a transcription factor, we first verified its subcellular localization, since this class of proteins is preferentially located or able to go to the cell nucleus, this import being a central step to regulate gene transcription [50],[51]. In silico analyses of the SmZF1 amino acid sequence did not predict any classical potential nuclear localization signal (NLS), but did reveal positively charged amino acids within the zinc finger motifs [35]. It has been demonstrated that zinc finger motifs are sufficient and sometimes essential for nuclear localization of ZF proteins, even without any canonical NLS detected in their amino acid sequences [51],[52]. Moreover, it is well known that small proteins (<40 Kda), like SmZF1, are sometimes able to passively diffuse into the nucleus [50].
Immunohistochemical analysis of the diverse parasite developmental stages demonstrated that SmZF1 was indeed localized in the nucleus of S. mansoni cercariae, schistosomula and adult male worms. This confirms previous results obtained by SmZF1 cDNA amplification [35] and reinforces our hypothesis that the protein is a transcription factor. An unexpected result was the lack of detection of SmZF1 protein in adult female worms when assayed by this technique. This differs from available transcriptome data, given the existence of one EST sequence (GenBank accession number BF936884) derived from an adult female worm cDNA library presenting 99% identity with SmZF1. Also, studies using oligonucleotide microarrays in which the SmZF1 sequence was spotted on the slide did not reveal this transcript as being differentially expressed between adult male and female worms [16],[20]. Based on these observations, we performed q-PCR experiments to analyze the SmZF1 mRNA expression. We were not able to detect differences in the levels of SmZF1 transcripts between adult male and female worms, indicating that the SmZF1 gene is being equally transcribed in adult female as it is in adult male worms. The fact that SmZF1 protein was not detected in adult female worms by immunofluorescence experiments suggests that a post-transcriptional mechanism regulates the gene. It is important to note that, apparently, SmZF1 mRNA levels are low in all parasite life cycle stages, as demonstrated by the number of ESTs matching SmZF1 cDNA present at dbEST (Table S1). Since the SmZF1 protein is highly abundant at the various stages, as verified by immunohistochemistry assays (except for the female adult worm), it can be hypothesized that the protein has a long half life and that the few existing mRNAs may possess a high translational rate. However, this picture might be different for female adult worms, in which the transcript could be less translated or translated in a non-efficient way. As a second hypothesis, the protein in females may present a higher turnover. Future experiments need to be done in order to clarify these points. In a recent study concerning S. japonicum, Liu and colleagues analyzed data obtained using either transcriptome or proteome approaches and found several genes with no direct correlation in their expression when comparing these two techniques [53]. The authors explained this fact by limitations in sensitivity of the proteomic technologies they employed, but also highlighted that some transcripts may be relatively stable, persisting throughout several stages and being translated in a shorter window. This could contribute to the discrepancy between the proteomic and transcriptomic data [53].
According to Hokke and colleagues, investigating proteins differentially associated with each sex could reveal important clues concerning the formation of sexually mature schistosomes and consequently leading to the description of novel chemotherapeutic targets acting in the maturation process [54]. Recently, different groups have used a myriad of approaches to describe schistosome genes expressed in a gender- or stage-enriched/specific fashion, emphasizing the importance of identification and characterization of proteins that may be controlling the transcription of these genes [8], [16]–[26]. Moreover, the sex-specific presence of a protein potentially capable of regulating the expression of a large number of other genes, as in the case of SmZF1, becomes undoubtedly important in this context. One molecule, SmLIMPETin appears to modulate gene expression in S. mansoni [55]. SmLIMPETin gene is less expressed in sexually mature adult females when compared to sexually immature adult females and sexually mature and immature adult males [55]. These observations suggest that the sex-specific expression of a transcription factor may be a common feature involved in the maintenance of this parasite life cycle.
The ability of SmZF1 to activate/repress transcription of a luciferase reporter gene in a cellular context was assessed using COS-7 cells. The first step was to confirm the expression, localization and activity of the fusion protein YFP-SmZF1, used for the assay. YFP-SmZF1 was clearly visualized in the COS-7 cells nuclei using fluorescent microscopy; however, the protein was also visualized as fibrous filaments dispersed at the perinuclear region, probably associated with the cells cytoskeleton. Furthermore, Western blot assays showed the preferential nuclear localization for YFP-SmZF1, although it was also detected to a lesser degree in the cytoplasmic extract fraction. One possible explanation for this finding is that the YFP portion of the fusion protein, considering its larger size, is interfering with the efficiency of its transport to the nucleus. Conversely, recombinant SmZF1-myc tag, a smaller protein, is detected exclusively in the nuclear portion of COS-7 cells.
The second step was to verify the protein activity, i.e., if the recombinant protein YFP-SmZF1 was able to bind to its target DNA. EMSA assays were performed using total COS-7 extracts incubated with the putative SmZF1 best binding sequence, D1-3DNA. The experiments showed that cell extracts expressing the YFP-SmZF1 recombinant protein retarded the migration of D1-3DNA in the gel. When anti-GFP or anti-SmZF1 antibodies were added to the extract-DNA samples, a supershift was observed, confirming the binding of YFP-SmZF1 to its target.
The transcriptional activity of SmZF1 was further tested using a luciferase reporter system. The results showed a 2-fold increase on the luciferase gene expression in COS-7 cells co-transfected with pGL3-zf-tk-luc and pEYFP-SmZF1. The small but significant increase in the luciferase gene expression observed might be due to the absence, in COS-7 cells, of additional proteins that are important for the proper arrangement of the transcriptional complex into the promoter region. Supporting this hypothesis, Emami and colleagues found that a species-specific interaction between TFIID and Sp1 was essential for transcriptional activation, thus suggesting a difference in transcriptional machinery between vertebrates and invertebrates [56]. As for SmZF1, if a binding partner was present, the increase in the transcriptional activation would probably be much more substantial. A similar scenario has been reported for the protein SmNR1 from S. mansoni. In a recent work, Wu and colleagues demonstrated that the SmNR1 protein alone is able to activate the transcription of a reporter gene in COS-7 cells, but when another protein already known to interact with it (SmRXR1) is present, this activation increases approximately 2-fold [28]. In order to better characterize SmZF1 action as a transcription factor, future experiments designed to detect the protein binding partners will be necessary.
In addition to the DNA/RNA specific binding ability of SmZF1 [36], the evidence of its nuclear localization, as well as its capacity to activate gene transcription, strongly suggest that SmZF1 is a S. mansoni transcriptional regulator. Additional experiments aimed at determining SmZF1 biological role are being performed. Recently our group used RNAi to conduct an in vitro phenotypic screening of 32 S. mansoni genes, including SmZF1, known to be expressed at the sporocyst stage [57]. In this study, miracidia were cultivated in vitro, transformed into sporocysts in the presence of specific dsRNAs and observed during 7 days, in order to evaluate phenotypic changes. The treatment of the S. mansoni larvae with SmZF1-dsRNA induced a reduction of 30% on the SmZF1 transcript levels, when assayed by q-PCR. This modest reduction on the transcript levels was accompanied by a shortening at the sporocyst length in two out of three independent experiments, when compared to a negative control in which a GFP-dsRNA was used. These results show that, even with a small reduction at the transcript levels the parasite phenotype was altered, demonstrating the importance of the SmZF1 gene expression for the parasite larval stage. We believe that the significance of these findings can be extended for the other life cycle stages.
|
10.1371/journal.pbio.1001988 | Protection of Armadillo/β-Catenin by Armless, a Novel Positive Regulator of Wingless Signaling | The Wingless (Wg/Wnt) signaling pathway is essential for metazoan development, where it is central to tissue growth and cellular differentiation. Deregulated Wg pathway activation underlies severe developmental abnormalities, as well as carcinogenesis. Armadillo/β-Catenin plays a key role in the Wg transduction cascade; its cytoplasmic and nuclear levels directly determine the output activity of Wg signaling and are thus tightly controlled. In all current models, once Arm is targeted for degradation by the Arm/β-Catenin destruction complex, its fate is viewed as set. We identified a novel Wg/Wnt pathway component, Armless (Als), which is required for Wg target gene expression in a cell-autonomous manner. We found by genetic and biochemical analyses that Als functions downstream of the destruction complex, at the level of the SCF/Slimb/βTRCP E3 Ub ligase. In the absence of Als, Arm levels are severely reduced. We show by biochemical and in vivo studies that Als interacts directly with Ter94, an AAA ATPase known to associate with E3 ligases and to drive protein turnover. We suggest that Als antagonizes Ter94's positive effect on E3 ligase function and propose that Als promotes Wg signaling by rescuing Arm from proteolytic degradation, spotlighting an unexpected step where the Wg pathway signal is modulated.
| The Wg/Wnt signaling pathway, found in most animals, is essential for regulating tissue growth and the formation of different cell types during development. Defects in the Wg/Wnt signaling relay can have serious consequences, ranging from aberrant organ patterning to malignant tumor formation. A pivotal step in the transmission of the Wg/Wnt signal is the stabilization of the protein Armadillo/β-Catenin, a key component of the pathway. However, the means by which the levels of this protein are regulated remain unclear. Here, we describe a novel control point of Armadillo/β-Catenin levels. Using RNA interference, we performed a screen in the fruit fly Drosophila melanogaster and identified Armless, a protein whose biological function was previously unknown, as a novel regulator of Wg/Wnt signaling, essential for the Wg/Wnt-dependent expression of downstream target genes. Our experiments suggest that Armless interferes with the tagging of Armadillo/β-Catenin with ubiquitin, thereby sparing it from proteasomal degradation. We also show that Armless directly interacts in vivo with Ter94, a ubiquitous ATPase involved in protein turnover. Our results suggest that Armless antagonizes Ter94's function in protein turnover, thereby acting as a positive regulator of Wg/Wnt signaling by promoting the stabilization of Armadillo/β-Catenin.
| The wingless (wg) gene was found nearly forty years ago with the characterization of a Drosophila mutant without wings [1]. The gene encodes a secreted glycoprotein, the founding member of the Wnt family of signaling proteins [2]. In the decades following its discovery, Wg/Wnt signaling has been shown to be essential during embryogenesis. Indeed, it is important throughout an organism's life, controlling also the homeostasis of different organs, for example, regeneration of epithelial cells in the intestine—the aberrant behavior of these cells in cancer is caused by constitutive Wg/Wnt signaling, which is consequently a key focus of medical and translational research [3],[4].
The relay of the Wg signal is controlled at different levels. However, the pivotal step is the regulation of the levels of Armadillo (Arm)/β-Catenin, the key transducer of the Wg/Wnt pathway. A multiprotein complex consisting of the scaffold proteins Axin and APC and the kinases Shaggy/GSK3β and Casein kinase I (CKI) recruits and phosphorylates Arm/β-Catenin. This marks Arm/β-Catenin for ubiquitination by the SCF/Slimb/βTRCP E3 ubiquitin ligase and subsequent degradation by the ubiquitin-proteasome system (UPS). When Wg/Wnt binds its receptors at the cell membrane, degradation of Arm/β-Catenin is prevented, presumably by protein interactions that lead to the dissociation of the E3 ubiquitin ligase from the Arm/β-Catenin destruction complex [5]. As a consequence, Arm/β-Catenin translocates into the nucleus, where it adopts its role as a transcriptional effector of Wg/Wnt signaling. Although this step is crucial, and is a potential point of regulation, little is known about the players involved in the processing of Arm/β-Catenin and its ultimate degradation.
In a genome-wide RNA interference (RNAi) screen we isolated Armless (Als) as a regulator of proximodistal growth of Drosophila limbs, and show in subsequent analyses that it exerts its function in the Wg pathway. Detailed genetic studies demonstrate that Als acts downstream of the destruction complex, at the level of the SCF/Slimb/βTRCP E3 Ub ligase. Cells depleted for Als exhibit strongly reduced Arm protein levels. Importantly, the activity of a constitutively active form of Arm, ArmS10, which cannot be phosphorylated and hence escapes ubiquitination and proteasomal degradation, is insensitive to depletion of Als. Using immunopurification and mass spectrometry analysis we found that Ter94 interacts with Als. Ter94 is an AAA ATPase associated with protein turnover and proteasomal degradation [6]. In sum, our data suggest that Als acts downstream of the Arm/β-Catenin destruction complex to positively regulate Arm protein levels, possibly by rescuing Arm from ubiquitination via Slimb. The human ortholog of Als, UBXN6, can substitute for Als in Drosophila, and Wnt target gene expression was impaired upon knock-down of UBXN6 in HEK-293 cells. We thus infer that Als and UBXN6 represent regulators of a conserved mechanism that ensures appropriate levels of Armadillo/β-Catenin by antagonizing its entry into the UPS.
Among its numerous developmental roles, Wg controls growth and patterning of the Drosophila wing. A hallmark phenotype of reduced wg function is nicked wings with reduced size. Positively acting components of the Wg pathway mostly display similar phenotypes when depleted via RNAi during larval stages. We established and used a tester system that allowed screening for genes involved in wing growth and margin formation. Our screen covered approximately 83% of all the predicted 14,306 genes of Drosophila. The details of the screen will be published elsewhere. CG5469 behaved as expected of a putative Wg signaling component: RNAi targeting its transcripts affected wing size in a non-allometric manner and caused a notched margin and loss of sensory bristles. This phenotype could be reproduced by the expression of nine independent UAS-CG5469RNAi lines with different Gal4 drivers (Figures 1B–1E and S1A–S1J). Expression of CG5469RNAi effectively reduced mRNA levels of CG5469 (Figure S1L, cf. Figure 1B–1E), and haplo-deficiency exacerbated the wing phenotypes (Figure S1M and S1N); these phenotypes were rescued upon overexpression of CG5469, as detailed in Text S1 and shown in Figure S2.
Wing notches could be caused by increased apoptosis or reduced cell proliferation. To discriminate between these possibilities, we experimentally accelerated cell divisions by string/cdc25 overexpression [7], and inhibited apoptosis by Diap1 overexpression. Only string/cdc25 overexpression restored marginal wing tissue loss caused by depletion of CG5469 (Figure S3F). This finding is consistent with a role of CG5469 in Wg signaling, as impaired Wg pathway activity affects primarily proliferation (Figure S3A–S3C; [8],[9]). Because of its mutant phenotypes, which are caused by severely reduced levels of Arm (see below), we refer to CG5469 as armless (als; for the nomenclature see also Text S1).
To further test whether als interacts with Wg signaling, we analyzed the effects of alsRNAi in different, sensitized backgrounds. Interfering with the Wg pathway at various levels leads to typical wing margin notches and loss of distal wing tissue (Figure S3G–S3J). These phenotypes were enhanced upon als depletion (Figure S3G′–S3J′). To test whether als is also involved in the Wg pathway in other developmental contexts, we expressed alsRNAi in the primordia of the dorsal thorax and the legs. Again, consistent with a role of als in the Wg pathway (Figure S4D and S4H; [10],[11]), thorax and scutellum sizes were affected (Figure S4B, S4C, S4F, and S4G, cf. Figure S4A), and showed reduced or missing microchaete, a slight misorientation of remnant micro- and macrochaete, and mild thoracic clefts (Figure S4B and S4C). Depletion of als in the leg discs caused a reduction of tarsal segments along the proximodistal axis and a dorsolateral shift of the sex combs—phenotypes characteristic of reduced Wg signaling (Figure S4J–S4M; [12],[13]). Stronger alsRNAi expression caused a dorsalization of the leg as judged from the mirror-image patterning of leg trichomes (Figure S4K and S4M). To analyze the phenotype at earlier stages of development, we expressed alsRNAi or AlsHA (a dominant-negative-like version; see Text S1) in the embryo with da-Gal4. We found that cuticles were shorter than in control embryos, with ventral denticle belts partially fused, which is indicative of a segment polarity defect (Figure S4S–S4V, S4X, and S4Y). These phenotypes are reminiscent of milder Wg loss-of-function phenotypes caused by armRNAi or UAS-Lgs17E expression (Figure S4P–S4R and S4W) by da-Gal4.
To further test whether als is required for Wg signaling output, we monitored the expression of well-known Wg target genes. Wg is secreted from a stripe of cells at the dorsoventral (D/V) border and triggers a transcriptional response in cells along the D/V axis of imaginal wing discs (the future proximodistal axis of the adult wing). Wg activates the proneural gene senseless (sens) in sensory organ precursors, which develop into the bristles of the adult wing margin [8],[14]. alsRNAi expression in the P-compartment resulted in a loss of Sens therein (Figure 2A and 2A′). Expression of Distalless (Dll), frizzled3 (fz3), and wingful (wf), which are activated by Wg in a broader region centered on the D/V border [15],[16], was strongly reduced or lost in response to reduction of als function (Figure 2B–2E). Conversely, arrow (arr), which is normally repressed by Wg signaling in central pouch regions (Figure 2F; [17]), became ectopically activated towards the central wing pouch (Figure 2G). In sum, these observations strongly suggest that als encodes a component with a positive role in the Wg pathway.
To try to gain insight into the pathways als might regulate, we further looked at als expression. In situ hybridization and lacZ reporter transgenes indicated that als is expressed at early third instar wing imaginal discs at the center of the pouch adjacent to the D/V border and, at faint levels, throughout the wing primordium (Figure S5A–S5C). Furthermore, the expression of UAS-alsRNAi in the P-compartment of wing imaginal discs caused a strong decrease in als expression (Figure S5B, inset). Interestingly, although normal when Wg signaling is inhibited, als expression could be upregulated when Wg signaling was activated by the overexpression of either Wg or ArmS10, and stimulation of Wg signaling by the inhibition of GSK3β caused elevated Als and als levels (Figures S5D and 8A). This is consistent with als functioning in the Wg pathway and having its activity fine-tuned by feedback mechanisms that ensure maximum pathway activity.
Wing notching is a phenotype characteristic not only of impaired Wg signaling, but also of reduced Notch signaling. The activities of both pathways are required to establish the D/V border and subsequent proximodistal wing growth. To determine whether als also plays a role in the Notch pathway, we analyzed the expression of the following Notch targets: (1) the proneural gene e(spl)m8 (Figure S6A; [18]), (2) a Notch responsive element (NRE) reporter (Figure S6A inset; [19]), and (3) the vestigial (vg) “boundary enhancer” (Figure S6B; [20]). In contrast to what we observed for the Wg targets above, the expression patterns of these Notch reporters were unaltered upon als depletion (Figure S6A′–S6B″). Moreover, neither did reduction of als function suppress the phenotypes caused by elevated Notch signaling (Figure S6D and S6F, cf. Figure S6C and S6E), nor did it ever mimic a neurogenic thorax phenotype or fused tarsal segments (Figure S4), which are both typical for reduced Notch signaling. Finally, we monitored the expression of the wg gene itself, which is under control of Notch activity [21]–[23]. Neither wg transcription nor Wg protein expression were affected at the D/V boundary upon alsRNAi expression (Figure S7A–S7G).
We also tested other pathways involved in wing growth and patterning, such as Hedgehog (Hh) signaling. Increased Hh signaling causes wing overgrowth predominantly along the anteroposterior (A/P) axis, along with ectopic veins in the A-compartment (Figure S6G). als depletion impaired wing growth along the proximodistal axis, which is under control of Wg signaling, but not along the A/P axis, and it did not alter the vein patterning defects caused by increased Hh signaling (Figure S6H). Furthermore, expression of the Hh target gene patched (ptc) was not affected by als depletion (Figures S6I, S6J, and 8E). In addition, we tested Upd/Jak/Stat signaling and EGFR/Ras signaling and found that none of these pathways was affected by als depletion (Figure S6K–S6N). Our findings therefore suggest a specific role for als in Wg signaling in the tissues that we have tested.
To investigate whether Als functions in Wg producing or receiving cells, UAS-alsRNAi was expressed in clones of wing imaginal cells marked by GFP co-expression. Expression of the Wg target Sens was lost in early induced clones (Figure 3A–3A″). This effect is cell-autonomous, i.e., it also occurs when mutant cells are close to wild-type cells that produce Wg. Adult wing clones, marked by UAS-forkedRNAi co-expression, had wing notches and thickened veins (Figure 3B–3E), both of which were tightly associated with mutant cells. We observed similar phenotypes when Wg signaling was cell-autonomously impaired (Figure 3F–3I).
To map where Als acts in the Wg pathway, we used phenotypic assays based on the activation or inactivation of Wg signaling in the eye or the wing. Expression of Wg under control of the sevenless (sev) enhancer causes small, narrow eyes that lack interommatidial bristles (Figure 4B; [24],[25]). This phenotype is partially suppressed when Wg signal transduction is hindered, for example, by the expression of a dominant-negative form of Lgs (Lgs17E [26]; Figure 4G). Impaired als function also causes a suppression of the sev-wg phenotype (Figure 4C–4F). In contrast, depletion of als did not alter the small eye phenotype caused by the overexpression of Eiger (Figure 4H–4J), a trigger of the apoptotic pathway [27].
We manipulated the Wg signaling cascade in the wing and the eye at successively more downstream positions and assessed the requirement of Als for the manifestation of the respective Wg gain-of-signaling phenotypes (see legend of Figure 5 for a detailed description of phenotypes). First, we found that Als depletion suppressed Wg pathway activation caused by RNAi of notum/wingful (encoding an extracellular inhibitor of Wg [28],[29]) or by overexpression of Arrow (co-receptor of Wg, [17]) or Dishevelled (Figure 5A–5B′, 5L, and 5L′). Next, we tested knock-downs of components of the destruction complex, both in the wing and the eye, and found that aberrant Wg output was suppressed by alsRNAi for Axin, APC, and Shaggy/GSK3β (Figure 5C, 5C′, 5D, 5D′, 5H–5K′, 5M, and 5M′). Only the phenotypes of RNAi against the SCF/Slimb/βTRCP E3 ubiquitin ligase [30] and expression of ArmS10, a constitutively active form of Arm/β-Catenin [31], could not be ameliorated by als depletion (Figure 5E–5F′ and 5O–5Q′). Interestingly, wild-type Arm could not overcome the alsRNAi phenotypes in the wing and the eye, and did not lead to the emergence of ectopic sensory bristles (Figure 5G, 5G′, 5R, and 5R′). We interpret these genetic findings to indicate that Als functions upstream of Arm, but downstream of the Arm/β-Catenin destruction complex, possibly at the level of the SCF/Slimb/βTRCP E3 Ub ligase.
Three different protein domains can be distinguished within Als: (1) an N-terminal coiled-coil domain, (2) a PUG domain, and (3) a UBX domain (Figure S9). Coiled-coil domains function in protein oligomerization, whereas the PUG und UBX domains are found in ubiquitin regulatory proteins [32]. To find out more about its molecular role, we tested where Als localizes within cells and which of its protein domains are functionally required. N-terminally tagged HAAls localized close to the cell membrane in a punctate ring pattern (Figure S8A–S8B′), with strongest expression in the apical region of the cell (Figure S8C, S8D, S8E). This localization is reminiscent of that of E-Cadherin, Arm, and the Wg receptor Frizzled2 (Fz2, Figure S8C–S8E″). We then expressed Als protein variants that lacked particular domains and assessed whether these could rescue the alsRNAi phenotypes. For this we created tagged and untagged constructs that were either sensitive or insensitive to the RNAi used. Expression of AlsΔN and AlsΔPUG did not alter the alsRNAi phenotype (Figure S9A–S9I), irrespective of their tag or targetability. However, expression of AlsΔUBX as well as full-length Als protein substantially rescued the wing phenotypes caused by alsRNAi (Figure S9J–S9L). This suggests that the UBX domain is dispensable for Als function, whereas the N-terminal and the PUG domains are both important. The requirement of the PUG domain implicates Als in playing a role in a ubiquitin-mediated protein degradation process.
To further our understanding of Als function, we expressed HA-tagged variants of Als in Drosophila Kc-167 cells, and after affinity purification we subjected the samples to shotgun liquid chromatography–tandem mass spectrometry (LC-MS/MS). In three independent experiments using AlsHA, HAAls, and HAAlsΔUBX, but not in the negative control sample, we identified Ter94 as the highest ranking protein interactor of Als (see Text S1), irrespective of the stimulation of the Wg pathway by Wg. Ter94 is the Drosophila ortholog of p97, an AAA ATPase found to be associated with ubiquitin-dependent protein degradation processes, and that can interact with different UBX domain proteins [33],[34]. To further validate the interaction between Als and Ter94 we co-expressed Ter94HA together with FLAGAls, FLAGAlsΔN, and FLAGGal4, which served as a negative control, in Kc-167 cells. Western blot analyses following reciprocal immunoprecipitations showed that (1) Ter94HA and endogenous Ter94 co-immunoprecipitated with FLAGAls and FLAGAlsΔN, and (2) FLAGAls, FLAGAlsΔN, and endogenous Als co-immunoprecipitated with Ter94HA (Figure 6A). When we expressed Ter94HA in wing imaginal discs, we found a strikingly similar sub-cellular distribution and co-localization with FLAGAls (Figure 6B–6B″). Importantly, we demonstrated a physical interaction between Als and Ter94 in vivo with bimolecular fluorescence complementation (BiFC) analysis (Figure 6C and 6C′; [35]; see also Materials and Methods), where the co-overexpression of Als-VC and Ter94-VN resulted in a Venus-YFP fluorescent signal. In sum, our biochemical and genetic results indicate that Als molecularly interacts with Ter94. The UBX domain is dispensable for Als to bind to Ter94, which is in agreement with our finding that Als lacking the UBX domain could still rescue the alsRNAi phenotype.
Our biochemical and genetic studies suggest that Als is functionally linked to the proteasomal degradation pathway. Since Als is positively required for Wg signaling, it could exert its function in the pathway either by suppressing the degradation of a positive component, or by enhancing the degradation of a negative component. We first analyzed the protein levels of key negative pathway components in wing imaginal discs. We expressed different alsRNAi lines in the wing primordium with nub-Gal4; however, neither Axin nor Shaggy nor Apc2 levels were affected upon als depletion (Figure S10A–S10L); [36]). In contrast, we found that the protein levels of Arm, the key positive signaling component, were dependent on als: when alsRNAi was expressed in the P-compartment under the control of hh-Gal4, Arm levels were strongly reduced in P but not A cells (Figure 7A–7A″). Even when we overexpressed Arm under the control of nub-Gal4, co-expression of alsRNAi led to strongly reduced Arm levels (Figure 7C and 7C′, cf. Figure 7B and 7B′). DAPI staining shows that all cells have normally shaped nuclei and are thus not apoptotic (Figure 7A′, 7A″, 7C, and 7C′). Constitutively active, non-degradable ArmS10 localizes like Arm to the apical membrane upon overexpression (Figure 7D and 7D′), and was also observed at higher levels in the cytoplasm (Figure 7F and 7F′). Importantly, we found that ArmS10 levels were not altered in als-depleted wing imaginal disc cells (Figure 7E, 7E′, 7G, and 7G′). This is consistent with our finding that ArmS10 overexpression could completely overcome the phenotypes caused by als depletion (Figure 5F, 5F′, 5Q, and 5Q′). Other positively acting components of the Wg pathway, Arrow and Fz2, did not exhibit alterations in protein levels upon alsRNAi expression (Figure S10M–S10P). The observation that Arm is subject to degradation in the absence of als function is in agreement with our observation that overexpressed Arm—in contrast to ArmS10—is not able to rescue alsRNAi phenotypes in wing and eye primordia (Figure 5G′ and 5R′).
The rescue experiments with Arm and ArmS10 suggested that Als acts upstream of Arm's proteasomal degradation. To further corroborate this idea, we reduced the levels of ubiquitin in addition to als depletion. This led to a suppression of the alsRNAi phenotypes in the wing and the eye (Figure S11I, S11L, and S11O, cf. Figure S11G, S11J, and S11M). Reduction of ubiquitin also suppressed Wg loss-of-function phenotypes that were based on elevated activity of the destruction complex by Axin overexpression (Figure S11A and S11C). In contrast, depletion of ubiquitin did not suppress the Lgs17E overexpression phenotype, which intersects the Wg pathway downstream of the proteasome (Figure S11D and S11F). These findings support the idea that Als acts upstream of Arm's proteasomal degradation.
To check whether Als directly interacts with Arm, we carried out selected reaction monitoring, a highly sensitive mass spectrometry approach that allows the detection of low amounts of peptides in a complex sample [37]. Als and Ter94 were used as bait. In none of the studies was an interaction between Arm and Als detected. Additionally, in shotgun affinity purification mass spectrometry using Arm or ArmS10 as a bait we did not find Als, whereas, as expected, components of the degradation complex such as Apc2, Axin, Sgg, and CKI were detected in the cytoplasm (Text S1).
We further analyzed the consequences of als depletion for Wg signaling in Kc-167 cells (Figure 8). Stimulation of the Wg pathway by the inhibition of GSK3β led to strong transcriptional upregulation of the Wg targets naked (nkd) and wingful (wf) (Figure 8B and 8C, bar 2), and an increase in Arm levels (Figure 8D, lanes 3 and 11). Consistent with the in vivo data, double-stranded RNA (dsRNA)–mediated knockdown of als (Figure 8A) caused a reduction in nkd and wf expression in stimulated cells (Figure 8B and 8C, bar 6, cf. bar 2). Importantly, the increase in Arm levels based on GSK3β inhibition was significantly reduced when als was knocked down (Figure 8D, lane 4, cf. lane 3, and lane 13 and 14, cf. lane 11). When the pathway was stimulated by inhibiting E1 Ub ligase activity, which caused a milder upregulation of nkd and wf than the GSK3β inhibition (Figure 8B and 8C, bar 3; see also Materials and Methods), transcriptional upregulation was still decreased upon als depletion (Figure 8B and 8C, bar 7), but to a far lesser extent; at the same time, Arm was reduced upon als depletion (Figure 8D, lane 6, cf. lane 5). Inhibition of the 26S proteasome increased Arm levels in both control and als-dsRNA conditions (Figure 8D, lanes 9 and 10).
When we depleted ter94 in wing imaginal discs and in Kc-167 cells, we found an upregulation of the positive Wg target genes nkd and wf, and a downregulation of the negative Wg target arr (Figure 8E). Moreover, in als-depleted cells, ter94 reduction led to an amelioration of the changes in Wg target expression (Figure 8E), and to a partial restoration of Arm levels (Figure 8D, lane 12). In contrast, the Hh target gene ptc was not altered (Figure 8E). Overexpression of ter94RNAi in wing imaginal discs, however, caused pupal lethality. Because of this, it was difficult to analyze the adult wing phenotype caused by the combined depletion of ter94 and als; because of ter94's pleiotropic role, the analysis of the consequences of ter94 depletion is challenging [38].
In addition, we found that overexpression of Ter94HA suppressed the strong dominant-negative-like effect of AlsHA overexpression in the wing and the eye primordium (Figure 8F–8I), while overexpression of Ter94HA could not suppress phenotypes caused by alsRNAi (Figure S2M). This is consistent with the finding that overexpression of Ter94HA alone did not cause any phenotype (Figure S2N); however, it suggests that an excess of Ter94HA can neutralize the excess of the dominant-negative AlsHA.
Together, these results suggest that Als functions downstream of GSK3β and downstream or at the level of E1 Ub ligase and that Als is required to antagonize Ter94 to allow normal Wg signaling. This is in agreement with our genetic epistasis experiments (Figure 5) and is corroborated by the further finding that overexpression of CSN6, an essential de-ubiquitinating enzyme and negative regulator of SCF/Slimb/βTRCP E3 Ub ligase [39], reverted the alsRNAi phenotypes in the wing and the eye (Figure S11H–S11N). Of note, CSN6 overexpression suppressed the Wg signaling defects caused by Axin overexpression (Figure S11A and S11B), but did not suppress Wg signaling defects caused by Lgs17E overexpression (Figure S11D and S11E).
To address the evolutionary conservation of Als's function, we tested whether the human ortholog of Als, UBXN6, can substitute for Als in wing and eye development. Overexpression of HAUBXN6 in imaginal discs could largely rescue the wing phenotypes and fully rescue the eye phenotypes caused by alsRNAi (Figure 9A–9D). Further, we found that HAUBXN6 localizes in cells of imaginal discs in a manner similar to HAAls (Figure 9E and 9E′, cf. Figure S8A–S8E). The UBXN6 and Als proteins share 33% identity and belong to the same subfamily of UBX domain proteins, which is specified by the presence of an UBX and a PUG domain (Figure 9F). We next asked whether UBXN6 is required for Wnt signaling in human cells. We stimulated the Wnt pathway in HEK-293 cells with mouse Wnt3a (mWnt3a) while targeting UBXN6 by small interfering RNA (siRNA). siRNA against CTNNB1/β-Catenin served as a control. To monitor the output of the Wnt pathway, we analyzed three established Wnt-responsive genes, SP5, AXIN2, and FZD1, by real-time PCR. All three targets were upregulated upon pathway stimulation (Figure 9G–9G″, green bars), and this response was abolished upon siRNA treatment of UBXN6 (Figure 9G–9G″, red bars). The depletion of β-Catenin caused a similar downregulation of Wnt target gene expression. Further, we found β-Catenin levels reduced upon UBXN6 depletion in HEK-293 cells (Figure 9H, lane 6, cf. lane 1). Similar to mWnt3a, the stimulation of the pathway by GSK3β inhibition caused an increase of Wnt target genes. Depletion of UBXN6 reduced this effect (Figure 9I). These findings suggest that UBXN6 is a functional ortholog of Als and that UBXN6 might play a similar role for Wnt signaling in human cells.
A prevalent mechanism for controlling information flow in signaling pathways is the alteration of the protein levels of key components. In the Wg/Wnt pathway, the Arm/β-Catenin destruction complex targets Arm/β-Catenin for ubiquitination by the SCF/Slimb/βTRCP E3 Ub ligase, resulting in proteasomal degradation and low cytoplasmic levels of Arm/β-Catenin in the Wnt pathway off state. If the pathway is turned on, Slimb-mediated ubiquitination is prevented, thus rescuing Arm from its proteasomal fate and causing a concomitant increase in Arm protein levels [5]. Here we describe Als as a new component of this control system; we found that Als is required to prevent the degradation of Arm/β-Catenin.
We identified als in a genome-wide in vivo RNAi screen in Drosophila. Because we did not succeed in isolating an EMS- or P-element-induced null allele and because another gene overlaps with als, we demanded, and obtained, particularly thorough evidence validating als gene function. (1) The alsRNAi phenotypes could be reproduced by nine different UAS-RNAi transgenes encoding independent RNA target sites. Together with an extended off target analysis, we could rule out unintentional RNAi as a cause for the als phenotypes. (2) RNAi-mediated inhibition of als expression was ascertained by monitoring als mRNA expression via real-time PCR and antisense mRNA in situ hybridization. (3) Expression of HAAls with different RNAi-insensitive rescue transgenes, as well as with its human ortholog UBXN6, rescued alsRNAi phenotypes.
Our analyses show that als encodes an essential positive Wg signaling component. This conclusion is based on the following evidence. als depletion caused wings with notched wing margins and loss of sensory bristles, which is characteristic of impaired Wg signaling. The distal wing region is most sensitive to als levels, as is the case for other positive components of Wg signaling. In agreement with this, we found increased als expression in the central wing pouch, at least in earlier L3 larval stages. Stimulation of the Wg pathway in wing imaginal discs or Kc-167 cells caused higher als expression, suggesting that als can be positively controlled by Wg signaling. However, Als levels must be precisely controlled since already mild overexpression of UAS-als elicits a dominant-negative effect on Wg signaling. The function of als for Wg signaling is not restricted to the wing: also in other tissues, such as the thorax, eyes, legs, and the embryo, alsRNAi phenotypes are identical to those seen when Wg signaling is disturbed. Also in human HEK-293 cells we found UBXN6/UBXD1, the ortholog of Als, to be required for Wnt signaling, and human HAUBXN6 largely rescued the alsRNAi phenotypes in Drosophila, which suggests their functional conservation. Depletion of als also enhanced Wg-sensitized phenotypes, further supporting the notion that its product is a Wg pathway component. Moreover, the expression of positively regulated Wg target genes is reduced or abolished upon loss of als function, while Wg-repressed target gene expression is ectopically activated. Importantly, while interfering with als function suppressed Wg signaling, it did not affect other pathways, such as Notch and Hh, Jak/Stat, or EGFR signaling. However, we cannot rule out that als is not required in another pathway in a different biological context. In humans, UBXN6 is reported to play a role in diverse scenarios: for example, it was shown to play a role in Caveolin turnover in human osteosarcoma U2OS cells [40]. This might indicate a broader role of UBXN6 in mammalians.
Our data show that Als regulates Armadillo protein levels. Based on our epistasis experiments, Als acts downstream of Shaggy/GSK3β and upstream of the SCF/Slimb/βTRCP E3 Ub ligase, which is known to ubiquitinate Arm, a prerequisite for proteasomal degradation. Consistent with this, the degradation-resistant form of Arm, ArmS10, could completely bypass the requirement for als, in contrast to the wild-type form of Arm. This suggests that proteasomal degradation acts downstream of als; however, we would like to point out that this cannot be taken as an unambiguous proof. Importantly, depletion of ubiquitin and overexpression of CSN6, a negative regulator of SCF/Slimb/βTRCP E3 Ub ligase, could ameliorate the als phenotype (as well as phenotypes based on the overexpression of Axin or Shaggy, which overactivate the destruction complex, thus resulting in enhanced Arm degradation). In contrast, altering these factors did not ameliorate the Lgs17E phenotype, which is caused by interfering more downstream in the Wg pathway. These findings suggest that als works upstream of proteasomal degradation. A further informative experiment was monitoring Wg pathway components with respect to protein levels: Arr, Fz, Axin, APC, Sgg, and Arm. The only change in the absence of Als function was Arm: its levels were strongly reduced upon als depletion. The effects on Arm levels could be due either to a direct effect on Arm or to an indirect effect on a negative component. Importantly, the rate-limiting factor Axin as well as other key negative components of the Arm/β-Catenin destruction complex were unaltered upon als depletion.
Some further mechanistic insight was obtained with our finding that Ter94 interacts in vitro and in vivo with Als. Interestingly, we found that Als-Ter94 localizes at the cell cortex, as was similarly observed for the Arm/β-Catenin destruction complex [41]. Our studies are consistent with earlier work that showed that the human ortholog of Ter94, p97, interacts with UBXN6 [42]. Ter94/p97/Cdc48 is a conserved and highly abundant AAA ATPase that was found to associate with SCF/Slimb/βTRCP E3 Ub ligases or proteasomal shuttle factors to mediate UPS-mediated protein degradation ([33],[42]–[45]; reviewed in [6],[34]). Specifications of the diverse activities of Ter94/p97 and the fate of its substrates are mainly exerted by UBX domain protein co-factors, which eventually either promote or hinder p97's function in protein turnover; an example of the latter involves the dissociation of the SCF/Slimb/βTRCP E3 Ub ligase complex, eventually leading to its inactivation [45]. Interestingly, it was recently reported that inactivation of the E3 ligase complex upon Wnt signaling is achieved by its dissociation from the destruction complex [5]. Based on our experiments and what is known about Ter94/p97, we suggest as a possible mechanism that Als antagonizes Ter94's positive effect on E3 ligase function, thereby rescuing Arm levels. We, however, did not observe increased protein levels of Slimb, Axin, Shaggy, or APC in our analyses; thus, our results favor a model in which Als antagonizes Ter94 to hinder the transfer of Arm to the proteasome by interfering with the SCF/Slimb/βTRCP E3 Ub ligase function or its assembly. Importantly, we did not find an interaction between Arm and Als. This is consistent with the finding that UBX domain family members lacking an UBA domain, such as UBXN6/Als, do not directly interact with substrate proteins [44], but are necessary for the activity or fate of the Ter94/p97.
Interestingly, Zhang et al. [38] found that ter94 depletion affected the partial proteolysis of Ci. However, they observed neither any typical consequence of disturbed Hh signaling per se (i.e., no alteration of Hh target gene expression in genes such as ptc) nor any phenotypical consequence upon overexpression of a dominant negative form of Ter94 (i.e., aberrant wing patterning and growth typical for Hh signaling). This is consistent with our data that neither Ci target expression nor Hh signaling was affected upon als or ter94 depletion (Figures S6G–S6J and 8E).
p97/Ter94 is known as a highly pleiotropic AAA ATPase associated with many cellular functions. Further, p97/Ter94 acts in multifaceted and large protein–protein complexes, and it is its regulatory co-factors, including UBX domain proteins, that render p97/Ter94 specific for a certain task in a particular cellular context. For example, p47/Shp1 is a co-factor of p97/Ter94 that blocks other co-factors from Ter94 binding [46],[47]. Interestingly, in our Kc-167 cell mass spectroscopy experiments, we found p47 in Ter94/Als protein complexes, but only in the absence of Wg stimulation (Text S1). On the other hand, we found als transcript and Als protein levels elevated upon Wg signaling (Figures 8A and S5D). These findings suggest a dynamic regulation of the Ter94 complex upon signaling inputs. The identification and functional analysis of all key components of the Als-Ter94 complex will be needed to obtain a refined insight into Als-Ter94's molecular mechanism.
Critically, our work spotlights an underappreciated facet in the control of the output of the entire canonical Wg/Wnt pathway—how Arm/β-Catenin is handed over to the proteasome— and the potential for regulating this step; our works also indicates that this step, in contrast to the conventional wisdom, is tunable. Our identification and characterization of the UBX protein Als as a positive regulator of Wg/Wnt signaling contributes to this layer of pathway control.
Ter94-VNm9 (Ter94 fused to the N-terminal part of Venus-YFP) and Als-VC155 (Als fused to the C-terminal part of Venus-YFP) were co-expressed and analyzed for interaction by BiFC [34]. Direct interaction of two proteins causes a reconstitution of the two non-fluorescent YFP subfragments into a functional YFP, resulting in a yellow fluorescent signal.
Images of immunostainings were generated with a Zeiss Lsm710 confocal microscope using 40× and 60× oil objectives. Images were analyzed with ImageJ software. Adult fly wings were mounted in Euparal, and images were generated with a Zeiss Axioplan microscope.
RNA probes were generated by run-off transcription of antisense mRNA with T7 RNA polymerase (Promega) from linearized plasmid templates with the PCR-amplified als coding region (full-length template as well as an EcoRV-truncated template, generating the first half of the coding sequence). Antisense mRNA was generated with a 3′ primer containing the T7 RNA polymerase binding sequence (5′-GAATTTAATACGACTCACTATAGG-3′).
UAS-alsRNAi was expressed under the control of c765-Gal4 (Figure S1L) in wing imaginal discs, which were collected at the late third instar larval stage for total RNA isolation (RNeasy Kit, Qiagen, or TRI-Reagent, Sigma-Aldrich). 1 µg of total RNA was used as a template for reverse cDNA transcription (Superscript-RT-II, Roche). Quantitative PCR reactions were carried out with ¼ of the real-time PCR reaction and als-specific primers, 25 cycles, in the presence of SYBR Green (Roche), which enabled quantitative detection of the amplicons. Primers were designed by Roche Universal ProbeLibrary and are intron-spanning (als primer pair a: forward: ggaaaagacactatatgactgcaaact; reverse: aaggagcgggtgtatcattg; als primer pair b: forward: gaaaccatgtccaagattaagaagt; reverse: atgccgctgccagttaaa). Real-time PCR was carried out on an Eppendorf Realplex Mastercycler. mRNA levels were calculated with the comparative CT (threshold concentration) method and were normalized for three external standards (actin5C, TBP, and gpdh). For each RNAi condition, data were obtained from three biological replicates (i.e., three independent but identical Gal4×UAS-RNAi fly crosses), and real-time PCR was carried out in technical triplicates per experiment.
Kc-167 cells were cultured in Schneider's Drosophila medium (Invitrogen) supplemented with 10% fetal calf serum and 1% penicillin/streptomycin at 25°C. For each experimental condition, 2× T75 (8 ml) culture flasks of 90% confluent Kc-167 cells were seeded at 60% confluency. 1 d after seeding, cells were transfected with the UAS transgenes (UAS-alsHA, UAS-ter94HA, UAS- FLAGals, UAS- FLAGΔN-als). UAS-GFP served as a control for transfection efficiency; empty pUAST-attB vector and UAS-FLAGGal4 vector served as negative controls; UAS transgenes were expressed under the control of tubulin-Gal4. DNA input per T75 flask was 0.5 µg of tubulin-Gal4, 0.75 µg of UAS-GFP, 0.5 µg of control vector, and 1.5 µg of the respective UAS transgene. Effectene (Qiagen) was used for transfection, according to the manufacturer's protocol. For stimulation of the Wg pathway, the supernatant of cells containing secreted Wg was added to the cell culture 24 h after transfection.
HEK-293 cells were seeded into a 96-well plate (40,000 cells/well). 1 d after seeding cells were transfected (Lipofectamine 2000, Invitrogen) with siRNAs targeting human β-Catenin, human UBXN6, or a negative control siRNA (Qiagen FlexiTube siRNA). The siRNA concentration per well was 0.05 µM. Co-transfected histone2B-RFP served as the control of the transfection efficiency. The Wnt pathway was stimulated by mWnt3a from co-transfected pcDNA3::mWnt3a. Each experimental condition was made in triplicate. 48 h after transfection, cells were harvested, and an aliquot was used for real-time PCR (1) to analyze the siRNA-mediated knockdown of β-Catenin or UBXN6 and (2) to analyze the transcriptional level of the Wnt targets SP5, AXIN2, and FRIZZLED-1. To analyze β-Catenin levels, siRNA (Qiagen FlexiTube siRNA, Thermo Scientific SMART pool) was applied two times (the second siRNA application was done 1.5 d after the first siRNA application), and cells were harvested 86 h after the first transfection.
|
10.1371/journal.pmed.1002200 | IL-7 Receptor Mutations and Steroid Resistance in Pediatric T cell Acute Lymphoblastic Leukemia: A Genome Sequencing Study | Pediatric acute lymphoblastic leukemia (ALL) is the most common childhood cancer and the leading cause of cancer-related mortality in children. T cell ALL (T-ALL) represents about 15% of pediatric ALL cases and is considered a high-risk disease. T-ALL is often associated with resistance to treatment, including steroids, which are currently the cornerstone for treating ALL; moreover, initial steroid response strongly predicts survival and cure. However, the cellular mechanisms underlying steroid resistance in T-ALL patients are poorly understood. In this study, we combined various genomic datasets in order to identify candidate genetic mechanisms underlying steroid resistance in children undergoing T-ALL treatment.
We performed whole genome sequencing on paired pre-treatment (diagnostic) and post-treatment (remission) samples from 13 patients, and targeted exome sequencing of pre-treatment samples from 69 additional T-ALL patients. We then integrated mutation data with copy number data for 151 mutated genes, and this integrated dataset was tested for associations of mutations with clinical outcomes and in vitro drug response. Our analysis revealed that mutations in JAK1 and KRAS, two genes encoding components of the interleukin 7 receptor (IL7R) signaling pathway, were associated with steroid resistance and poor outcome. We then sequenced JAK1, KRAS, and other genes in this pathway, including IL7R, JAK3, NF1, NRAS, and AKT, in these 69 T-ALL patients and a further 77 T-ALL patients. We identified mutations in 32% (47/146) of patients, the majority of whom had a specific T-ALL subtype (early thymic progenitor ALL or TLX). Based on the outcomes of these patients and their prednisolone responsiveness measured in vitro, we then confirmed that these mutations were associated with both steroid resistance and poor outcome.
To explore how these mutations in IL7R signaling pathway genes cause steroid resistance and subsequent poor outcome, we expressed wild-type and mutant IL7R signaling molecules in two steroid-sensitive T-ALL cell lines (SUPT1 and P12 Ichikawa cells) using inducible lentiviral expression constructs. We found that expressing mutant IL7R, JAK1, or NRAS, or wild-type NRAS or AKT, specifically induced steroid resistance without affecting sensitivity to vincristine or L-asparaginase. In contrast, wild-type IL7R, JAK1, and JAK3, as well as mutant JAK3 and mutant AKT, had no effect. We then performed a functional study to examine the mechanisms underlying steroid resistance and found that, rather than changing the steroid receptor’s ability to activate downstream targets, steroid resistance was associated with strong activation of MEK-ERK and AKT, downstream components of the IL7R signaling pathway, thereby inducing a robust antiapoptotic response by upregulating MCL1 and BCLXL expression. Both the MEK-ERK and AKT pathways also inactivate BIM, an essential molecule for steroid-induced cell death, and inhibit GSK3B, an important regulator of proapoptotic BIM. Importantly, treating our cell lines with IL7R signaling inhibitors restored steroid sensitivity. To address clinical relevance, we treated primary T-ALL cells obtained from 11 patients with steroids either alone or in combination with IL7R signaling inhibitors; we found that including a MEK, AKT, mTOR, or dual PI3K/mTOR inhibitor strongly increased steroid-induced cell death. Therefore, combining these inhibitors with steroid treatment may enhance steroid sensitivity in patients with ALL. The main limitation of our study was the modest cohort size, owing to the very low incidence of T-ALL.
Using an unbiased sequencing approach, we found that specific mutations in IL7R signaling molecules underlie steroid resistance in T-ALL. Future prospective clinical studies should test the ability of inhibitors of MEK, AKT, mTOR, or PI3K/mTOR to restore or enhance steroid sensitivity and improve clinical outcome.
| Although modern treatment protocols have drastically increased the cure rate among patients with T cell acute lymphoblastic leukemia (T-ALL), nearly 40% of patients require the most aggressive treatment regimen, significantly increasing the risk of harmful treatment effects later in life. These detrimental effects can include growth defects, bone necrosis, heart failure, and an increased risk of developing secondary malignancies. Moreover, treatment outcome for relapsed T-ALL patients is extremely poor.
Steroids are the cornerstone chemotherapeutic drug in the treatment of acute lymphoblastic leukemia (ALL), including T-ALL. However, steroid resistance is common among patients and is associated with poor outcome and an increased risk of relapse.
The mechanisms underlying steroid resistance in patients with ALL are poorly understood.
Therefore, we performed an unbiased, comprehensive genetic analysis of pediatric T-ALL, as well as in vitro functional analyses to validate associations between the identified mutations and steroid resistance.
We performed whole genome and targeted exome sequencing in patients with T-ALL and identified mutations in 151 genes, many of which are involved in cytokine signaling, transcriptional regulation, cell death, cell cycle, chromatin modification, and cellular transport.
Mutation data were integrated with changes in chromosomal copy number and were correlated with the patients’ clinical features and underlying biological characteristics. Mutations in the IL7R signaling components JAK1 and KRAS were correlated with steroid resistance and poor outcome.
Sequencing of IL7R signaling molecules in a larger pediatric T-ALL cohort revealed mutations in 32% of patients.
Expressing specific mutant and/or wild-type IL7R signaling molecules in two steroid-sensitive T-ALL cell lines induced steroid resistance via robust downstream signaling through MEK-ERK and AKT, thereby reducing steroid-induced apoptosis. Moreover, treating these cells with inhibitors of IL7R signaling restored steroid sensitivity.
Primary T-ALL cells obtained from patients were treated with steroids either alone or in combination with IL7R signaling inhibitors. We found that including these inhibitors significantly enhanced steroid-induced cell death.
These results should be tested further in prospective patient cohorts, to investigate the possibility that including IL7R signaling inhibitors in treatment regimens could restore or enhance steroid sensitivity in patients with ALL, thereby improving clinical outcomes.
| In children with acute lymphoblastic leukemia (ALL), response to therapy, including in vitro or in vivo steroid response, is a strong predictor of survival and cure [1–3]. ALL can be classified as T cell ALL (T-ALL) or B cell precursor ALL (BCP-ALL): T-ALL, particularly, has a high risk of relapse and is refractory to further treatment due to acquired therapy resistance. The mechanisms that underlie steroid resistance are poorly understood. In contrast to cell lines, which often harbor mutations and/or deletions in the steroid receptor NR3C1 [4], mutations are relatively rare among patients with ALL [5,6]. Upon steroid binding, NR3C1 translocates to the nucleus and drives the expression of target genes [7]. To date, steroid resistance has not been associated with reduced NR3C1 expression, expression of NR3C1 splice variants [8–10], or reduced expression of chaperone proteins [11,12]. Therefore, steroid resistance seems to be independent of changes in the NR3C1 gene itself in most patients with steroid-resistant T-ALL. Several mechanisms have been proposed to explain steroid resistance in T-ALL including activation of AKT1, which phosphorylates serine 134 of NR3C1, thereby preventing nuclear translocation [13]. Also, elevated MYB and BCL2 concentrations may promote survival following steroid treatment [14]. Activated NOTCH1 may confer steroid resistance by repressing expression of NR3C1 and PTEN [15]. Mutations in RAS have been shown to be associated with steroid resistance in BCP-ALL and are prevalent in relapsed patients [16–18]. Recently, CASP1 and its activator, NLRP3, were also shown to be associated with steroid resistance in ALL [19].
In this study, we aimed to provide an unbiased and comprehensive analysis of the molecular mechanisms that drive T-ALL and to resolve the cellular mechanisms that underlie steroid resistance. For this, we performed whole genome sequencing (WGS) and targeted exome sequencing (TES) in diagnostic patient samples obtained from pediatric T-ALL patients. Mutation data were integrated with copy number changes as determined by array comparative genomic hybridization (aCGH) to capture the full complexity of genomic mutations in T-ALL. Identification of steroid resistance mechanisms may provide therapeutic treatment options to improve sensitivity to this cornerstone chemotherapeutic drug in ALL treatment, improve cure rates, and help reduce detrimental late side effects of intensive treatment schedules through dose reduction.
This study did not have a protocol or prospective analysis plan. An outline of this study is provided in Fig 1. Briefly, to obtain insight into the genetic landscape of pediatric T-ALL, we performed WGS on paired diagnostic–remission samples from 13 patients covering all of the most predominant genetic subtypes in T-ALL. Recurrence of identified mutations was then established by applying a TES approach to a cohort of diagnostic samples from 69 well-characterized pediatric T-ALL patients, and these mutation data were further integrated with copy number data for mutant genes as obtained by aCGH. The mutation/aberration statuses of 151 genes that were identified were then correlated with the patients’ clinical features and underlying biological characteristics including in vitro drug response, T-ALL subtype, and outcome. We found that mutations in components (KRAS and JAK1) of the IL7R signaling pathway correlated with steroid resistance and poor outcome. We then used a PCR–Sanger sequencing approach to identify mutations in other IL7R signaling components, including IL7R, JAK1, JAK3, NF1, NRAS, KRAS, and AKT genes, in an expanded cohort of diagnostic patient samples including these 69 patients and 77 additional T-ALL patients. The association between mutant signaling components and steroid resistance was then functionally explored in two steroid-sensitive T-ALL cell lines (SUPT1 and P12 Ichikawa), and the ability of IL7R signaling inhibitors to revert steroid resistance in these models was tested. Moreover, the ability of these inhibitors to restore or enhance steroid sensitivity was then investigated in primary leukemic cells isolated from 11 T-ALL patients.
Diagnostic primary leukemia samples from 162 pediatric T-ALL patients were used for this study (Fig 1; S1 Table). In the discovery phase, we used DNA isolated from matched pre-treatment (diagnostic) and post-treatment (remission) sample pairs from 13 pediatric T-ALL patients who enrolled in the Dutch Childhood Oncology Group (DCOG) ALL-10 protocol between 2004 and 2012. In the expansion phase of this study, we used DNA from diagnostic, pre-treatment patient material from 69 patients who enrolled in the German Co-operative Study Group for Childhood Acute Lymphoblastic Leukemia (COALL) protocol between 1997 and 2003 (COALL-97). Initial findings were then confirmed in a larger cohort of diagnostic, pre-treatment T-ALL patient materials including DNA from the previously mentioned 69 COALL patients plus that of five additional patients who also enrolled in the COALL-97 protocol and 72 additional T-ALL patients who enrolled in the DCOG protocols ALL-7/8 (n = 30) or ALL-9 (n = 42). Functional validation was done on viably frozen pre-treatment leukemia cells from peripheral blood or bone marrow from eight of the 74 patients who enrolled in the COALL-97 protocol (as mentioned above) plus two additional patients who enrolled in the COALL-03 study and one patient who enrolled in the DCOG ALL-10 study. Median follow-up for patients who enrolled in the DCOG ALL-7/8/9 (1988–2004) and COALL-97 protocols was 67 and 52 mo, respectively. The patients’ parents or legal guardians provided informed consent to use leftover diagnostic material for research with approval from the institutional review board of the Erasmus MC Rotterdam and in accordance with the Declaration of Helsinki. Leukemia cells were harvested from blood or bone marrow samples and enriched to a purity of at least 90% as described previously [20].
Sequencing of 13 T-ALL tumor pairs in the discovery cohort (S1 and S2 Tables) was performed at Complete Genomics (Mountain View, California) using unchained combinatorial probe-anchor ligation chemistry on arrays of self-assembling DNA nanoballs producing 35-bp paired-end reads [21]. The average gross mapping yield for the 26 genomes was 183 Gb. On average 96.5% of the genome was called with 55× coverage or higher. Structural variants in the whole-genome-sequenced DNA of the 13 patients were detected using Complete Genomics’ cgatools (version 2.0.2.17) and were compared to the NCBI reference genome build 37. For this, junctions were identified that were not adjacent according to the reference sequence. Subsequently, somatic junctions were discovered as junctions detected in the tumor samples that were absent in the matching normal samples. Finally, we filtered for somatic high-confidence junctions using the following criteria: (1) there are at least ten mate pairs in the discordant reads cluster; (2) the de novo assembly of the junction is successful; (3) the junction exhibits high mapping diversity, i.e., the distance between the first position of the left-most mate read and the last position of the right-most mate read in the discordant reads cluster is more than 70 bp; (4) known underrepresented repeat sequences are not involved, e.g., ALR/Alpha; (5) the variant is absent in dbSNP build 132; (6) the variant does not result from deletion events of transposable elements (AluY and L1 subclasses); and (7) the variant is absent in 52 normal genomes that served as the baseline reference set (ftp://ftp2.completegenomics.com/Baseline_Genome_Set/SVBaseline/).
For mutation detection, reads were aligned to the NCBI build 36 reference genome by applying a local de novo assembly approach, and variations were called using Complete Genomics software v1.8 and v1.12. In each sample, 3.4 × 106 single nucleotide variants (SNVs), 221.7 × 103 small insertions, 234.5 × 103 small deletions, and 77.3 × 103 substitutions were detected on average. For this study, we focused on tumor-specific, non-synonymous mutations in exons, excluding sequence polymorphisms as present in NCBI dbSNP databases or the 1000 Genomes Project databases. This revealed 460 SNVs and 808 small insertions or deletions (INDELs) (S1 Fig) in the exons of the 13 T-ALL patients in total. To optimize thresholds for Complete Genomics quality parameters, we validated 46 mutations by PCR and Sanger sequencing. We found two quality parameters informative to distinguish true from falsely called mutations (S2A Fig). One is total score (TS), representing the confidence in the called mutation. The other is somatic score (SS), representing the confidence that the mutation is present in the tumor and absent in the matched remission sample. These two scores were calculated using Complete Genomics software v1.8 and v1.12 and CGA Tools 1.4.0. Based on the receiver operating characteristic (ROC) curves (S2B Fig), SS ≥ 0.1 and TS ≥ 100 were used as thresholds to reliably call somatic mutations with high confidence. Based on these thresholds, 137 genes were found to carry 178 high-confidence SNV or INDEL mutations (S1 Fig). The WGS sequence data of 26 genomes, aligned to the human reference genome (NCBI build 36), have been deposited in the European Nucleotide Archive with accession numbers ERS934791–ERS934816.
PCR reactions were performed using 25–50 ng of genomic DNA, 300 nM primers, 200 μM dNTPs, 2 mM MgCl2, and 1.25 units of AmpliTaq Gold (Applied Biosystems) in 1× PCR Buffer II (Applied Biosystems) in a volume of 50 μl. PCR products were purified with the Millipore Vacuum Manifold filter system and sequenced (BigDye Terminator v3.1 Cycle Sequencing Kit, Applied Biosystems) on the ABI PRISM 3130 DNA Analyzer (Applied Biosystems).
To compare WGS with TES variant calling and to validate and determine quality parameters for WGS and TES, respectively, we performed TES for 410 mutated genes as found by WGS in the diagnostic samples of the 13 discovery cohort patients (S1 and S2 Tables). These genes included all 137 genes with high-confidence mutations and 273 mutated genes that had lower WGS quality scores.
Enrichment of exonic DNA sequences from genomic DNA was performed using Agilent SureSelect MP4 arrays (2-Mb capture region, with >2× average coverage). Sequencing was performed at ServiceXS. (Leiden, The Netherlands) using the Illumina HiSeq 2000 platform, which produced 100-bp paired-end reads. Image analysis, base calling, and quality check were performed using the Illumina data analysis pipeline RTA v1.13.48 and/or OLB v1.9 and CASAVA v1.8.2. Prior to alignment, reads were filtered based on the following quality thresholds: read trimming at Phred quality score 20 and minimal read length 36. The reads were then aligned to the NCBI build 37 reference genome using a short read aligner based on Burrows-Wheeler Transform [22] with a mismatch rate of 4%. SNV and INDEL mutations were identified using a ServiceXS in-house pipeline based on Bayesian statistics with the following thresholds: minimal coverage of 5× for INDELs and 10× for SNVs, with a minimal variant frequency of 30%.
Mutations detected by both WGS and TES platforms can be regarded as true mutations and hence helped to define the quality threshold parameters for the TES approach. More than 60% of predicted exonic, non-synonymous SNVs identified by TES had not been identified by WGS (S3 Fig). Exploiting the overlapping mutations detected by both platforms, we trained a one-class classifier to determine the quality parameter boundaries for mutation detection by TES. The TES quality parameters include read depth and the variant overall quality score. For SNV calling, we trained a Gaussian one-class classifier using ddtools [23] (S3A Fig). The performance of the Gaussian classifier is presented by the ROC curve in S3B Fig. The boundary was chosen to call mutations with a false negative rate (FNr) of 0.10, resulting in a false positive rate (FPr) of 0.28. Given the distribution of the INDEL training set as shown in S3C Fig, a simple decision tree classifier for calling INDEL mutations was built that resulted in a FNr of 0.10 and a FPr of 0.63, confirming the low concordance of INDEL mutations by both platforms.
To evaluate the recurrence of 137 high-confidence mutated genes and seven validated low-confidence mutated genes as identified by WGS (S2 Table), we sequenced the diagnostic samples of 69 genetically and clinically well-annotated T-ALL patients by TES for these 144 genes, together with an additional set of 110 genes that are recurrently mutated in leukemia [24,25] (S1 and S3 Tables). The mean coverage by TES was 561×, and 89% of all captured exons were covered by more than 100 reads. A stringent filtering strategy, as used for WGS (S4 Fig), was then applied to the TES results. Polymorphisms present in an expanded panel of genetic variation databases (listed in the next section) were excluded, providing an effective filter to exclude germline variants in the expanded cohort of 69 T-ALL patients for whom paired normal samples were not available [26]. The TES sequence data of the 13 discovery cohort patients and the 69 expansion cohort patients have been aligned to the human reference genome (NCBI build 37) and are deposited in the European Nucleotide Archive with accession numbers ERS935731–ERS935812.
We increased the number of genetic variation databases to filter out polymorphisms from the mutation dataset as determined by TES by including dbSNP [27] versions 130, 131, 132, and 135; 1000 Genomes Project database [28] versions 2010 November and 2012 April; the ESP6500 database (http://evs.gs.washington.edu/EVS/); the Core and Diversity Panels in Huvariome [29], and an in-house dataset of 1,302 exomes from Radboud University Medical Center, Nijmegen, The Netherlands [30]. In addition, we excluded mutations that were also identified in the 13 remission samples by WGS. We retained mutations that represented polymorphisms according to the databases mentioned above when variations involving the same amino acids were annotated in the COSMIC database. By applying stringent filtering against a broad range of genetic variation databases, we aimed to remove as many germline variants as possible. To segregate germline polymorphisms from somatic mutations in patient samples in the absence of paired normal control samples, variations observed in thousands of unrelated individuals provide an effective filter [26].
We performed copy number analysis based on aCGH using SurePrint G3 Human CGH 2×400K arrays (Agilent Technologies) in diagnostic leukemia samples of 53 out of the 69 expansion cohort patients whose genomes were sequenced by TES (S3 Table). The array images were processed to obtain the log10 ratio of red and green channel signals, after background correction and dye normalization using Agilent Feature Extraction software (version 10.5.1.1). The probe chromosomal locations were annotated using the NCBI reference genome build 36. Afterwards, the log ratios were subjected to a noise-reduction algorithm, the Waves aCGH Correction Algorithm [31], to reduce the wave artifact characterized by an undulating aCGH profile along the chromosome. This algorithm corrects biases that are caused by differences in labeling intensities of DNA fragments, which depend on GC content and fragment size as well as the hybridization efficiencies of labeled fragments to corresponding hybridization probes on the array. Afterwards, the log10 ratios (L) were converted to copy number (CN) as CN = 2 × 10L. Amplifications and deletions were then called if three consecutive probe sets in a gene had a CN value beyond average CN ± 2 × the standard deviation of CN of the corresponding probe sets among all samples.
The allele-specific copy number of the diagnostic sample from patient #10793 was obtained using Affymetrix SNP Array 6.0 with 680-bp median intermarker spacing. For this, 500 ng of genomic DNA isolated from the diagnostic and remission samples was used as input. Raw signal intensities were analyzed using Partek Genomics Suite. First, probe intensities were adjusted for a number of properties that are correlated with intensity, including fragment length, GC content, and other sequence-based hybridization bias. After quantile normalization, paired analysis was performed to generate allele-specific copy numbers by comparing the diagnostic sample to the remission sample.
Pathway enrichment analysis was performed on 127 genes that had mutations or copy number aberrations in more than one patient sample in the expansion cohort. Two approaches were applied. One used the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7. The functional annotation clustering reports of enriched biological pathways and molecular functions annotated by Gene Ontology can be found in S4 Table. The other approach used Ingenuity Pathway Analysis to identify enriched canonical pathways in these 127 recurrent genes. Default settings were applied except for setting the species to human only. The p-values indicated are calculated using a two-sided Fisher’s exact test.
The gene expression profiling dataset of the 117 pediatric T-ALL cases as produced by microarray [32] is available at GEO (http://www.ncbi.nlm.nih.gov/geo/) under accession number GSE26713.
For the 69 COALL cohort samples in the expansion phase, we related the mutational statuses and copy number aberrations in 151 genes (S3 Table) to three clinical/biological parameters: T-ALL subtype (as defined previously [32]), in vitro prednisolone LC50 level, and survival outcome. In these analyses, the presence of mutations (Mut), deletions (Del), amplifications (Amp), and the combinations MutDel, MutAmp, or MutAber (i.e., mutations and/or copy number aberrations including deletions and amplifications) in each gene was tested against the clinical/biological parameters. Each gene with a specific feature (e.g., JAK1_Mut) was tested individually. The features are binary (present/absent), which means multiple incidences of the same feature in the same gene for the same patient sample are aggregated. For example, JAK1_Mut = “present” in a patient sample means that there is at least one JAK1 mutation identified in the sample. Associations between any of these six features and T-ALL subtype were calculated using Fisher’s exact test. Associations with in vitro prednisolone LC50 were calculated using the Kruskal-Wallis test. All p-values are two-sided. Associations with survival were calculated using the log-rank test. For all test results, p-values were used to prioritize our hypotheses on associations and therefore remained nominal without multiple testing corrections. A threshold of nominal p < 0.05 was used to present findings that are potentially more relevant. Using this threshold, genes that are associated with reduced drug sensitivity, poor event-free or relapse-free survival, or particular T-ALL subtypes are summarized in S5 Table. Given the data availability of the patient samples (S1 Table), the number of tests varies for different associations and is indicated in S5 Table.
In the integration phase, we validated the associations identified in the expansion phase. For this, we performed PCR–Sanger sequencing for IL7R, JAK1, JAK3, NF1, NRAS, KRAS, and AKT genes (i.e., IL7R signaling pathway genes) in the confirmation cohort of 146 patients comprising the 69 patients from the expansion phase and 77 additional pediatric T-ALL patients. The presence of mutations in these genes as well as available information on genetic aberrations affecting the PTEN gene were used to segregate patients into three groups, i.e., those with mutations in the IL7R signaling pathway, those with mutations in PTEN, and the rest of the patients. Association between IL7R pathway mutations and T-ALL subtype was tested by Fisher’s exact test. Association between IL7R pathway mutations and in vitro prednisolone LC50 was tested by Kruskal-Wallis test in 97 patients for whom in vitro prednisolone response data were available. Association between IL7R pathway mutations and survival was tested using the log-rank test.
Gateway multi-site recombination (Invitrogen) was used to simultaneously clone multiple DNA fragments into our Gateway-adapted lentiviral pLEGO-iC2 destination vector (Addgene). First, the parental pLEGO-iC2 vector was converted into a Gateway destination vector by replacing the pSFFV-MCS-IRES-mCherry insert with the Gateway Cassette A. Cassette A, comprising the ccdB and chloramphenicol resistance genes flanked by attR recombination sites, was amplified by PCR and cloned into the ApaI/PciI-digested pLEGO-iC2 vector. Lentiviral expression vectors were assembled using Gateway recombination of the lentiviral destination vector with four synthetically synthesized entry vectors (Eurogentec): (1) attL1/attR5-flanked doxycycline-inducible promoter (third generation, Clontech); (2) attL5/attL4-flanked human cDNA sequence; (3) attR4/attR3-flanked DDK-tag followed by a stop codon, WPRE sequences, and a constitutive pSFFV promoter; and (4) attL3/attL2-flanked TETon-T2A (Thosea asigna virus 2A peptide)–puromycin resistance cassette or TETon-T2A-LNGFR (truncated or ΔNGFR) reporter. LR-recombination reactions were performed according to the manufacturer’s instructions. For the shRNA experiments, PLKO.1-puro lentiviral shRNA constructs directed against the human NR3C1 gene were selected from the MISSION T shRNA Library (Sigma-Aldrich). For lentivirus production, HEK293T cells were transfected with lentiviral expression vector DNA and pMD2.G (VSV-G), pMDLg/pRRE, and pRSV-REV support vectors (Addgene) using X-tremeGENE HP DNA Transfection Reagent (Roche). Transfection was performed in DMEM supplemented with 10% heat-inactivated fetal calf serum (FCS), 1× Glutamax, 1% penicillin/streptomycin, and 0.25 μg/ml Fungizone, and the HEK293T cells were cultured overnight in a humidified incubator at 37°C and 5% CO2. Following transfection, lentivirus particles were produced and collected in serum-free Opti-MEM1 (Thermo Fisher Scientific) for up to 48 h. Culture medium containing lentiviral particles was collected, filtered through a 0.45 μM Minisart filter (Sartorius), and concentrated by centrifugation at 4°C using a VIVASPIN 20 concentration column (Sartorius). Viral particles were stored at −80°C.
Lentiviral transduction was used to obtain SUPT1 and P12 Ichikawa T-ALL cell lines containing a variety of expression vectors. For this, viral batches were serially diluted in 96-well plates in a total volume of 50 μl of Opti-MEM1 medium, and 50,000 cells were added in Advanced RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 2% heat-inactivated FCS, 1× Glutamax, 1% penicillin/streptomycin, and 0.25 μg/ml Fungizone. Cells were incubated at room temperature for 30 min on a gently shaking platform, and further incubated for several hours in a humidified incubator at 37°C and 5% CO2. For overnight incubation, FCS was added to a final concentration of 10%, after which the medium was refreshed. The transduction efficiency was determined 4 d later. To avoid multiple integrations per cell, only cells that were transduced by optimal virus concentrations (maximal transduction rate of 50%) were further cultured in bulk and enriched by selection in medium containing 1–2 μg/ml puromycin. The bulk-selected cells were used for subsequent experiments.
Bulk transduced lines were maintained in culture medium at a concentration of 0.25–1.5 × 106 cells/ml and refreshed twice weekly. To induce expression from the lentiviral vector, cells were grown in the presence of 0.5 mg/ml doxycycline prior to the cytotoxicity testing. Cytotoxicity was tested in the presence of a single concentration or a serial dilution of a variety of drugs or inhibitors as indicated: prednisolone (15 mg/ml–0.007 μg/ml), L-asparaginase (100–0.032 IU/ml), vincristine (2.5–0.10 ng/ml), 2 μM ruxolitinib (JAK1 inhibitor, Selleck Chem #S1378), 2 μM MK2206 (AKT inhibitor, Selleck Chem #S1078), 10 μM CI1040 (MEK inhibitor, Axon Medchem #Axon 1368), and 2 μM CAS 667463-62-9 (GSK3 inhibitor IX, Santa Cruz Biotechnology #sc-202634). Cells were incubated for 72 or 96 h in a humidified incubator at 37°C and 5% CO2 as specified in figure legends. Viable cell numbers were counted based on forward- and side-scatter parameters using flow cytometry (MACSQuant, Miltenyi Biotec), and data were analyzed using FlowJo software v10 (Treestar). Percent of viable cells after drug exposure was normalized to the percentage of viable untreated control cells, whereas percent of viable cells after exposure to a drug in combination with an inhibitor was normalized to the viability of inhibitor-treated control cells. The selected concentration of the inhibitors was such that the viability of inhibitor-treated control cells deviated not more than 20% compared to the viability of untreated control cells. Graphs were made using Graphpad PRISM 6 software.
Primary antibodies used for Western blot detection of proteins were obtained from Cell Signaling Technology, unless specified otherwise: phospho-AKT S473 (#9271), phospho-JAK1 (#3331), phospho-STAT5 (#9351), phospho-MEK1/2 (#9154), phospho-ERK1/2 (#4370), phospho-mTOR S2448 (#2971), phospho-p70S6 Kinase T421/S424 (#9204), BAD (Abcam #AB32455), phospho-BAD S136 (#4366), BIM (Abcam #AB15184), phospho-BIM S55 (#4550), phospho-BIM S69 (#4581), phospho-CREB S133 (#9198), phospho-ATF1 (#9198), BCL2 (Santa Cruz Biotechnology #sc-130308), BCLXL (#2764), MCL1 (Sigma #HPA008455), PUMA (Abcam #AB33906), cMYB (#12319), phospho-IKKab S176/S177 (#2978), phospho-IKKab S176/S180 (#2697), GSK3AB (#5676), phospho-GSK3AB S21/9 (#9331), phospho-p38/MAPK T180/Y182 (#4511), GCR (Santa Cruz Biotechnology #sc-1003), phospho-GCR S211 (Abcam #AB3579), DYKDDDDK Tag Antibody (#2368), CD127 (IL7R) (R&D Systems #MAB306), RAS (Millipore #05–516), and β-actin (Abcam #AB6276). Following staining of Western blots using IRDye fluorescent secondary antibodies, fluorescent intensities were scanned and quantified using an Odyssey imager (LI-COR) and normalized to β-actin expression levels.
Single drug or inhibitor cytotoxicity assays were carried out as described previously in 384-well plates for 72 h [33]. Briefly, 10,000 primary T-ALL patient cells from viably frozen stocks were plated in 45 μl of culture medium (Advanced RPMI 1640 medium [Thermo Fisher Scientific] containing 20% heat-inactivated fetal bovine serum, 1× Glutamax, 1% penicillin/streptomycin, 25 μg/ml gentamycin, and 0.25 μg/ml Fungizone) 2 h before drugs or inhibitors were added. To determine the IC50, each drug or inhibitor was dissolved in 100% DMSO in approximately 10,000-fold the estimated IC50 concentration, diluted in a nine-point dilution series in duplicate in DMSO, which was then 100-fold diluted in HEPES buffer (pH = 7.4). For the assay, 5 μl of each single drug or compound dilution was added to the preplated cells in a final concentration that ranged from 1 nM to 10,000 nM for all compounds in 0.1% DMSO. As an indirect measure of total cell viability after 72 h of incubation, the intracellular ATP content was measured. The plates were cooled to room temperature, and the cells were incubated with 25 μl of ATPlite 1step solution (PerkinElmer) per well. Luminescence was recorded on an Envision multimode reader (PerkinElmer), and results were normalized to the measurements of control wells containing only 0.1% DMSO. IC50 values were fitted manually by nonlinear regression using XLfit5 software (IDBS). Maximum and minimum signals were locked, where appropriate, to obtain the best fit as indicated by the F-test as implemented in XLfit5. If IC50 values did not fall within the tested concentration range, drugs or inhibitors were retested after further dilution. For curve shift synergy experiments of drug–inhibitor combinations, inhibitor stocks were diluted in DMSO on the day of the experiment to concentrations of 10,000 times their IC50 values, as determined in the single agent experiments. Single drug or inhibitor dilutions as well as drug–inhibitor combination mixtures in 1:1, 4:1, and 1:4 ratios were prepared, and diluted in DMSO into seven-point dose–response series in duplicate. Further dilution and addition to the cells was performed as described above. Because of the IC50 matching, all stock solutions and mixtures would be equipotent in the absence of synergy or antagonism. For the single drug or inhibitor experiments, final assay concentrations were between 10 and 0.01 times their IC50 values. For single drug or inhibitor curves, IC50 values were fitted on the percent-effect data by nonlinear regression using XLfit5. To correct for interassay variation, the IC50 values of the single agents in the synergy experiment were used to calculate the synergy for the drug–inhibitor combination experiments. The combination index (CI) as a quantitative measure of synergy for drug–inhibitor combinations [34] was calculated as described previously [35]. CI values lower than 1.0 indicate synergy of drug–inhibitor combinations; CI values lower than 0.3 indicate strong synergy. CI values higher than 1.5 indicate antagonistic effects of drug–inhibitor combinations.
To determine the mutational landscape of pediatric T-ALL patients, we used an integrated approach combining mutation data, copy number data, and clinical information obtained from 82 pediatric T-ALL patients (Fig 1; S1 Table). First, we performed WGS on paired diagnostic–remission samples from 13 patients at Complete Genomics [21]. These patients covered all known genetic subtypes including two early thymic progenitor ALL (ETP-ALL) cases; cases harboring TLX3 (five cases), TLX1 (one case), NKX2-1 (one case), HOXA (one case), KTM2A/MLL (one ETP-ALL case), or TAL1 (one case) chromosomal rearrangements; and cases for which driving oncogenic events remain unknown (three patients, including one ETP-ALL case) [32]. Extensive filtering (S1 Fig) and first quality parameter optimization was performed that was validated by PCR–Sanger sequencing (S2 Fig). We established the optimal quality parameter thresholds for WGS analysis to robustly call high-confidence mutations as being 0.1 and 100 for SS and TS, respectively. We further confirmed these thresholds by TES using the Illumina HiSeq 2000 platform of 410 genes from the diagnostic samples of these same 13 T-ALL patients (S1 and S2 Tables). These genes included 137 genes containing 178 high-confidence mutations (85 SNVs and 93 INDELs) and 273 genes containing 377 predicted low-confidence mutations (139 SNVs and 238 INDELs). We confirmed 74 out of 85 (87%) high-confidence SNVs and two out of 93 (2%) high-confidence INDEL mutations, in contrast to seven out of 139 (5%) low-confidence SNVs and zero out of 238 low-confidence INDEL mutations. These findings indicate that the chosen quality parameter thresholds for WGS were suitable to identify high-confidence mutations. Moreover, the overlap between WGS and TES results was further used to optimize the TES quality parameter threshold setting (S3 Fig).
In addition to mutations, we identified by WGS 185 somatic intrachromosomal breakpoint junctions (average 14 per patient, range 5–25) and 40 interchromosomal breakpoint junctions (average three per patient, range 2–11) that represented 183 predicted rearrangements including 78 deletions, 28 duplications, 16 inversions, 16 translocations, and 45 complex rearrangements in these 13 patients (S2 Table). In all, 182 out of 225 chromosomal junctions had evidence for inclusion of random nucleotides (average 18 nucleotides, range 1–221) that point to the involvement of the RAG recombination machinery. This presence of random nucleotides was also the case for most chromosomal junctions identified in the two ETP-ALL patients, both of whom arrested at early thymic progenitor stages that precede RAG-mediated T cell receptor gene recombination events. Breakpoints for known driving oncogenic rearrangements as predicted using fluorescence in situ hybridization or real-time quantitative PCR were identified in ten out of 13 T-ALL patients. We could not pinpoint any specific oncogene-driving event from the various rearrangements identified in two out of three T-ALL cases for which driving oncogenic rearrangements were unknown (#9255 and #9343). One ETP-ALL patient (#10793) had evidence for chromothripsis in Chromosomes 7 and 14 (Fig 2A). Breakpoints for deletions, insertions, duplications, complex rearrangements, and translocations were identified in both chromosomes that flank areas of allelic losses or gains according to SNP-based microarray analysis (Fig 2B). Multiple breakpoints were identified close to the HOXA gene cluster. Multiple clustered breakpoints frequently affected TCRAD and BCL11B loci and pointed to a rather organized form of chromothripsis in this particular patient rather than random reassembly of disintegrated chromosomal segments as frequently observed in patients with chromothripsis [36].
We then expanded our mutation analysis by performing TES on a cohort of diagnostic samples from 69 well-characterized pediatric T-ALL patients (Fig 1; S1 Table) [37]. In total, 254 genes were sequenced at a mean coverage of 561×, including the 144 validated genes identified in the discovery cohort and 110 additional selected genes that are frequently mutated in ALL (S3 Table) [24,25]. Following stringent filtering (S4 Fig), 401 mutations were identified in 151 genes (S3 Table). These mutations were then integrated with aCGH copy number data, which were available for 53 out of 69 patients (S3 Table). The median number of genes harboring mutations and/or copy number changes was six per patient (range 0–51). ETP-ALL patients had more aberrations than other subtypes (p < 0.001, median 12, range 5–51) [32,38,39], while TALLMO patients had fewer aberrations (p = 0.003, median 4, range 0–42). The median number of aberrations in TLX and proliferative subtypes was 8.5 (range 4–10) and 6 (range 3–27), respectively. Sixty-six of the mutated genes have been previously identified in T-ALL, and 83 mutated genes have been previously found in other types of leukemia or nonhematopoietic tumors. Two genes, RABL6 (a GTP-binding member of the Ras superfamily of small GTPases) and IGHV3-64, have not previously been reported in mutated form in human cancer. Mutations mostly affected genes that are involved in cytokine signaling, transcription regulation, cell death, cell cycle, or chromatin modification or that encode transporters (Fig 2C; S4 Table).
The mutation/aberration statuses of the 151 genes were then correlated with clinical and biological parameters of these 69 patients: in vitro drug response, T-ALL subtype, and outcome. Several mutations and aberrations were associated with poor survival or drug resistance (S5 Table). Interestingly, mutations in IL7R signaling molecules—including JAK1 and KRAS—correlated with prednisolone resistance and reduced survival (Fig 3A–3C). To validate these associations, we expanded our mutation analyses by PCR–Sanger sequencing to the IL7R, JAK1, JAK3, NF1, NRAS, KRAS, and AKT genes in the diagnostic samples of these 69 patients and 77 additional T-ALL patients. Mutations were identified in 47 out of 146 patients (32%) and were associated with the ETP-ALL and TLX subtypes (p < 0.001; see S1 and S6 Tables for details). In vitro prednisolone response data on these diagnostic samples were available for 97 patients, and in 28 out of 97 cases IL7R signaling mutations were associated with steroid resistance (p = 0.033; Fig 3D; S6 Table). These patients also had a significantly worse outcome than patients lacking these mutations (p = 0.009; Fig 3E). Inactivating events in PTEN in 18 out of 97 patients are associated with the TALLMO subtype [20,40], and these patients had prednisolone sensitivity levels that were similar to those of patients lacking IL7R signaling or PTEN mutations. These data therefore suggest a potential relationship between IL7R signaling mutations and steroid resistance.
To functionally explore whether IL7R signaling mutations may drive steroid resistance, we tested whether expression of mutant IL7R signaling molecules—versus their wild-type counterparts—can confer steroid resistance in two steroid-sensitive T-ALL lines, namely SUPT1 and P12 Ichikawa. All mutant signaling molecules expressed from doxycycline-inducible expression constructs—including IL7RRFCPH, JAK1R724H, JAK1T901A, JAK3M511I, JAK3R657Q, NRASG12D, and AKTE17K—were able to trigger IL3-independent growth in Ba/F3 cells, in contrast to their wild-type isoforms, confirming that these mutations can transform cells [41]. In SUPT1 and P12 Ichikawa T-ALL backgrounds, expression of the cysteine mutant IL7RRFCPH conferred steroid resistance, while the wild-type IL7R and the non-cysteine IL7RGPSL mutant receptors did not (S5 Fig). Both mutant JAK1 molecules—but not wild-type JAK1—conferred steroid resistance (Fig 4A–4C). Surprisingly, JAK3 mutations did not confer resistance (Fig 4E and 4F), while expression of wild-type NRAS, NRASG12D, and wild-type AKT strongly conferred resistance to steroids (Figs 4G and S5). Although AKTE17K behaves as an activating mutant [42] that supports IL3-independent growth in Ba/F3 cells [41], it did not confer steroid resistance in SUPT1 or P12 Ichikawa cells. We therefore denoted bulk lines that expressed IL7RRFCPH, JAK1R724H, JAK1T901A, NRASG12D, or wild-type NRAS or AKT as the “steroid-resistant panel,” whereas lines that expressed wild-type IL7R, JAK1, or JAK3 or mutant JAK3M511I, JAK3R657Q, or AKTE17K as the “steroid-sensitive panel.” The expression of particular (mutant) IL7R signaling molecules specifically affected steroid response since all lines displayed comparable sensitivities to vincristine and L-asparaginase treatments (Figs 4H, 4I and S5).
We next investigated whether the expression of IL7R signaling molecules that give rise to steroid resistance affects nuclear shuttling of the steroid receptor NR3C1, rendering it incapable of activating downstream target genes upon steroid exposure. We found no difference among steroid-resistant and steroid-sensitive lines in their abilities to activate NR3C1 target genes following steroid exposure, including NR3C1, TSC22D3/GILZ, BCC3/PUMA, KLF13, BCL2L11/BIM and FKBP5 (S6 Fig). Therefore, the steroid resistance provoked by expression of mutant IL7R, JAK1, or NRAS molecules, or wild-type NRAS or AKT, is independent of the NR3C1 response following steroid exposure.
To examine the underlying mechanism that confers steroid resistance, we performed Western blot analyses on steroid-resistant and steroid-sensitive lines to measure the activities of the IL7R signaling and downstream pathways (Figs 5A and S7). Steroid-sensitive and -resistant lines had equal concentrations of total NR3C1 and equal serine 134 phosphorylation (Fig 5A and 5B). In contrast to steroid-sensitive lines, steroid-resistant lines had higher activation levels of the RAS-MEK-ERK and AKT pathways (Fig 5C–5E). Steroid-resistant lines also displayed higher levels of phosphorylated (activated) p70-S6K and (inactivated) GSK3B (Fig 5F), and higher activation of the CREB and NFκB pathways downstream of AKT, which resulted in significantly higher concentrations of antiapoptotic MCL1 and BCLXL (Fig 5G). Thus, robust IL7R pathway activation in steroid-resistant lines triggered a strong survival response, which may have overridden the proapoptotic NR3C1 response (Fig 5H). Although BIM is essential for steroid-induced apoptosis [43–46], total BIM concentrations were comparable among steroid-sensitive and -resistant lines. All steroid-resistant lines had higher levels of phosphorylated BIM except for the steroid-resistant AKT line (Fig 5A). Consistent with this, GSK3B, an important regulator of proapoptotic BIM [47,48], was strongly inactivated in all these resistant lines.
We then tested whether inhibitors of IL7R signaling can restore steroid sensitivity in steroid-resistant cells. Inhibitors of JAK1 (2 μM ruxolitinib), MEK (10 μM CI1040), and AKT (2 μM MK2206) were tested for their specificity in blocking IL7R signaling in steroid-resistant lines expressing IL7RRFCPH, JAK1T901A, AKT, or NRAS and their abilities to revert steroid resistance. Ruxolitinib blocked IL7R downstream signaling in both IL7RRFCPH and JAK1T901A lines, but was ineffective in blocking signaling in AKT or NRAS lines, as expected (Figs 6A, 6B and S8). Consistently, ruxolitinib treatment increased steroid sensitivity in IL7RRFCPH and mutant JAK1 lines (Fig 6C and 6E) but not in wild-type NRAS or AKT lines (Figs 6D, 6E and S8). The MEK inhibitor CI1040 blocked ERK activation in all four lines tested, reducing downstream activation of mTOR and p70-S6K. CI1040 treatment enhanced activation of GSK3B and caused a shift from phosphorylated to non-phosphorylated BIM in these lines, except for the AKT line. CI1040 treatment also resulted in elevated levels of active AKT, possibly due to a cellular feedback rescue mechanism. CI1040 partially restored steroid sensitivity in most steroid-resistant SUPT1 lines and was most effective in the wild-type NRAS and mutant NRASG12D lines (Figs 6C, 6D and 6F). CI1040 treatment efficiently enhanced steroid sensitivity in resistant P12 Ichikawa lines that expressed IL7RRFCPH or wild-type NRAS or AKT (S8 Fig). CI1040 also increased steroid sensitivity in some steroid-sensitive lines (Figs 6F and S8). The AKT inhibitor MK2206 blocked AKT signaling and reduced phosphorylated levels of downstream mTOR and p70-S6K (Fig 6A and 6B). MK2206 restored the steroid sensitivity phenotype in all steroid-resistant lines (Figs 6C, 6D, 6G and S8). Like CI1040 treatment, MK2206 treatment also increased steroid sensitivity in steroid-sensitive lines, possibly by inhibiting endogenous AKT. Similar effects were observed for treatment with the MEK-AKT inhibitor combination (Fig 6H). In contrast, inhibitor IX, a blocker of GSK3B activation, provoked resistance in the majority of steroid-sensitive lines as well as some steroid-resistant lines (Fig 6I), providing further evidence that GSK3B is a key regulator of steroid responsiveness.
We then further tested whether clinically relevant inhibitors of the IL7R signaling pathway affected the prednisolone sensitivity of primary leukemia cells isolated from 11 pediatric T-ALL patients at diagnosis (Tables 1 and S1). Leukemia cells were incubated with serial dilutions of prednisolone or single inhibitors as well as their combinations at different ratios based on estimated IC50 concentrations. The level of synergism was determined by calculating the CI for serial dilutions of three drug–inhibitor combination mixtures (1:1; 1:4, and 4:1) at effective doses that were lethal for 50% (ED50) or 75% (ED75) of the leukemia cells (S9 Fig). Three out of 11 patients harbored IL7R mutations, whereas one patient had a KRAS mutation. Leukemic cells from four other patients did not harbor mutations in IL7R signaling components but had other mutations, whereas the mutation status of three additional patients was unknown (Table 1). These 11 patient samples differed in their responses to prednisolone based on their IC50 values or the percentage of total responding cells (Table 1), and patients #2911 and #10110 were refractory to steroid treatment. Surprisingly, in contrast to previous results on cell lines, the JAK inhibitor ruxolitinib had no effect on the prednisolone response in the eight T-ALL patient samples for which CI values could be calculated. Treatment with the MEK inhibitor AZD6244 or trametinib synergistically enhanced prednisolone responsiveness in 5/9 and 9/10 patient samples, respectively. Neither inhibitor enhanced steroid responsiveness in the leukemia cells from steroid-resistant patient #2911, whereas trametinib synergistically enhanced the prednisolone response in leukemia cells from steroid-resistant patient #2322. AKT inhibitor MK2206 synergistically enhanced prednisolone sensitivity in leukemia cells from 8/11 patients, while mTOR inhibitors AZD8055 and everolimus acted synergistically when combined with prednisolone in 10/11 and 8/8 patient samples, respectively, including both steroid-resistant patients. The PI3K inhibitor NVPBEZ235 synergistically enhanced prednisolone sensitivity in 10/11 patient samples.
Here, we report the identification of mutations in IL7R signaling molecules in 32% of pediatric patients with T-ALL. These mutations are associated with reduced steroid sensitivity and poor clinical outcome. In addition, we provide functional evidence that these mutations reduce steroid-induced cell death by activating the downstream signaling pathways MEK-ERK and AKT. Activation of these pathways causes (1) upregulation of the antiapoptotic proteins MCL1 and BCLXL; (2) inactivation of the proapoptotic protein BIM, an essential component in steroid-induced cell death; and (3) inactivation of GSK3B, a key regulator of BIM. Importantly, in cell lines and primary patient samples, inhibitors of IL7R signaling restored and enhanced steroid sensitivity, respectively.
WGS is a powerful technique used to identify breakpoint junctions of chromosomal rearrangements that result in the activation of oncogenes or create fusion proteins. It validated chromosomal translocations in all ten patients that were predicted by fluorescence in situ hybridization or other techniques, and led to the identification of various chromosomal rearrangements that had not been previously identified in T-ALL. Pediatric T-ALL patients on average have three (range 0–11) interchromosomal and 14 (range 5–25) intrachromosomal junctions as a consequence of translocations, inversions, deletions, duplications, or complex rearrangements. Multiple and clustered chromosomal junctions were identified in two patients that provided evidence for chromothripsis of Chromosomes 7 and 14 (ETP-ALL patient #10793) or Chromosomes 1 and 5 (proliferative T-ALL patient #10943). Most intra- and interchromosomal junctions, including those found in ETP-ALL patients, contain insertions of non-template-derived nucleotides, which strongly implies the involvement of the RAG1/2-mediated recombination machinery in these rearrangements.
Based on the WGS data for 13 patients and TES data from an expanded cohort of 69 T-ALL cases that were integrated with LOH data, we identified 151 mutated genes, of which two genes had not been observed before in human cancer (RABL6 and IGHV3-64). Integrating this dataset with clinical and biological parameters identified IL7R signaling mutations as being associated with steroid resistance in pediatric T-ALL patients; these mutations may serve as biomarkers for reduced steroid response. Mutations affecting the IL7R signaling pathway and downstream AKT and MEK-ERK pathways were also associated with decreased relapse-free survival for pediatric T-ALL patients who had received DCOG or COALL treatment. Little is known about the prognostic significance of mutations in the IL7R-JAK/STAT pathway in T-ALL [49], and various studies have failed to demonstrate adverse effects for pediatric T-ALL patients harboring IL7R or IL7R/JAK mutations [50–52]. Our results are in line with the poor prognosis for patients with JAK1 mutations originally reported for adult T-ALL [53] and with the poor prognosis for patients with NRAS/KRAS mutations reported for adult T-ALL patients treated in the GRAALL-2003 and GRAALL-2005 trials [54]. One explanation for varying results among studies may be that mutation rates for individual signaling components are low. To the best of our knowledge, no study has performed a comprehensive analysis of the prognostic significance of the full spectrum of IL7R signaling mutations in T-ALL before. These signaling mutations affect 32% of T-ALL patients, and are most recurrent in ETP-ALL patients ([40,52] and this study) and patients with the TLX subtype of T-ALL, which mostly harbors HOXA-activating aberrations or TLX3 translocations [32].
Functional validation experiments in two T-ALL cell lines demonstrated that mutated IL7R signaling molecules robustly activate the downstream molecules AKT and MEK-ERK, thereby inducing steroid resistance. These include mutations in IL7R and JAK1, but not JAK3. Mutations in JAK3 only modestly activate signaling, presumably due to their dependence on JAK1 [55]. The IL7R signaling mutations do not affect NR3C1 signaling directly, including its nuclear translocation or its ability to transactivate target genes upon steroid exposure. These findings are consistent with our previous finding that steroid treatment activates NR3C1 target genes similarly between steroid-sensitive and steroid-resistant ALL patients [56], supporting the notion that resistance mechanisms are downstream—or independent—of steroid-induced NR3C1 transactivation. We found no evidence of differential NR3C1 serine 134 phosphorylation between the steroid-sensitive and steroid-resistant groups (NR3C1 serine 134 is phosphorylated by AKT and prevents nuclear translocation of NR3C1) [13]. We did not find increased MYB or BCL2 concentrations, previously proposed as steroid-resistance mechanisms in select xenograft ALL models [14].
Our data suggest that the NR3C1-driven proapoptotic response is reduced by an AKT/MEK-ERK-driven antiapoptotic response resulting in steroid resistance. AKT drives the expression of antiapoptotic MCL1 and BCLXL. In our resistant panel of cell lines, high MEK-ERK signaling was associated with the presence of phosphorylated (and inactivated) GSK3B and BIM, which are essential for steroid-induced death [43–46]. BIM is phosphorylated by ERK [57,58], and, in line with this, BIM was not phosphorylated in steroid-resistant lines upon treatment with the MEK inhibitor CI1040. GSK3B is an important regulator of proapoptotic BIM [48] and may prevent phosphorylation of BIM, thereby preserving steroid responsiveness. In addition to activated MEK-ERK, AKT can phosphorylate and inhibit GSK3B, as observed in the steroid-resistant panel of cell lines. Therefore, GSK3B may play an important role in the regulation of steroid responsiveness in general. Consistent with this, the GSK3B inhibitor IX strongly provoked steroid resistance in most steroid-sensitive cell lines. Our data therefore suggest that any cellular mechanism that may activate MEK-ERK and AKT can lead to steroid resistance in T-ALL, and may explain some steroid-resistant T-ALL cases that lack apparent IL7R signaling or NR3C1 mutations. The recently described NR3C1 cleavage by CASP1 might provide an alternative explanation for steroid resistance [19].
The finding that steroid resistance arises from activated MEK-ERK and AKT suggests that a therapeutic regime combining steroids with MEK, AKT, or mTOR inhibitors may increase steroid responsiveness. Importantly, these inhibitors might also increase the steroid response in steroid-sensitive patients. Interestingly, even though three of our patient samples harbored IL7R mutations, we found no proof of the effectiveness of ruxolitinib to reverse steroid resistance. This may possibly be due to a lack of cellular proliferation during in vitro drug sensitivity testing, and implies that dormant leukemic stem cells are not killed by ruxolitinib. Ruxolitinib has demonstrated some effectiveness in treating ETP-ALL in xenograft models [59,60], although a major effect was not achieved.
Our finding may also be highly relevant to BCP-ALL patients—who often have mutations in IL7R signaling molecules [61,62]—although this remains to be proven. High levels of AKT activity have been associated with steroid resistance and poor outcome in BCP-ALL [63]. Mutations in NRAS and KRAS are common in ALL and are associated with steroid resistance, central nervous system involvement, and poor outcome [16,17]; these mutations are preferentially acquired at relapse [18]. Connectivity mapping based on expression profiles of steroid-resistant ALL [64] suggests that mTOR inhibitors restore steroid sensitivity by reducing the levels of antiapoptotic MCL1 [65,66].
One possible limitation of our study is the relatively small cohort size. However, given the extremely low incidence of T-ALL (approximately 15 new patients identified in the Netherlands each year), it was not feasible to obtain larger numbers of clinically and biologically documented patient samples. A second limitation is that we were unable to obtain data regarding the effectiveness of inhibiting IL7R signaling inhibitors in vivo with respect to restoring steroid sensitivity in primary leukemic cells, e.g., using patient-derived T-ALL xenograft mouse models. In this respect, it is important to note that a recent preclinical study found that the PI3K/mTOR inhibitor NVPBEZ235 enhanced steroid sensitivity in a T-ALL xenograft model [67].
In conclusion, we provide evidence that steroid resistance in pediatric T-ALL patients is associated with mutations in IL7R signaling molecules. Because treatment success and clinical outcome are highly dependent upon steroid sensitivity in this setting, our findings suggest that small-molecule inhibitors of MEK, PI3K, AKT, and/or mTOR should be tested in order to restore or enhance steroid sensitivity in patients with ALL.
|
10.1371/journal.pgen.1003987 | Expanding the Marine Virosphere Using Metagenomics | Viruses infecting prokaryotic cells (phages) are the most abundant entities of the biosphere and contain a largely uncharted wealth of genomic diversity. They play a critical role in the biology of their hosts and in ecosystem functioning at large. The classical approaches studying phages require isolation from a pure culture of the host. Direct sequencing approaches have been hampered by the small amounts of phage DNA present in most natural habitats and the difficulty in applying meta-omic approaches, such as annotation of small reads and assembly. Serendipitously, it has been discovered that cellular metagenomes of highly productive ocean waters (the deep chlorophyll maximum) contain significant amounts of viral DNA derived from cells undergoing the lytic cycle. We have taken advantage of this phenomenon to retrieve metagenomic fosmids containing viral DNA from a Mediterranean deep chlorophyll maximum sample. This method allowed description of complete genomes of 208 new marine phages. The diversity of these genomes was remarkable, contributing 21 genomic groups of tailed bacteriophages of which 10 are completely new. Sequence based methods have allowed host assignment to many of them. These predicted hosts represent a wide variety of important marine prokaryotic microbes like members of SAR11 and SAR116 clades, Cyanobacteria and also the newly described low GC Actinobacteria. A metavirome constructed from the same habitat showed that many of the new phage genomes were abundantly represented. Furthermore, other available metaviromes also indicated that some of the new phages are globally distributed in low to medium latitude ocean waters. The availability of many genomes from the same sample allows a direct approach to viral population genomics confirming the remarkable mosaicism of phage genomes.
| Prokaryotic species contain extremely large gene pools (pan-genome) the study of which has been constrained by the difficulties in getting enough cultivated representatives of most of them. The situation of their viruses, also known as phages, that provide part of this genomic diversity and preserve it, is even worse. Here we have found a way to bypass the limitation imposed by pure culture to retrieve phage genomes. We obtained large insert clones (fosmids) from natural communities that are undergoing active viral attack. This has allowed us to triple the number of genomes of marine phages and could be similarly applied to other habitats, shedding light into the biology of the most numerous and least known biological entities on the planet. They exhibit a remarkable degree of variation at one single geographic site but some seem also to be prevalent worldwide. Their frequent mosaicism indicates a high level of promiscuity that goes beyond the already remarkable hybrid nature of prokaryotic genomes.
| Prokaryotic viruses, often referred to as phages, are one of the largest reservoirs of underexplored genetic diversity on Earth. They are more numerous than any other biological form on the planet, and the astronomical values put forward for their numbers are to the tune of 1030, difficult to comprehend even by metaphoric abstractions [1]. Such estimates have contributed to an increasing appreciation of the role of this poorly charted component in the global carbon and energy cycling in the oceans [2], [3]. The high prevalence of phages in the environment also raises important questions about their local and global population diversity, the dynamics of interaction within themselves and their hosts, and the evolutionary implications of these relationships [4], [5].
A critical bottleneck for the study of phages is the need to obtain their hosts in axenic cultures before they themselves can be cultured. Consequently, as most marine bacteria remain uncultured [6]–[8], so too do their phages. Obtaining genomic DNA for uncultured microbes has been relatively easy, and sequencing of numerous oceanic metagenomes and single cell genomes have provided an extraordinarily detailed view of the real world of marine microbes [9]–[14]. Similar progress, though, has been elusive for marine phages. Even though they are estimated to be 10-fold as numerous as prokaryotic cells, recovering viral DNA in amounts sufficient for sequencing has proven difficult although recently methods have been devised to improve the process [15]–[17]. Phage genomes are much smaller than cellular ones and total phage DNA per volume is relatively low [18] compared to their cellular hosts. As a result, DNA amplification is normally a necessary step before metavirome sequencing, which probably biases the product significantly [19], [20]. Still, with all these caveats, the nascent field of marine metaviromics has provided an insight into the marine viral world [21]–[26]. Therefore, most of our knowledge about complete marine phages' genomes stems from cultured representatives [27]–[30], isolated only because of success in culturing the host, which themselves, in several cases took painstaking years to be adapted for growth in the laboratory e.g. Ca. Pelagibacter [31], [32].
Cloning of environmental DNA into fosmids, used successfully for studying prokaryotic genomic fragments of uncultured microbes [9], [10], has opened an alternative to obtain complete genomes of phages [18], side-stepping completely the previously mandatory availability of a cultivated host. This has been possible due to the observation that inserts cloned in fosmids from metagenomic DNA have a significant representation of phage genomic fragments [9], [10]. Actually, a replicating phage in the course of its natural lytic cycle in a cell, provides a natural amplification that is reminiscent of laboratory cloning or other methods of genome amplification, such as multiple displacement amplification (MDA) [33]. Formerly, metagenomic fosmids have been shown to capture major marine phage lineages like cyanophages [10], [18] and the SAR11 viruses [32].
The deep chlorophyll maximum (DCM) is the site of maximal phototrophic cell density in oligotrophic open ocean waters. It is a seasonal phenomenon in temperate waters forming at the middle of the photic zone during the summer stratification of the water column [34], [35] and a permanent feature in tropical latitudes. Supported by the high number of microbial cells, the number of infecting phages is also expected to be high. We have sequenced and assembled ∼6000 metagenomic fosmids obtained from the Mediterranean DCM (MedDCM) cellular fraction (>0.2 µm). Among them more than a thousand genomic contigs were derived from marine phages that were actively replicating and are described here. Two hundred and eight of them represented novel complete genomes, and some were very different from any phage known to date. Furthermore, the examination of the genomes has allowed assigning putative hosts to many of these previously unknown phages. This collection also provides a unique opportunity to examine concurrent phages from the same natural habitat, en masse. The sequences reveal the existence of multiple, highly related coexisting lineages for each phage type, likely matching or exceeding the multiple prokaryotic lineages of their host genomes [36]. From the same site a metavirome (from the viral size fraction) has also been directly sequenced by Illumina (MedDCM-Vir) to assess the relevance of these phages in the viral sized fraction.
From the sequenced fosmids, we manually selected 1148 virus-like contigs (size range 5–48 Kb, average size 23 Kb, GC% range 27–57) based on their resemblance to known phages and/or presence of key phage genes using the Phage Orthologous Groups [37] (see Materials and Methods). As is typical for viral genomes, nearly half of the proteins from this collection of fosmid contigs (40%) did not present any hits to the NR database, reflecting the novelty of the phage genomes described here. Thirty six percent were similar only to hypothetical proteins. Of the remaining 24% that could be attributed a function, most (19%) were clearly phage-related, 1% were cellular-like (host) proteins and 4% were unclassified. Among host-related genes, we identified auxiliary metabolic genes (AMGs) commonly found in phages such as photosystem related genes (psbA, psbD), 6-phosphogluconate dehydrogenase (gnd), Glucose-6-phosphate 1-dehydrogenase (zwf) and transaldolase (talC) [30], [38], [39].
The presence of a large number of predicted proteins characteristic of tailed phages (e.g. terminase, tape measure protein, tail formation and baseplate related proteins) indicated that 935 contigs clearly originated from the order Caudovirales. For the remaining contigs, in which these specific genes could not be identified, further comparisons suggested that they were also tailed bacteriophages. Given that these contigs could be reliably assigned to head-tail phages and are derived from the cellular fraction (between 5 and 0.2 µm) selective for prokaryotic cells, we have focused this work only on tailed phages and not on other types of viruses (e.g. eukaryotic viruses) that might also be present in the fosmid library. Phylogeny of the essential terminase gene has been used to resolve different phage groups and define new ones [18], [40]. We found a remarkable diversity of this phage packaging gene in our assembled contigs. A phylogenetic analysis showed that these contigs not only recaptured several known lineages (e.g. T4-like or T7-like) but also defined many novel major branches (Figure 1).
We organized the 1148 contigs into sequence identity clusters (see Materials and Methods) to group together genomic fragments of the same or highly related phage lineages (more than 95% nucleotide identity over at least 20% overlap). It seems apparent from the examination of the contigs that the DNA from which they derive are not individual phage genomes but the concatamer that appears as an intermediate stage during the replication of most Caudovirales [41]. This has allowed us to assess genome completeness when one fosmid covered more than one complete genome in the cellular concatamer or two identical clones overlapped (Figure S1). Even though the insert size of fosmid clones (30–40 kb) limits the maximum size, two hundred and eight such complete genome representatives (henceforth referred to as CGRs) could be recovered. We have largely focused on their analyses, although the other contigs have been also used when they could provide additional information. All contigs were named in a way to reflect their origin, completeness, and sequence similarity amongst themselves (see Materials and Methods for details).
To establish the relationships of the novel phage genomes with known phages, we performed a large-scale whole genome comparison with several reference genomes, including all marine phages. A purely genomic approach to classify phages has been proposed before and actually recapitulates several features of traditional phage classification [42]–[44]. Our slightly modified genomic approach similarly agrees well with both methods (Figure S2, S3 and S4). The whole genome comparison of the 208 CGRs shows that while some of them cluster with known phages, there are several instances of completely novel phage groups (Figure 2, Figure S5) as already hinted by the terminase phylogeny (Figure 1). Using the tree obtained, we have organized these CGRs into 21 sequence groups (G1 to G21) (Figure 2, Table 1). Within each group there was also a large degree of variation, showing protein identities typically in the range of 50–70% (see below), in effect akin to different genera of phages, i.e. within each group there was more than one phage genus. As an example, G21 groups together different genera of phages from the marine Bacteroidetes Cellulophaga [45] and Persicivirga [46] (Figure S5).
Another way to classify phages is by the host upon which they prey. Although the identification of hosts of uncultured viruses is non-trivial, phage genomes sometimes display features that divulge the identity of the host. Well known amongst such features are AMGs, metabolic genes that are frequently phage versions of host genes, e.g. photosystem genes carried by cyanophages that help boost phage replication during infection [47]–[50]. Actually, the photosystem genes (psbA and psbD), apart from unequivocally linking a phage to cyanobacteria, have been shown to discriminate not only between phages of different environments (e.g. marine or freshwater), but even different phage types (e.g. podoviruses or myoviruses) [51], [52]. In absence of such signature genes, another tell-tale feature may be simply high sequence identity to phages with known hosts, which is likely only if the phages share a common host species. For some phages, presence of CRISPR spacers in uncultured phage genomes and concordant matches in a host genome may also be used as evidence of a phage-host relationship [53]. In our case this last approach did not help, probably due to the scarcity of CRISPR systems among marine microbial genomes.
A less explored link between phages and their hosts is related to the putative temperate nature of several phages, particularly their integration into host tRNA genes. Integration into a host genome requires that the phage carries an integrase, an excisionase and a repressor [54], [55]. Phages integrating into tRNAs carry a phage attachment site (attP) that is an exact match of a host tRNA gene (bacterial attachment site, attB). For example, the Prochlorococcus phage P-SS2 contains an integrase gene, and an attP site (53 bp), which is an exact match of 36 bp to the host tRNA (attB) of Prochlorococcus MIT9313 [56]. Along these lines, a phage carrying an integrase and a putative attP site identical to a host tRNA gene fragment is highly suggestive of a host-phage relationship. As proof of principle for this method, we used two cyanophage contigs from our collection identified clearly due to presence of photosystem genes (psbA in this case). Both these contigs also carried an integrase gene and a partial tRNA gene. Comparisons to the cyanobacterial genomes of Prochlorococcus and Synechococcus revealed that the tRNA gene fragment in both of these contigs was identical to the tRNA-Leu of Prochlorococcus marinus MED4 (42 bp exact match) and Synechococcus CC9605 (39 bp exact match), linking them to these putative hosts. Phylogenetic analysis of the psbA gene additionally supported this specific prediction (Figure S6 and Figure S7). Another such prediction could be made for a CGR that was >80% identical (in nucleotides) along its entire length to pelagiphage HTVC019P (Figure 3). Such high sequence identity already suggests that this CGR represents a novel pelagiphage. This CGR also carries an integrase gene and a fragment of a tRNA-Leu gene that is identical (46 bp) to the tRNA-Leu gene in Ca. Pelagibacter HTCC7211. It is important to emphasize that, given the high conservation of the tRNA gene among closely related species, the predictions based on this method alone are expected to provide only a broad taxonomic assignment, i.e. the phylum or class (e.g. SAR11 cluster or Verrucomicrobia). However, when supplemented with supporting evidence in the form of characteristic host genes (e.g. psbA for cyanophages), or high nucleotide identity to cultivated phages, it may be possible to be more specific in the predictions.
Using a combination of these approaches, applied to all the phage contigs, we were able to assign putative hosts to 527 contigs (Data S1). Several CGRs could be associated with a known host (Figure 2, Table 1, Data S1) (see below). Many of which are as yet uncultured microbes known only by their genome sequences. They represent a wide variety of important marine microbes like Cyanobacteria (Prochlorococcus and Synechococcus), members of the SAR11 clade (Ca. Pelagibacter and the Alpha proteobacterium HIMB114), SAR116 representatives (Ca. Puniceispirillum), Verrucomicrobia and the recently described low-GC clade of marine Actinobacteria [57]. However, it is important to underscore that host-association of a single CGR in a group in no way implies that the entire group to which it belongs infects the same host. With all these caveats and after comparison of the closest known phages for each group in Figure 2, inferences regarding putative hosts could be made for 64 of the 208 CGRs. Using these host assignments and the genomic properties of the phages we classified them as follows.
Group G2 contains CGRs that appear to be cyanophages, likely infecting Prochlorococcus. They are related (>75% nucleotide identity in several regions) to the known Prochlorococcus phages MED4-117 and MED4-184, both dwarf myoviruses. Groups G7, G8 and G9 were closely related and actually all belong to the subfamily Autographivirinae. They all possess an RNA polymerase that is the hallmark gene of this family, among other characteristic structural and replication genes [43] (Figure 3). All RNA polymerase containing CGRs could be classified in one of these three groups. G7 contains a novel CGR that, from the psbA gene phylogeny (Figure S6) and similarity to Synechococcus phage RIP2 (>75% identity across the genome), likely preys on Synechococcus. Group G8 contained a CGR that could be classified as a new pelagiphage (infecting SAR11) by both sequence similarity (>75% nucleotide identity along the entire genome) to HTVC019P [32] and the integrase/att relationship. CGRs in group G9 are novel phage genomes for which no host assignment was possible.
Of the 31 CGRs in group G15, nine seem to be related to the recently cultured Pelagibacter phage, HTVC010P, which was shown to be the most abundant phage in the oceans [32]. These CGRs shared high nucleotide identities (>80%) in large regions with HTVC010P, suggesting they are also pelagiphages. In particular, two of these CGRs are highly similar along their entire lengths to HTVC010P, effectively making them Mediterranean variants of this phage, which was isolated from Bermuda (Hydrostation S) (Figure 4). Additionally, several of these new phage genomes were linked to the SAR11 cluster by the integrase/att relationship (Table 1, Data S1).
We defined several distinct groups of phages for which there are no known related genomes available. However, it was still possible to predict hosts for several CGRs in these novel phage groups. For example, one of the seven CGRs in group G11 could be linked to Verrucomicrobia using evidence from integrase/att identity to the single-cell amplified genome (SAG) SCGC AAA300-K03 [14] recently described as belonging to this phylum. The GC content of this phage genome (43.8%) also matches very well the cellular genome GC content (42.3%). To our knowledge this is the first report of a marine Verrucomicrobia phage.
G17 and G19 contained CGRs that were putative pelagiphages unlike any others known before. There is evidence for them infecting SAR11 cluster microbes from both integrase/att relationship and small regions of high nucleotide identity with HTVC010P. In addition, some of the CGRs in G19 were nearly fully syntenic to a prophage locus in the genome of the SAR11 alpha proteobacterium HIMB114, albeit at a protein sequence identity in the range of 40–50% (Figure S8).
Along the same lines, several of the CGRs from the group G16 (Table 1, Figure 2) could prey upon SAR116. The first SAR116 phage (HMO-2011) has been recently described [58]. However, these CGRs are unrelated to HMO-2011, which is related to group G12 instead (Figure 2). Only the integrase/att relationship connected these CGRs of group G16 to Candidatus Puniceispirillum marinum [59] and other uncultured SAR116 representatives [14].
One phage genomic fragment (not a CGR) could be putatively assigned as an actinobacterial phage, the most likely host being Ca. Actinomarina minuta, the smallest free-living microbial cells described so far [57]. The fragment carries an integrase and also a 43 bp att site that is 100% identical to tRNA-Val of the putative host genome. This match is so specific that the att site sequence only retrieves Ca. Actinomarina minuta sequences from the complete GenBank collection. In addition, a WhiB transcriptional regulator found only in Actinobacteria, was also found in this phage fragment. This gene has been found previously in mycobacterial phages (e.g. TM4), where it has been shown to have a growth inhibitory and a super-exclusion effect [60]. This phage genomic fragment appeared most closely related by sequence (Figure 2) to the G13 group of CGRs to which no other host could be assigned.
An essential question is how relevant are the phages represented by our CGRs in a DCM phage population. To this end, a different DCM sample from the same location (and retrieved four years later) has been processed to generate a metavirome (MedDCM-Vir). The DNA from the viral fraction in the sample was amplified by MDA and sequenced by Illumina to provide nearly 18 Gb of sequence data. We used this metavirome, along with several others [21], [22], and some representative metagenomes, to assess relative abundance of known marine phages (133 reference genomes) and the CGRs. The most abundantly recruiting genomes are shown in Figure 5 and Figure S9. As expected, recruitment from metagenomes is much less than from metaviromes, reasserting the viral nature of CGRs. Among the top recruiting genomes there is a large representation of the CGRs, with only a few cultivated Ca. Pelagibacter and Prochlorococcus phages reaching comparable values. Although most CGRs recruited more in their habitat of origin (MedDCM-Vir) they also recruited very well in other datasets, such as the Sargasso Sea. Reciprocally, several phages isolated from the Sargasso (e.g. P-SSP2, P-GSP1 and P-SSP7), not only recruited a high number of reads from the Sargasso Sea metavirome, but also from the MedDCM-Vir.
How much of the viral diversity at the DCM was recovered in our fosmids? To answer this question, we have used very relaxed criteria for recruitment (BLASTN, minimum alignment length 50 bp and e-value 0.01). As a control, we used multiple genomes of several abundant DCM microbes (Prochlorococcus, Synechococcus, Ca. Pelagibacter, SAR86, Group II Euryarchaeota, adding up to a total of 75 Mb sequence data) to recruit reads from the metavirome MedDCM-Vir. Only 0.14% of reads could be matched indicating a negligible contamination with cellular DNA. On the other hand, the 1148 phage contigs described here recruited about 1.54% of all the reads of the MedDCM-Vir, an order of magnitude more, but still suggesting they represent a small minority in the Mediterranean DCM virome. The 133 reference genomes recruited even less (only about 0.36%). It has been recently suggested, using microscopic techniques, that natural marine viral populations may be dominated (up to 92%) by non-tailed phages [61], providing a potential explanation for such low recruitment levels.
It is important to underscore here that the metaviromes are always amplified by MDA. There is evidence that MDA acts much more efficiently with single stranded DNA so that extant metaviromes could be over representing ssDNA viruses [19], [20] and are consequently biased against dsDNA genomes. However, even the recently described 608 genomes of marine, circular ssDNA viruses [26] recruited only 1.5% of the MedDCM-Vir reads. Such results are not restricted to the Mediterranean DCM metavirome, as they recruited similarly low levels from the Sargasso metavirome (0.89%) [21], and nearly nothing from the Pacific Ocean Virome [22]. Therefore, it appears that the vast majority of the marine virome sequence space is as yet unsampled.
With this large collection of complete phage genomes from the same place and time, it becomes possible to examine concurrent diversity, and patterns of variability, that have traditionally been analyzed by repeated and independent phage isolation in culture. Firstly, we have found several examples of nearly identical phage genomes. Using very restrictive similarity criteria (95% nucleotide identity over 95% overlap) we identified 519 contigs (out of the total of 1148) that clustered in groups of 2 to 22 members. Analysis of these highly similar clusters revealed several examples of nearly identical phage genomes with minor differences only, clearly showing that there are numerous recently diverged concurrent variants. For example, cluster C12B (Figure S10) contains nine contigs that were >98% identical over the overlapping regions. In this comparison, some contigs are nearly identical with only minor indels, such as contigs 6 and 7 in Figure S10. However, other contigs/regions diverged much more (similarity down to 75–80%, for example contigs 7 and 8). This is reminiscent of the flexible genomic islands of the prokaryotic genomes [62] and had previously being shown for cultivated phages. In a number of published studies [27], [53], tail proteins and other host recognition structures have been described as highly variable. This phenomenon was attributed to diversity of host recognition specificity among different phage lineages. However, the variations that we have found in the closely related genomes, although including structural host recognition features, do not appear to be restricted to any specific functional role. For example, internal virion proteins, terminases and capsid proteins were all observed within variable regions.
Another frequent pattern is the presence of a hybrid architecture in which large divergent regions are present together with regions of high identity. An example is shown in Figure S11. Such genomes clearly belong to phages infecting a common host that have exchanged genomic fragments during a mixed infection. Moreover, given the identical nature of several of these regions, it does appear that these exchanges are recent events. Similar results have been obtained by comparing cultured phage isolates. For example, the study of several isolated staphylococcal phages strongly suggested the exchange of large segments of genomes among them [63]. Whether or not such recombinations are facilitated by the presence of linker regions [64] or are random rearrangements followed by selection for function has not been established. The sheer amount of phage infections occurring in the marine habitat at any given time [1], [2] makes it likely that any of these events are feasible.
Given the high sequence identities found between the Mediterranean pelagiphages and the first pelagiphage HTVC010P isolated from the Sargasso Sea (Figure 4), we searched for more such examples in our contig collection. Identical phage sequences have been found before in geographically distant marine samples, but these were based on small genomic fragments (200–600 bp) [65], [66]. We found two cyanophage contigs from our collection that were >97% identical along their entire lengths to cyanophages isolated as far as the Pacific Ocean (Figure S12). Both of these contigs were nearly 40 Kb long (nearly complete fosmids) and originate from myoviruses that are >170 Kb long. These are remarkable examples of global distribution of viruses that suggest a rapid global phage circulation, likely along with oceanic currents.
It has been clear for some time that culture, although instrumental in the development of Microbiology, cannot provide an adequate representation of the real diversity of prokaryotic microbes and their phages in a sensible timeframe. New technologies based on high-throughput sequencing and direct nucleic acid retrieval from communities or single cells provide critical short-cuts for advancing in the discovery of the cellular microbes. However, an equivalent short-cut for the phage sequence space has been missing. Phage isolation and sequencing is very important in studying the natural diversity of phage populations, yet it is tightly constrained by the burden of obtaining host cultures. These limitations are not only relevant for the study of biodiversity alone. The population genomics of prokaryotic microbes and their phages, i.e. their evolution and microdiversity, suffer from similar handicaps. Here we have provided the largest collection of concurrent phage genomes ever described for any habitat so far by using metagenomic fosmids. This opens a route towards phage population genomics that can be based on complete genomes, rather than small genomic fragments. It appears to be the simplest and most effective high-throughput method to obtain complete phage genomes from a natural habitat yet. In addition, we have been able to assign putative hosts to many by using sequence based criteria that appear quite reliable and could prove instrumental as the field of metaviromics evolves further.
On the other hand, it is quite obvious that we have only retrieved a small fraction of the full diversity of phages living in the habitat of choice (the Mediterranean DCM). First of all our method is limited by the size of fosmid clones so that large viral genomes could not be retrieved. We could have tried to use overlapping fosmids but they would probably lead to unreliable chimeric assemblies. The genuine examples of mosaicism detected here, make artifactual assemblies from metaviromes a possibility that needs to be considered. For now we decided to focus mostly on the bona fide complete genomes (CGRs). One possible way to bypass the size limitation would be to use larger insert vectors such as BACs [67], or apply long-read sequencing directly to the samples when it is available [68]. Another obvious limitation of our method is that only replicating viruses, and apparently, those using the concatamer mode of replication, have been captured. Although it appears to narrow the window of the kinds of phages detected, it provides a confirmation of their active role in the ecology of the environment. It is possible that some phage particles are just remnants of past lytic events [1], which are not relevant for the current habitat ecosystem functioning despite their presence in the metavirome. Those are excluded in our methodology. Finally, both single stranded DNA or RNA phages [69] are obviously omitted from detection by our technique. As mentioned before, traditional metaviromes might also be highly biased [19], [20], [70], and in this sense both methods might be complementary. In spite of all these caveats, we nearly tripled the number of marine phage genomes.
Given the recovery of nearly identical genomic fragments across the globe, it is already evident that there is very weak (if any) phage biogeography in temperate and tropical latitudes. Therefore, an in depth study of a single location can contribute enormously to our knowledge of phage biodiversity. Furthermore, coming from a single sample, we have shed light into the dynamics of genome change in concurrent phages. Using phage contigs highly related to the globally distributed pelagiphage HTVC010P, remarkable sequence conservation and variation patterns were discernible. There are aspects that are reminiscent of similar phenomena in prokaryotes [71]–[73], such as the flexible genomic islands, i.e. the presence of several concurrent lineages that differ only in small genomic regions. Some of these regions are probably involved in host specificity at the level of clonal lineages [5], [27]. However, some unexpected genes were subjected to high microdiversity (capsid protein and terminases), the reasons for which are for the moment obscure. Capsid proteins could be involved in host recognition but it is not likely that the terminase has any connection with such specificity. In addition, swapping of genome fragments amongst phage lineages appears to be a central theme in phage evolution. Overall there seems to be more creativity in concurrent, highly identical (over 95%), phage genomes compared to cellular genomes, that sometimes involves the replacement of large genomic segments, likely by recombination with distant lineages that share the same host. Similar phenomena had been detected before in cultivated phages [63]. Not being strictly fitness constrained as the cellular compartment, phages might embark onto more adventurous evolutionary trajectories. Actually, there is little doubt that phages may represent a significant part of the prokaryotic pan-genome [74] that could outsource risky, but highly innovative, evolutionary paths to their accompanying phage populations.
The availability of large numbers of closely related genomes and the discernible patterns in their diversity and distribution increases our appreciation towards the enormous variety that exists, much of which was only partially captured before by isolated phage genomes. Importantly, it opens up a view of the phage world where instead of observing phage genomes as discrete entities, we can begin to look upon them as vast, constantly churning global continuums.
The sample from which the fosmid library was constructed was taken on October 15, 2007 from the DCM (50 m depth) off the coast of Alicante, Spain (38°4′6.64″N 0°13′55.18″W) with a Niskin bottle. The sample was filtered through 5 µm polycarbonate and 0.22 µm Sterivex filters. DNA from 0.22 µm filters was used to create a fosmid library of ∼13000 clones. A 454 metagenome from the same filter, and results of sequencing of ∼1000 fosmids have been described previously [10]. For this work, DNA from ∼6000 metagenomic fosmids was extracted and pooled in 24 batches, with ∼250 fosmids in each batch. These were sequenced using Illumina PE 300 bp reads in a single lane (∼175× coverage for each fosmid). Each batch was assembled independently using Velvet [75] (k = 51). Several criteria were employed to identify phage genomic fragments, for example, multiple hits to all known phages, presence of key phage genes using Phage Orthologous Groups [37], availability of multiple related fragments, and manual examination of each contig. POGs are clusters of orthologous genes from bacteriophages that can be used to identify viral genes and a virus quotient (VQ) quantifies the phage specificity of each gene (the closer it is to 1, more phage specific the gene is). The VQ profile of the POGs of selected MedDCM contigs was very similar to the one obtained for the known phage genomes (majority of the POGs with VQ equal to 1), suggesting that those contigs indeed represent true phage genome fragments. A total of 1148 (lengths ranging from 5–48 kb) contigs were finally selected for the final analysis. The presence of the vector sequence (ranging to 16–67 bp) on both sides of 139 assembled contigs indicated that these contigs represented the complete fosmid sequence. The lengths of the majority of the complete fosmids were between 30–40 kb. Genes were predicted using prodigal [76], and annotated using BLAST against the NR database, Pfam [77], COGs [78], TIGRfams and POGs [37]. All complete genome representatives were manually examined and annotated using the HHpred server [79]. All contigs were named according to the nomenclature described below.
Seawater (20 L) collected from the DCM of the Mediterranean Sea (65 m deep) on August 29th, 2011, was filtered through a 0.2 µm filter (Millipore GVWP2932A). Subsequently, phages were concentrated using tangential flow filtration (TFF) with a 30 kD polyethersulfone membrane from Vivaflow (VF20P2). The resulting phage concentrate was ultracentrifuged (Optima XL 1000K Ultracentrifuge, Beckman) for 1 h at 4°C using a Type 70 Ti rotor (Beckman) at 30,000 rpm (92,600 g). The pellet was resuspended in 1 mL of the seawater supernatant and treated with 2.5 units DNase I at 37°C for 1 hr, and 70°C for 10 min to remove bacterial DNA. The phages were then lysed in 0.50 mg/mL Proteinase K and 1.0% SDS at 56°C for 1 h followed by two rounds of phenol/chloroform/isoamyl alcohol extraction. The aqueous phase was then chloroform/isoamyl alcohol extracted and ethanol precipitated and resuspended in sterile water. DNA quantity and quality was determined using gel electrophoresis and Picogreen. Multiple amplification displacement (Illustra GenomiPhi V2 DNA Amplification Kit, GE Healthcare) was performed using ca. 30 ng of DNA for each of five reactions. The resulting DNA (ca. 5 µg) was sequenced in one third of an Illumina lane, yielding approximately 18 Gb of sequenced data (paired end reads, 300 bp insert size) with a total of ∼180 million reads.
An all-vs-all comparison, using BLASTN [80] was performed for all contigs. Only >95% identical hits and with lengths >50 bp were retained. Overlapping hits, if any, were merged together using the mergeBed program in the BEDtools package [81]. The total length of these hits was then used to compute percentage coverage of the contig length. All pairs of contigs selected satisfying the coverage criteria (of 20% in the first round of clustering and 95% in the second round) were visualized in Cytoscape as a connected network [82]. Groups of connected contigs in these networks were considered as valid clusters. The 1148 contigs were clustered first using a criterion of >20% coverage but with very high nucleotide sequence identity (>95%). 117 clusters (containing 914 contigs) were obtained, and 236 contigs remained unclustered. In the next step, the contigs in each cluster were clustered at an even stricter criterion of at >95% coverage and >95% nucleotide identity to identify nearly identical contigs. Further examination of the 102 subclusters obtained after this second step, allowed us to identify 208 complete phage genomes indicated by the circular-like organization of two or more contigs of a cluster. Similarly, end redundancy in contigs that were unclustered was used to identify complete genome representatives.
As described above, an all-vs-all nucleotide comparison was used first to cluster all viral contigs using cut-off of 95% sequence identity and 20% coverage. Contig clusters formed in this step were named given a cluster number (e.g. C1, C2 etc). Unclustered contigs were tagged with a “U”, for “unclustered”. In the second round of clustering, we used the same sequence identity (95%) but a higher cut-off to coverage (95%) to identify the most highly related and syntenic contigs within each cluster. At this stage, if multiple clusters were obtained within a single cluster (say C1), they were tagged alphabetically, e.g C1A, C1B, C1C etc. Contigs within a cluster (C1), but not part of any further subclusters were not tagged again. All clusters (both clusters and subclusters) were examined manually for completeness. For example, if a complete genome representative (CGR) was identifiable in subcluster C1A, it was tagged as a CGR-C1A. If a CGR was identifiable in a cluster, it was tagged as CGR-C1. If a CGR was found in a cluster, all other contigs that were not identified as CGRs, were tagged as CGF (complete genome fragment). The naming scheme is described in detail below.
Following the suggestions made for the nomenclature of viruses, we have used the following procedure for the nomenclature of uncultured viruses described in this work: (1) uv - uncultured virus (2) MED - three letter abbreviation in capitals indicating origin of the sample (3) CGR/CGF/GF - field indicating if the contig refers to a complete genome, a fragment of a complete genome, or just a genomic fragment. CGR (complete genome representative) is a contig that is assumed to be a complete phage genome. There may be more than one CGR in a cluster. CGF (complete genome fragment) is a contig that cannot be inferred to represent a complete phage genome, but is part of a cluster that contains a CGR). GF (genomic fragment) is similar to a CGF but without any CGR. Such GF contigs can have an extra name field (see below) indicating they are unclustered (U). (4) C1/C1A/C1B/U - indicates the clustering status of the contig. (5) Field containing the contig identifier, e.g. MedDCM-OCT-S14-C437. An example of a complete identifier is uvMED-CGR-C1-MedDCM-OCT-S17-C19, enabling quick identification of several key features of a phage genome/contig.
Several well-classified reference phage genomes, identified using the ICTV classification (http://www.ictvonline.org) were downloaded from NCBI. In addition, all known marine phage genomes were included in the comparison. Each genome was compared to another using TBLASTX [80] using the BLOSUM45 matrix. A hit was considered significant if it had >30% sequence identity, a minimum length of 30 aa and an e-value of at least 0.01. The bit score of all such selected hits in a comparison was summed up to give a comparison score for a pair of genomes. Closely related genomes get higher comparison scores. To normalize for different genome sizes each phage genome was also compared to itself to obtain a self-score. The Dice coefficient, which is a similarity metric ranging from 0 to 1, was computed as follows, Dice = (2*AB)/(AA+BB), where AB is the comparison score of phage A with phage B, AA and BB are the comparison scores of phages A and B with themselves respectively. This metric was transformed to a dissimilarity metric by subtracting it from one. A neighbor joining tree was constructed from the complete distance matrix using the PHYLIP package [83].
Separate initial comparisons were run for well classified podoviruses, myoviruses and siphoviruses (classification obtained from http://www.ictvonline.org) to examine the validity of the approach (See Figure S2, Figure S3 and Figure S4). The tree shown in Figure 2 was created using a comparison of all reference phages and with the complete genome representatives (208 CGRs) identified in this study. In the comparison of all tailed phages to each other (Figure 2), several well described phage groups are separable, e.g. Autographivirinae, Tevenvirinae, Spounavirinae etc.
For the terminase tree, Pfam domains, COGs, POGs, TIGRfams were searched using hmmsearch program in the HMMER3 package [84] (evalue 1e-5), in addition to NCBI BLAST [80] to identify large subunit terminase sequences in the entire dataset. 401 unique sequences were identified in the contigs. In addition, 125 reference sequences and 105 terminase sequences from marine phages were included. A total of 631 terminase sequences were used for the alignment. For the phylogenetic trees of photosystem genes psbA and psbD, protein sequences were extracted from the annotated metagenomic fosmids and compared to NCBI NR database using BLASTP to recover additional sequences. Several previously described sequences were also used. All alignments were created using Muscle [85], manually inspected and trimmed as necessary, and maximum likelihood trees were constructed using the program FastTree2 [86] using a JTT+CAT model and an estimation of the gamma parameter. Bootstrapping was performed using the Seqboot program in the PHYLIP package [83].
We used both metagenomes (MedDCM [10], Global Ocean Sampling [11]) and metavirome datasets from the Sargasso Sea, British Columbia coastal waters, Gulf of Mexico, Arctic Ocean [21] and the Pacific Ocean [22]. In addition, the metavirome (MedDCM-Vir) constructed in this study was also used. For depicting comparative recruitment across metaviromes and metagenomes (as shown in Figure 5), a hit was considered if it was at least 50 bp long, had an e-value of less than 1e-5 and more than 95% identity. The number of hits to each phage contig was divided by the length of the contig (in kb) and also by the size of the database (number of reads recruited per kb of contig/size of the database in Gb), which provides a normalized measure to compare recruitments by differently sized contigs versus several metagenomes. This measure is abbreviated as RPKG (Reads per Kb per Gb).
All 1148 contigs assembled in this study have been submitted to DDBJ and are available using the accession numbers AP013358-AP014505. The metavirome has been deposited in NCBI SRA with the Bioproject number PRJNA210529.
|
10.1371/journal.ppat.1006778 | Prohibitin plays a critical role in Enterovirus 71 neuropathogenesis | A close relative of poliovirus, enterovirus 71 (EV71) is regarded as an important neurotropic virus of serious public health concern. EV71 causes Hand, Foot and Mouth Disease and has been associated with neurological complications in young children. Our limited understanding of the mechanisms involved in its neuropathogenesis has hampered the development of effective therapeutic options. Here, using a two-dimensional proteomics approach combined with mass spectrometry, we have identified a unique panel of host proteins that were differentially and dynamically modulated during EV71 infection of motor-neuron NSC-34 cells, which are found at the neuromuscular junctions where EV71 is believed to enter the central nervous system. Meta-analysis with previously published proteomics studies in neuroblastoma or muscle cell lines revealed minimal overlapping which suggests unique host-pathogen interactions in NSC-34 cells. Among the candidate proteins, we focused our attention on prohibitin (PHB), a protein that is involved in multiple cellular functions and the target of anti-cancer drug Rocaglamide (Roc-A). We demonstrated that cell surface-expressed PHB is involved in EV71 entry into neuronal cells specifically, while membrane-bound mitochondrial PHB associates with the virus replication complex and facilitates viral replication. Furthermore, Roc-A treatment of EV71-infected neuronal cells reduced significantly virus yields. However, the inhibitory effect of Roc-A on PHB in NSC-34 cells was not through blocking the CRAF/MEK/ERK pathway as previously reported. Instead, Roc-A treated NSC-34 cells had lower mitochondria-associated PHB and lower ATP levels that correlated with impaired mitochondria integrity. In vivo, EV71-infected mice treated with Roc-A survived longer than the vehicle-treated animals and had significantly lower virus loads in their spinal cord and brain, whereas virus titers in their limb muscles were comparable to controls. Together, this study uncovers PHB as the first host factor that is specifically involved in EV71 neuropathogenesis and a potential drug target to limit neurological complications.
| A close relative of poliovirus, Enterovirus 71 (EV71) causes Hand, Foot and Mouth Disease and has been associated with neurological complications in young children. The lack of effective therapeutics is largely due to our limited understanding of the mechanisms involved in EV71 neuropathogenesis. Here, using a proteomics approach, we have identified a unique panel of host proteins that were modulated during EV71 infection of motor-neuron cells, which are found at the neuromuscular junctions where EV71 enters the central nervous system. Among the candidate proteins, we focused our attention on prohibitin (PHB), a protein that is involved in multiple cellular pathways and the target of anti-cancer drug Rocaglamide (Roc-A). We demonstrated that cell surface-expressed PHB is involved in EV71 entry into neuronal cells specifically, while mitochondria-associated PHB is required for viral replication. Furthermore, Roc-A treatment of EV71-infected neuronal cells led to significantly lower virus yields. In vivo, EV71-infected mice treated with Roc-A survived longer than the vehicle-treated animals and had significantly lower virus titers in their spinal cord and brain. Together, our work uncovers PHB as the first host factor that is specifically involved in EV71 neuropathogenesis and a possible drug target to limit neurological complications.
| Enterovirus 71 (EV71) is a non-enveloped, positive-sense, single-stranded RNA virus, and causes hand, foot and mouth disease (HFMD) in humans. Being a close relative of poliovirus, EV71 is deemed as an important neurotropic virus worldwide [1]. Since its first isolation in California in 1969, several major outbreaks have been reported in China, Singapore, Korea, and Japan [2–5]. Although the clinical manifestations are generally mild and self-limiting, including HFMD and herpangina, severe neurological complications have been consistently reported with EV71-associated infections, causing brainstem encephalitis, acute flaccid paralysis, pulmonary edema and cardiopulmonary failure [6,7]. In addition, some patients who have recovered from severe disease have been reported to develop long term neurologic and psychiatric disorders [8]. There are currently no effective prophylactic or therapeutic agents against EV71. Although several vaccines have completed Phase III clinical trials [9], regulatory issues may limit their widespread utilization. In addition, as these vaccine candidates consist of inactivated virus from a single EV71 genotype (C4), cross-protection against other genotypes may be limited [10,11].
The increasing awareness of life-threatening EV71 infections has boosted research in recent years to further understand virus-host interactions and develop effective antiviral strategies [12–17]. However, the neuropathogenesis of EV71 is still poorly understood. Infection occurs when the virus enters the body upon ingestion and/or inhalation. The virus multiplies initially in the alimentary tract mucosa and rapidly reaches the deep cervical and mesenteric lymph nodes via the tonsils and Peyer’s patches [18]. After a short transient systemic dissemination phase, the virus accumulates and actively replicates in muscles where it is believed to infect motor neurons at the neuromuscular junctions. Experimental evidence supports that EV71 migrates to the brainstem via retrograde axonal transport as previously described for its close relative poliovirus [1,2,19–21]. However, the molecular mechanisms involved in EV71 infection of motor neurons to access the central nervous system (CNS) have not been studied. Indeed in vitro studies aiming at studying EV71 neurovirulence have employed neuroblastoma cell lines that may not reflect accurately infection in motor neurons. To address this gap, we have recently reported a novel in vitro model of EV71 infection in the murine motor neuron cell line NSC-34 [22]. NSC-34 cells originate from the fusion between murine neuroblastoma and spinal cord cells, and possess motor neuron-like properties, such as generation of action potentials and production of acetylcholine [23], therefore making it a relevant model to study the mechanism of EV71 neuropathogenesis. We demonstrated that NSC-34 cells are permissive to EV71 clinical isolates and found that, unlike any other mammalian cell types so far reported, EV71-infected NSC-34 cells do not undergo apoptosis and lysis. Instead we showed that the virus exits the cells via a non-lytic mode, a phenomenon that has also been previously described for poliovirus [21,24,25]. These unique features thus suggested that the infection cycle of EV71 in NSC-34 cells involves host pathways and partners that are likely to be different from those previously identified in other mammalian cell types such as muscle cells and neuroblastoma cells.
In this work, using a proteomics approach coupled with mass spectrometry, we have identified a panel of cellular proteins that were dynamically regulated during EV71 infection of NSC-34 cells. Among the host protein candidates that were up-regulated, we focused our attention on prohibitin (PHB) and characterized its role during EV71 infection in NSC-34 cells. We also demonstrated the importance of PHB during EV71 infection in a symptomatic mouse model of EV71 infection.
To identify the host proteins involved in EV71 infection cycle in NSC-34 cells, a 2DE proteomic approach was undertaken. NSC-34 cells were infected with EV71 at M.O.I. 10, and the cell lysates were harvested at 6, 24, 48 and 72 hours for downstream proteomic analysis in which a range of 350–800 spots were resolved.
By using PDQuest 2-D Analysis Software (BioRad), a total of 81 protein spots (Fig 1a) that displayed at least 0.5-fold differential expression (p<0.05, two-tailed Student’s t-test) compared to uninfected controls, were excised for in-gel digestion and MALDI-TOF MS analysis. The peptide fingerprints were then searched against NCBInr mouse genome database for protein identification using MASCOT program (http://www.matrixscience.com/). The protein candidates were then categorized based on their primary functional class indicated in UniprotKB database (S1 Table).
To illustrate the dynamic regulation of host proteins during the viral infection, a heat map was generated using MultiExperiment Viewer (MeV), with the distance between proteins represented by Euclidean average linkage clustering (Fig 1b). This clustering analysis revealed that proteins that were up-regulated (Fig 1c) during infection are mainly involved in motility (23%) and catalytic processes (20%), while proteins that participate in RNA processing (32%) and energy biosynthesis (20%) generally displayed a down-regulation trend during the course of infection (Fig 1d).
Functional interactions among the selected host proteins were analyzed by STRING (Search Tool for the Retrieval of Interacting Genes/Proteins). This platform allows establish protein-protein interactions based on published literature, online databases, predicted functional associations using genomic information or observations made with other organisms [26]. The protein network obtained was significantly enriched with the p value of less than 0.05, suggesting that the interactions are highly associated and unbiased (Fig 2; S2 Table). Furthermore, some of the selected host proteins appear to have strong associations among each other as indicated by the thickness of connecting lines which reflects the confidence level of the interactions [26].
Using GO annotations for biological processes, molecular functions, cellular compartments and protein classes, the protein candidates were localized within the cytoplasm (29.2%), organelles (20.8%) and macromolecular complexes (13.9%) (S1a Fig). In addition, they were found to be involved in various biological processes including mitochondrial biogenesis, proteolytic activity, cytoskeletal machinery and RNA processing (S1a Fig), consistent with the protein clusters observed in the STRING network (Fig 2). Finally, molecular function analysis indicates that majority of these proteins contribute to nucleic acid binding transcription factor activity (34.7%), structural molecule activity (25%) or binding (22.2%) (S1a Fig).
A meta-analysis with other selected proteomic studies of EV71-infected muscle and neuronal cells [13–15,17,27–29] revealed minimal overlap between NSC-34, RD and other neuronal cell types with 4 protein candidates only, namely ACTB, TUBB, PDIA3 and ENO1, suggesting that the host-pathogen interactions in NSC-34 cells are unique (S1b Fig). The limited overlap may also be partly explained by differential proteomics approaches. ACT and TUBB are involved in maintaining cytoskeletal structure, and they are found highly modulated during viral infection to facilitate virus internalization and transportation [30–32]. ENO1 has been shown previously to interact with cytoskeletal proteins in intermediate filaments framework rearrangement [33]. On the other hand, PDI functions in catalyzing reduction and oxidation processes and protein folding [34]. It has also been demonstrated to be involved in humoral immune response [35] or viral replication (DENV) [36] and entry (HIV) [37]. Not surprisingly, greater overlap was seen between motor-neuron NSC-34 and other neuronal cells (26 shared hits) than between NSC-34 and RD cells (5 shared hits). Importantly, neuro-specific proteins such as PRPH and UCHL1 were only identified from profiling studies in NSC-34 and other neuronal cells, thus validating our 2DE proteomic approach.
Seven protein candidates namely, alpha-enolase (ENO1), DEP domain-containing mTOR-interacting protein (DEPTOR), peripherin (PRPH), phosphatidylethanolamine-binding protein 1 (PEBP1), prohibitin (PHB), stomatin-like protein 2 (STOML2) and protein disulfide-isomerase A3 (PDIA3) were selected for validation of the proteomic findings by gene knockdown. These host proteins have been previously shown to be associated with various steps in the life cycle of viruses, such as entry [37–40] and replication [15,41,42], or to be involved in autophagy [43–45] and axonal transport [46–48]. Silencing of each selected gene target was achieved by reverse transfecting the On-TARGETplus siRNA SMARTpool into NSC-34 cells, prior to viral infection. siRNA SMARTpools consist of four highly potent gene-specific siRNA molecules which have been modified to minimize off-target activity and enhance gene specificity [49].
Cytotoxicity of the siRNAs SMARTpools was first established. Apart from the siRNA pool targeting PDIA3, no significant cytotoxicity was observed with the other siRNA pools at 25 and 50 nM with cell viabilities greater than the 70% viability threshold (Fig 3a). The PDIA3-specific siRNA pool concentrations were lowered to 5 and 10nM to avoid cytotoxicity (Fig 3a). Virus titers in the culture supernatants of siRNA-transfected cells were then determined at 48 h.p.i. Results indicated that silencing of STOML2, PRPH, PHB and DEPTOR led to significantly lower virus titers, whereas PDIA-, PEBP1- and ENO1-knocked down resulted in increased viral titers in the supernatants of EV71-infected NSC-34 cells compared to control (Fig 3b). Importantly, both trends were dose-dependent. Therefore, these results validate the 2D-proteomics approach as a powerful way to identify host proteins that play a role during EV71 infection in NSC-34 cells.
Prohibitins belong to a highly conserved protein family present in unicellular and multicellular eukaryotes [50]. Prohibitin (PHB; BAP-32) and prohibitin 2 (PHB2, REA, BAP-37) are two highly homologous members of this family and are ubiquitously expressed in multiple cellular compartments including the mitochondria, nucleus, and the plasma membrane. Prohibitins have been involved in multiple cellular functions including cell proliferation and maintenance of the functional integrity of the mitochondria [50]. In addition, PHB specifically has been previously reported to be involved in the entry step of alphavirus chikungunya (CHIKV) [40], and flaviviruses Dengue (DENV) [38] and Hepatitis C (HCV) [39], and to interact with envelope proteins from white spot syndrome virus to prevent infection [51]. However, there has been no report so far on the role of PHB during EV71 infection. First, modulation of PHB expression during EV71 infection in NSC-34 cells was confirmed by western blot and showed an overall up-regulation of PHB during the course of infection compared to uninfected control (S2 Fig). Next, the impact of PHB gene silencing on virus production was further analyzed using a wider range of siRNA pool concentrations including 5, 10, 25 and 50 nM. Efficacy of the gene silencing was assessed by Western blot and showed a dose-dependent decrease of PHB expression in NSC-34 cell lysates (Fig 3c). This dose-dependent PHB knockdown correlated well with a dose-dependent reduction in the viral titers measured in the culture supernatant (Fig 3d), therefore supporting the role for PHB in EV71 infection cycle. To address the possibility of false positive or off-target effects of the siRNA pool, PHB gene silencing was performed with the individual siRNAs species from the pool. Western blot showed that each individual siRNA was capable of silencing PHB expression significantly (S3a Fig) which correlated with reduced viral titers measured in the culture supernatants of EV71-infected NSC-34 cells (S3c Fig) with minimal cytotoxicity (S3b Fig). Finally, to further examine the role of PHB during EV71 infection cycle, PHB was over-expressed in NSC-34 cells. Western blot confirmed the upregulation of PHB by 2.6 fold compared to controls (S3d Fig). A significantly higher viral titer was observed in the culture supernatant of PHB over-expressing cells compared to controls (S3e Fig), thus further demonstrating the involvement of PHB in virus production in NSC-34 cells.
To determine if PHB mediates entry of EV71 into NSC-34 cells, a competition assay was performed using commercially available anti-PHB antibody. Incubation of NSC-34 cells with anti-PHB antibody prior to infection led to significant reduction of the viral titer in a dose-dependent manner (Fig 4a). To demonstrate a physical interaction between cell surface-expressed PHB and EV71, a proximity ligation assay (PLA) was performed. In this assay, PHB and EV71 are recognized by specific primary antibodies raised in two different species, which are in turn recognized by species-specific secondary antibodies conjugated to a probe and a target, respectively. Should EV71 and PHB be in close proximity, the probe and the target ligate, amplify and result in emission of a fluorescent signal. SCARB-2 which has been previously demonstrated as the main receptor for EV71 in RD cells [52] was used as positive control and a red fluorescent signal was readily detected in EV71-infected RD cells (Fig 4b). A positive signal was also detected with NSC-34 cells incubated with anti-PHB and anti-EV71 antibodies thus supporting the close proximity between EV71 virus particles and surface-expressed PHB (Fig 4b). In contrast, and expectedly, no signal was detected with mouse SCARB-2 (mSCARB2) and EV71 antibodies (Fig 4b), since we have shown previously that entry of EV71 into NSC-34 cells is not mediated by mSCARB-2 [22].
The physical interaction between EV71 and cell surface-expressed PHB was further assessed by performing a co-immunoprecipitation experiment. NSC-34 cells were incubated with EV71 for 2 hours at 4°C to allow viral adsorption onto the cell surface but no internalization. The total cell lysate was obtained and a pulldown was carried out using antibody specific to PHB or an isotype IgG antibody control. The immunoprecipitates were then analyzed by Western blot using anti-EV71 primary antibody. A discrete band at the expected size was obtained when pulldown was performed with the anti-PHB antibody whereas no band was seen when pulldown was done with the IgG isotype (Fig 4c). These findings thus support that EV71 physically interacts with cell surface-expressed PHB, and suggest that PHB may serve as a receptor for EV71 entry into NSC-34 cells.
In addition to its association to the plasma membrane at the cell surface, PHB is also present intracellularly [50]. To study the role of intracellular PHB in EV71 infection cycle, PHB-knocked down NSC-34 cells were transfected with EV71 RNA genome and the viral titers were determined at 6, 12, 18 and 24 hours post-transfection, in order to assess virus production within a single cell infection cycle. Transfection of viral genome was meant to bypass the virus entry step which we have shown involves cell surface-expressed PHB. No viral titer was obtained at 6 h.p.t. in the PHB-knocked down and control cells (Fig 5a). From 12 h.p.t onwards the viral titers detected in the culture supernatant from PHB-knocked down cells were consistently lower than those measured in siNTC or non-treated cells (Fig 5a). This result thus demonstrates that PHB plays a role in the intracellular virus infection cycle.
To further confirm this hypothesis, a luciferase EV71 (lucEV71) replicon transfection assay was performed. In this replicon, the viral structural genes have been replaced with a luciferase-encoding gene while the other parts of the viral genome are retained (Fig 5b). Upon transfection, the replicon undergoes a single replication cycle with no production of virus progeny since it is deficient in viral structural proteins. The luminescence measured from the cell culture is proportional to the amount of luciferase produced inside the cell thereby reflecting the replication activity of the replicon. Here, a significant reduction in the luminescence signal was observed in PHB-knocked down NSC-34 cells compared to siNTC-treated and non-treated cells (Fig 5c).
Thus, together the data support that intracellular PHB is involved in EV71 viral replication.
To further investigate the role of intracellular PHB during EV71 infection cycle, immunostaining was performed on EV71-infected NSC-34 cells probing for PHB, EV71 capsid proteins VP0/VP1 and the viral replication intermediate dsRNA. Co-localization between PHB and dsRNA, and between PHB and EV71 capsid proteins was readily observed (Fig 6a), supporting that intracellular PHB could be involved in the viral replication and/or assembly processes.
To further study the role of intracellular PHB in viral replication, co-immunoprecipitation was carried out using antibody against PHB. Pull down with anti-PHB antibody followed by Western blot using anti-EV71 3D/3CD antibody led to the detection of a 72kD band that corresponds to the EV71 3CD protein complex and a 53 kDa band (3D polymerase) which co-migrated with IgG heavy chain (Fig 6b). Taken together, these data support a physical interaction between intracellular PHB and EV71 non-structural proteins 3D and 3CD, indicating that intracellular PHB is likely involved in viral replication.
Previous studies have shown that the main replication sites of picornavirus are located at the Golgi apparatus and endoplasmic reticulum (ER) [53]. We have demonstrated that intracellular PHB co-localizes with dsRNA and is closely associated with the EV71 3D polymerase. Given that intracellular PHB is abundantly and mainly expressed on mitochondria [54], we speculated that in NSC-34 cells mitochondria may be exploited by EV71 as replication site. Consistently, co-localization of PHB and EV71 with mitochondria was observed by IFA (Fig 7a and 7b). Furthermore, co-localization of PHB, dsRNA and mitochondria was also readily detected, thus indicating that mitochondrial PHB is associated with the viral replication complex (Fig 7c). To exclude the possibility that the replication complexes detected were actually associated to the ER, which is in close proximity to mitochondria, the mitochondrial fraction was prepared from EV71-infected NSC-34 cells and Western blot analysis revealed the presence of viral capsid protein VP1 (38 kD), 3D (53 kD) and 3CD (72 kD) proteins (Fig 7d). Furthermore, the mitochondrial fraction was shown to be free of cytoplasmic contamination, as evidenced by the presence of mitochondrial marker (ATPB, 52 kD) and lack of ER marker (Calreticulin, 48 kD). Similar observation was made with the mitochondrial fraction prepared from EV71-infected RD cells (Fig 7d), suggesting that EV71 is able to exploit various cellular organelles as replication scaffolds in various mammalian cell types from different species. This finding is consistent with a previous study where EV71 VP1 was found to be associated with mitochondria in Hela cells [55]. To further support the close proximity and likely interactions between the viral replication complexes and mitochondria, transmission electron microscopy was performed on EV71-infected NSC-34 cells. Clustering of mitochondria surrounding viral replication complexes could be seen (electron-dense like structures) in the infected cells, and examination at a higher magnification indicated a close association between mitochondria membrane and virus complex/virus particle (Fig 7e).
Collectively, the data strongly indicate that mitochondria in NSC-34 cells are exploited by EV71 as a replication scaffold and that mitochondria-associated PHB plays a role in this process.
Recent studies have shown that PHB activity is inhibited by a group of phytochemicals called rocaglamides, which are derived from the traditional Chinese medicinal plants Aglaia [39, 56–58]. We therefore investigated whether Roc-A could interfere with EV71 infection cycle by blocking PHB activity in NSC-34 cells. Incubation of Roc-A with virus prior to NSC-34 cell infection (co-treatment) did not result in any significant reduction in viral titer (S4a Fig). When cells were pre-treated with Roc-A prior to EV71 infection (pre-treatment), less than 1 log PFU/mL of decrease in viral titer was observed at the highest drug concentration only (500 nM) (S4b Fig). In contrast, a dose-dependent decrease in the viral titer was seen when Roc-A treatment was applied after infection (post-treatment) at concentrations ranging between 10–100 nM (Fig 8a). Western blot analysis of the cell lysates further confirmed the dose-dependent reduction of intracellular viral capsid protein and PHB (Fig 8b). Taken together, the data suggest that the antiviral effect of Roc-A on EV71-infected NSC-34 cells targets the viral replication step and not the entry step, in contrast to a previous study with HCV [39].
Prior studies focusing on cancer have reported that the mechanism by which Roc-A targets and inhibits PHB activity involves blocking the CRAF/MEK/ERK pathway [56,57], and this was also described in the HCV study [39]. To investigate whether the CRAF/MEK/ERK signaling pathway was inhibited in Roc-A treated NSC-34 cells, Western blot analysis was performed. The results showed that in Roc-A-treated NSC-34 cells only CRAF phosphorylation was impaired while the activation of MEK remained at basal level (S5 Fig). In addition, ERK phosphorylation was not observed under any conditions (S5 Fig). These findings thus indicate that the antiviral effect of Roc-A in NSC-34 cells is likely independent on the CRAF/MEK/ERK signaling pathway.
To decipher the mode of action of Roc-A in NSC-34 cells, immunostaining of Roc-A treated NSC-34 cells was performed. Decreased signals for PHB and mitochondria were observed with increasing concentrations of Roc-A (Fig 8c), suggesting that Roc-A might affect mitochondrial integrity. We thus assessed the mitotoxicity and cytotoxicity of Roc-A using the Mitochondrial ToxGlo assay (Promega). While cytotoxicity remained generally minimal over the range of Roc-A concentrations tested, the intracellular ATP levels were significantly reduced in a dose-dependent manner, thus indicating functional impairment of the mitochondria in Roc-A-treated NSC-34 cells (Fig 8d). Consistently, using the membrane-permeant JC-1 dye as an indicator of mitochondrial membrane potential, Roc-A-treated NSC-34 cells displayed marked and dose-dependent mitochondrial depolarization as evidenced by the decrease of red fluorescent J-aggregates, compared to untreated cells (S6 Fig). Together, these observations suggest that Roc-A treatment in NSC-34 cells results in reduced levels of PHB, leading to mitochondrial destabilization and lower ATP production. One could thus speculate that the lack of intact mitochondria and reduced intracellular ATP levels might eventually impact negatively on EV71 replication efficacy.
The role of PHB in EV71 infection cycle was also studied in human muscle (RD) and neuronal (SK-N-SH) cell lines. As human (GI246483) and murine (GI6679299) PHB display high similarity in their amino acid composition, most of the anti-PHB antibodies commercially available demonstrate good cross reactivity with cell lines of both species. We first showed by flow cytometry comparable levels of surface expression of PHB on RD, SK-N-SH and NSC-34 cells (S7 Fig). However, both PHB gene silencing and PHB receptor blocking experiments performed in human muscle RD cells did not impact the viral titer (S8a and S8b Fig). On the contrary, PHB silencing in the human neuroblastoma cells SK-N-SH led to a significant dose-dependent reduction in viral titer in the culture supernatant (Fig 9a). In addition, reduced virus titers were observed with SK-N-SH cells pre-treated with anti-PHB antibodies, thus supporting that cell surface-expressed PHB is involved in EV71 entry into this human neuroblastoma cell line (Fig 9b). To investigate if intracellular PHB is also involved in viral replication in SK-N-SH cells, transfection of lucEV71 replicon into PHB-silenced SK-N-SH cells was performed. Results showed a significant reduction in the luminescence signal compared to controls (Fig 9c). Finally, the effectiveness of Roc-A treatment in EV71-infected SK-N-SH cells was also assessed. Similar to our observations with NSC-34 cells, a significant dose-dependent decline in viral titers was observed (Fig 9d).
Taken together, these findings thus strongly indicate the specific involvement of PHB in both viral entry and replication of EV71 in neuronal cells from both human and murine origins.
The role of PHB was further investigated in vivo, using the mouse model of EV71 infection that we established previously where 2-week old AG129 mice (deficient in Type I&II IFN pathways) infected with EV71 display progressive limb paralysis and spatio-temporal virus accumulation in the limb muscles, spinal cord and brainstem [59]. Here, immunohistochemical analysis showed that PHB was readily detected in the limb muscles, brainstem and spinal cord at day 4 p.i. (Fig 10a). Furthermore, some co-localization with EV71 was observed (Fig 10a).
Next, the in vivo anti-EV71 efficacy of Roc-A was assessed by treating therapeutically EV71-infected mice with Roc-A at day 1 and 3 p.i. The development of clinical manifestations was clearly delayed in the Roc-A-treated mouse group which resulted in increased survival time compared to the untreated or vehicle-treated control groups (Fig 10b). In addition, the viral loads in limb muscles, spinal cord and brain in both Roc-A-treated and vehicle-treated mice were determined. Comparable viral loads were detected in the limb muscles from both groups (Fig 10c). In contrast, viral titers in the spinal cord and brain from the Roc-A-treated animals were significantly lower compared to the vehicle-treated group (Fig 10c), thus supporting that Roc-A treatment specifically impairs EV71 neuropathogenesis. These findings correlate well with our in vitro data showing that the role of PHB during EV71 infection cycle is specific to neuronal cells.
Together, the in vivo data support that PHB plays a critical role in EV71 neurovirulence, and that Roc-A represents a potential therapeutic strategy to limit EV71 neuropathogenesis, thereby minimizing neurological manifestations and complications.
The re-emergence of neurotropic enteroviruses in recent years has motivated investigations into EV71 transmission in the neuronal system. Understanding the interplay between virus and host proteins is likely to result in the identification of potential novel drug targets and development of novel antiviral strategies.
Here, using a proteomics approach, we have identified a panel of host factors that displayed dynamic regulation during the course of EV71 infection in the motor neuron NSC-34 cells. The host protein candidates are mainly involved in cytoskeletal structure maintenance, RNA processing and mitochondrial biogenesis. By employing a siRNA gene silencing approach, we have shown that some of these host factors either facilitate or limit EV71 productive infection. Among these host factors, PHB was found to exert a pro-viral effect as evidenced by the reduced viral titers measured in the culture supernatant of NSC-34 cells when the expression of PHB was down-regulated, and by an increased viral titer in PHB over-expressing cells. PHB is mainly localized on plasma membrane, mitochondria and nucleus, and has been involved in multiple signaling pathways regulated by growth factors, immune response, mitochondrial biogenesis, cell migration, proliferation and survival [50,58]. Interestingly, knockdown in NSC-34 cells of STOML2, which was shown to interact with PHB and participate to mitochondria biogenesis [60], resulted in significant reduction of viral titer, similar to that seen with PHB-knocked down cells. This further supports the involvement of mitochondrial proteins during EV71 infection cycle in NSC-34 cells.
In addition, previous studies have reported the association of PHB with internalization of several viruses, including HCV [39], CHIKV [40], DENV [38], and coronavirus (SARS-CoV) [61]. Furthermore, PHB was shown to promote HIV replication by interacting with the HIV-1 glycoprotein [62].
Using various experimental approaches, we have demonstrated that cell surface-expressed PHB is physically associated with EV71 and is involved in the entry of the virus into NSC-34 cells. On the other hand, by employing a lucEV71 replicon, we have shown that intracellular (mitochondrial) PHB plays a role in EV71 replication activity. This was further supported by the observation that mitochondrial PHB co-localizes with the replicating viral genome (dsRNA) and the non-structural proteins 3D polymerase and 3CD complex. Co-immunoprecipitation and TEM approaches also confirmed the physical proximity and likely interactions between viral complexes/viral particles and mitochondria. These data thus led us to propose that mitochondria could serve as replication site for EV71 in NSC-34 cells. The association of EV71 with mitochondria was reported in a previous study where it was proposed that EV71 could potentially interact with some mitochondrial signaling proteins to evade host anti-viral innate immunity [55].
Previous studies have reported that membrane-bound PHB binds to RAS in a GTP-dependent manner, which in turn activates CRAF kinase and eventually triggers the MAPK pathway [58]. Similar to other flavaglines, Rocaglamide (Roc-A) is a natural product that displays insecticidal, anti-fungal, anti-inflammatory and anti-cancer activities [63]. Mechanistically, Roc-A was found to inhibit CRAF-PHB interactions in tumor cells [64–66], and in an in vitro model of HCV infection [39]. In EV71-infected NSC-34 cells, incubation with nM concentrations of Roc-A resulted in a dose-dependent reduction in virus titers. However, the mechanism by which Roc-A exerts its antiviral effect against EV71 in NSC-34 cells does not seem to be mediated by blocking the CRAF/MEK/ERK pathway, given that phosphorylated ERK proteins could not be detected in uninfected NSC-34 cells, suggesting that this pathway is not functional in these cells. Instead, we found that Roc-A-treated cells displayed reduced expression of mitochondrial PHB and lower levels of intracellular ATP, which suggests that mitochondria integrity/functionality is impaired in Roc-A treated NSC-34 cells. Since we showed that mitochondrial PHB is involved in EV71 replication and that mitochondria serve as replication site for this virus, the impact of Roc-A on PHB expression and mitochondria integrity could represent the basis of its antiviral activity. Further investigation is necessary to decipher the molecular mechanisms by which Roc-A affects the expression of mitochondrial PHB.
Interestingly, we found that the role of PHB in EV71 entry and replication was limited to cells of neuronal origin, thus supporting a role of PHB specifically in EV71 neuropathogenesis. This neuro-specific phenotype was also observed in vivo where Roc-A-treatment resulted in reduced virus loads in the CNS (spinal cord and brain) only but not in the limb muscles from infected mice, although PHB was readily detected in the muscle cells as well. The cell type-dependent involvement of PHB during EV71 infection likely reflects differential intracellular events with different host factors being engaged during EV71 intracellular life cycle. Since EV71 is known to be able to use multiple receptors to enter host cells, one could speculate that the host factors that are engaged during EV71 infection depend on the entry receptor that is being used by the virus. Further study is necessary to explore this idea.
In conclusion, our work has uncovered a novel host factor that is specifically involved in EV71 neurovirulence. In addition, our data support that Roc-A, a previously established anti-cancer drug that targets PHB, could represent a therapeutic approach to limit EV71 neuropathogenesis, and thus prevent or limit associated neurological complications. Given the current attrition in effective antiviral drugs against EV71, Roc-A repurposing is worth considering seriously.
All the animal experiments were carried out under the guidelines of the National Advisory Committee for Laboratory Animal Research (NACLAR) in the AAALAC-accredited NUS animal facilities. The animal experiments described in this work were approved under the NUS Institutional Animal Care and Use Committee (IACUC) protocol number 16–0136. Non-terminal procedures were performed under anesthesia, and all efforts were made to minimize suffering.
Murine motor neuron NSC-34 cells (CELLutions Biosystems, CLU140), human rhabdomyosarcoma (RD) cells (ATCC CCL-136) and human neuroblastoma SK-N-SH cells (ATCC HTB-11) were used in this study. All cell lines were cultured in Dulbecco’s Modified Eagle’s medium (DMEM) (Gibco) containing 10% fetal bovine serum (FBS) (Gibco) at 37°C with 5% CO2. Non-mouse-adapted EV71 S41 (5865/SIN/00009, Accession No.: AF316321), kindly provided by Prof. Chow V. T. K. at National University of Singapore, was isolated from the lymph node of a EV71-infected patient who died of encephalitis and pulmonary edema in Singapore [67]. The virus stocks were made in RD cells and the viral titers were determined by plaque assay on RD cells.
Generally, viral infection was performed at a multiplicity of infection (M.O.I.) of 10 (NSC-34 cells) or 1 (SK-N-SH cells) for 48 hours, prior to downstream viral titer determination or cell lysate harvesting. RD cells (M.O.I. 1) were infected with EV71 for 12 hours before the culture supernatant or cell lysate was collected for further processing.
NSC-34 cells (107 cells/flask) were seeded onto T-75 culture flask overnight prior to infection at MO.I. 10. After 1 hour incubation, unbound viruses were removed by washes and fresh DMEM with 2% FBS was added to the cells. At 6, 24, 48 and 72 hour post-infection (h.p.i.), the culture medium was removed and cells were washed twice with wash buffer (ProteoExtract Complete Mammalian Proteome Extraction Kit, Millipore). The cells were then gently scrapped off using cell scrapper in 1 mL of wash buffer and spun down at 150 ×g for 10 minutes at 4°C. The cell pellet was stored at -80°C until further processing.
Total proteins extract from infected cells was prepared using ProteoExtract Complete Mammalian Proteome Extraction Kit (Millipore). Briefly, the cell pellet was thawed by resuspension in ice-cold Resuspension Buffer and the proteins were extracted with Extraction Buffer at room temperature (RT). Benzonase and reducing agent were added during protein extraction to minimize nucleic acid contamination and to remove disulphide bonds, respectively. The solubilized protein suspension was subjected to centrifugation at 25,000 ×g for 30 minutes at 4°C to remove the remaining insoluble material. The recovered cell extract was stored at -20°C until further analysis.
The protein samples (200 μg, quantified using RC DC Bradford assay, BioRad) were loaded onto individual lanes of the isoelectronic focusing (IEF) tray with pre-wetted electrode wicks. Passive rehydration was performed for each protein sample using 11cm pH4-7 ReadyStrip IPG strips (BioRad) for 12 hours at RT with gel side down configuration. After rehydration, the protein sample was subjected to IEF on Protean IEF Cell i11 (BioRad) according to the following conditions: 250 V for 20 minutes with linear ramp, 8,000 V for 2.5 hours with linear ramp and 8,000 V for 30,000 V-hours with rapid ramping. IPG strips equilibration was achieved by incubating the strips with pre-warmed DTT Equilibration Buffer I (BioRad) followed by iodoacetamide-supplemented Equilibration Buffer II (BioRad) for 10 minutes each on orbital shaker. Equilibrated IPG strips were then transferred onto 12.5% Tris-HCl Criterion gel (BioRad) and overlaid with ReadyPrep Overlay Agarose (BioRad). Electrophoresis was run at 200 V for 65 minutes.
After electrophoresis, the gels were stained with InstantBlue (Expedeon) for 1 hour and submerged in MilliQ water overnight to remove background signal. Gels were scanned using GS-800 Calibrated Densitometer (BioRad). Gel images were further processed using PDQuest 2-D Analysis Software (BioRad), whereby the different gel images from three independent experiments were matched and the intensities of detected spots were measured. Protein spots that showed at least 0.5-fold change in spot intensity (p<0.05, two-tailed Student’s t-test), compared to the uninfected control sample, were excised for MALDI-TOF MS. The fold change was calculated using the equation: (MeanofSpotExpressioninInfectedSamplesMeanofSpotExpressioninUninfectedSamples) for each time point.
In-gel digestion and MALDI-TOF MS of the excised protein spots were done by Protein and Proteomics Centre, National University of Singapore (Singapore). The data was search against the murine and viruses National Centre for Biotechnology Information non-redundant (NCBInr) database using a MASCOT program (http://www.matrixscience.com). No threshold was applied to the MS/MS fragment ions intensities.
Data mining of the identified proteins was done by searching in PANTHER (http://www.pantherdb.org/) and Swiss-Prot/TrEMBL (http://www.uniprot.org/) databases. The enrichment analysis of protein-protein interactions was performed using STRING network analysis version 10 (http://string-db.org/). Hierarchical clustering and classification were performed using MultiExperiment Viewer version 4.9 (http://mev.tm4.org/#/welcome).
On-TARGET plus siRNA SMARTpools targeting genes encoding for prohibitin (PHB), peripherin (PRPH), phosphatidylethanolamine binding protein 1 (PEBP1), enolase-1 (ENO1), stomatin-like protein 2 (STOML2), protein disulfide isomerase family A member 3 (PDIA3), DEP domain containing MTOR-interacting protein (DEPTOR) and non-targeting siRNA control (NTC) were purchased from Dharmacon (GE Life Sciences). The siRNA SMARTpool sequences are shown in S3 Table. Briefly, various concentrations of siRNAs were prepared using DharmaFECT Cell Culture Reagent in a total volume of 50 μL (DCCR) (Dharmacon, GE Life Sciences) and incubated for 5 minutes at RT. DharmaFECT 1 Transfection Reagent (1 μL) (Dharmacon, GE Life Sciences) was then added to the siRNA mixture and topped up to final volume of 100 μL with DCCR. After 30 minutes incubation with transfection reagent at RT, NSC-34 cells (2.5 × 105 cells/ 400 μL) were seeded onto 24 wells plate and reverse transfected with the siRNA constructs. At 48 hour post-transfection (h.p.t.), cellular viability was assessed using alamarBlue cytotoxicity assay (Invitrogen) and the cells were subjected to EV71 infection at M.O.I. 10. Culture supernatant and cell lysate were harvested at 48 h.p.i. for viral titer determination and Western blot analysis. Gene silencing of PHB was also performed in RD and SK-N-SH cells following the same procedure.
RD cells (105 cells/well) were seeded onto 24 wells plate. Culture supernatant from EV71-infected samples was serially diluted (10-fold) with DMEM containing 2% FBS prior to infection. The cell monolayer was incubated with 100 μL of the diluted viral suspension for 1 hour at 37°C. The cells were then washed twice with PBS and replaced with 1 mL DMEM containing 2% FBS and 1% carboxymethyl cellulose (CMC, Sigma Aldrich). After 3 days incubation at 37°C, the infected monolayers were fixed and stained with 4% paraformaldehyde/ 0.1% crystal violet solution (Sigma Aldrich). The number of plaques was scored visually and viral titers were expressed as plaque-forming units (PFU) per milliliter (PFU/mL).
Drug treated or siRNA-transfected cells (2.5×104 NSC-34 or 5×104 SK-N-SH cells/well) were washed twice with PBS and 1× alamarBlue reagent (Invitrogen) diluted with DMEM containing 2% FBS was added. After 4 hours incubation at 37°C, the fluorescence signals were determined using microplate reader (Infinite 2000, Tecan) at Ex570nm and Em585nm. Percentage of viable cells was calculated using non-treated cells as control.
NSC-34 and SK-N-SH cells (2×105 cells/well) were seeded onto 24-well plate and incubated at 37°C overnight. The cells were pre-treated with anti-PHB antibody (PA5-27329, Invitrogen) at 5, 15 and 30 μg/mL in DMEM and 2% FBS for 1 hour at 37°C. The cell monolayer was then washed twice with PBS and infected with EV71 at M.O.I. 10 (NSC-34) or 1 (SK-N-SH) for 1 hour. Culture supernatant was harvested at 48 h.p.i. for viral titer determination.
EV71 viral genome was extracted from infected cell culture supernatant using QIAamp Viral RNA Mini Kit (Qiagen), according to the manufacturer’s instructions. Viral RNA was diluted to 0.25 μg with OptiMEM (Invitrogen) in a total volume of 50 μL and incubated for 5 minutes at RT. After incubation, 1 μL of Lipofectamine 2000 (Invitrogen) was added to the RNA mixture and topped up to 100 μL, followed by 30 minutes incubation at RT. NSC-34 cell suspension (105 cells/ 400 μL) was mixed with 100 μL of transfection mixture and added into each well. At 48 h.p.t. the cells were directly transfected with 50 nM PHB siRNA constructs. The culture supernatant was harvested at different time points for viral titer determination.
LucEV71 replicon or pCMV6-PHB (Origene)-harbouring E. coli was grown in LB broth supplemented with kanamycin (50 μg/mL). Both plasmids were extracted using QIAprep Spin Miniprep Kit (Qiagen). LucEV71 was linearized with Mlu1 restriction enzyme (R0198S, New England Biolabs). The linearized plasmid was then purified using chloroform/phenol/isoamyl (CPI 24:25:1) and chloroform (Sigma Aldrich). Purified linear plasmid (1 μg) was then subjected to in vitro transcription using MEGAscript T7 Transcription Kit (AM1334, ThermoScientific), according to the manufacturer’s instructions. The RNA product was further cleaned-up with chloroform/phenol/isoamyl (CPI 24:25:1) and chloroform. NSC-34 and SK-N-SH cells (2.5×104 cells) seeded in a 96-well plate were reverse-transfected with 50nM siPHB, siNTC or left untreated for 48 hours. At 48 h.p.t. 1 μg of lucEV71 RNA was added into each well and incubated for 48 hours. The luminescence signal was then captured using Nano-Glo Luciferase Assay System (Promega, N1103). Signal was normalized against non-transfected cells. The pCMV6-PHB plasmid (0.5 μg) was transfected into NSC-34 cells for 48 hours prior to viral infection. The culture supernatant was harvested at 48 h.p.i. for determination of viral titer by plaque assay.
NSC-34 and RD cells (105 cells) were seeded onto a coverslip, prior to cell surface viral adsorption at M.O.I. 20 (NSC-34 cells) and 5 (RD cells) for 2 hours at 4°C. The cells were then fixed with 4% PFA for 30 mins at RT. The cells were incubated first with 2% (w/v) BSA in PBS for 1 hour at 37°C and probed with primary antibodies (S4 Table) for 1 hour at 37°C. Next, the cells were incubated with two PLA probes (DUO92101, Sigma Aldrich) for 1 hour at 37°C, prior to ligation for 30 minutes. Signal amplification of the PLA probes was achieved by incubating the cells with polymerase for 100 minutes at 37°C. The coverslips were then mounted on glass slides with Duolink In Situ Mounting Medium with DAPI and viewed under Olympus IX81 fluorescence microscope.
NSC-34 cells (2×107 cells/flask) were seeded onto T-175 culture flask. For study of the interaction between EV71 and surface-expressed PHB, the cells were incubated with EV71 at M.O.I. 30 for 2 hours at 4°C before they were lysed using iced cold lysis buffer (ThermoScientific). For study of the interaction between intracellular PHB and EV71, infected protein extracts (M.O.I. 10) were prepared using iced cold lysis buffer at 48 h.p.i. After conjugating Dynabeads Protein G (Life Technology) with anti-PHB antibody or IgG isotype control (S5 Table) for 20 mins at RT, the protein complexes were pulled down and subjected to Western blot analysis (S6 Table). Briefly, the protein extracts were incubated with the conjugated magnetic beads and further incubated for 3 hour at 4°C with constant rotating, before eluting using Laemmli buffer. Isotype antibody pull-down and uninfected cells were used as controls.
NSC-34 (107 cells/flask) and RD (107 cells/flask) cells were seeded onto T-75 culture flask overnight, prior to EV71 infection at M.O.I. 10 and 1, respectively. At 48 h.p.i. (NSC-34) and 12 h.p.i. (RD), the cells were lysed and the mitochondria-enriched fraction was obtained using Mitochondria Isolation Kit for Cultured Cells (89874, ThermoScientific). Both cell fraction and mitochondrial fraction were stored separately for Western blot analysis using relevant antibodies (S4 Table).
NSC-34 (2×105 cells/well) and SK-N-SH (105 cells/well) cells were seeded onto 24-well plate and incubated overnight. For pre-treatment condition, the cells were incubated with Roc-A (SML0656, Sigma Aldrich) at various concentrations for 3 hours prior to EV71 infection at M.O.I. 10. For post-treatment condition, the cells were infected with EV71 at M.O.I. 1 (NSC-34 cells) or 1 (SK-N-SH cells) for 1 hour and then treated with Roc-A for 48 hours. For co-treatment, cells were incubated with EV71 and Roc-A simultaneously for 1 hour prior washing and replaced with fresh 2% DMEM. Culture supernatant was harvested at 48 h.p.i. for viral titer determination.
NSC-34 (5×104 cells/well) cells were seeded onto 96-well white opaque plate and incubated overnight. The cells were treated with various concentrations of Roc-A (diluted with 2% DMEM) for 48 hours. Cytotoxicity and mitotoxicity of Roc-A were then assessed using Mitochondrial ToxGlo Assay (G8000, Promega). Briefly, 20 μL of 5× Cytotoxicity reagent were added into each well and incubated at 37°C for 30 minutes. Cytotoxicity was measured using fluorescence at Ex485nm and Em525nm. After equilibrating the assay plate to RT for 10 minutes, 100 μL of ATP detection reagent were added into each well and mitotoxicity was measured by luminescence. All readings were normalized against non-treated cells. Sodium azide (S2002, Sigma Aldrich) and staurosporine (S6942, Sigma Aldrich) were used as mitochondrial toxin and cytotoxin positive controls, respectively.
NSC-34 (105 cells/well) cells were seeded onto 8-wells chamber slides (Ibidi) and incubated overnight. The cells were treated with various concentrations of Roc-A (diluted with 2% DMEM) for 48 hours. JC-1 dye (Invitrogen) (10 μg/mL in 2% DMEM) was added into each well and incubated for another 15 mins, prior to imaging. NSC-34 cells incubated with sodium azide NaN3 at 1 μM in 2% DMEM (S2002, Sigma Aldrich) was used as positive control.
Cell lysate was prepared using M-PER Mammalian Protein Extraction Reagent containing 1% of Halt Protease Inhibitor Cocktail and 1% of 0.5 M EDTA (ThermoScientific). Protein quantification was performed using Quick Start Bradford Protein Assay (BioRad). Denatured proteins (5 μg) were resolved in 10% SDS-PAGE gel and transferred electrophoretically onto nitrocellulose membrane using Trans-Blot Turbo Transfer System (BioRad). After blocking with 5% (w/v) milk in TBST (TBS buffer with 0.01% Tween 20) for 1 hour at RT, the membrane was probed using specific primary antibodies and relevant secondary antibodies (S6 Table). The chemiluminescence signal was visualized using Clarity Western ECL Substrate (BioRad) on X-ray films. Densitometric quantification was performed using ImageJ and the relative band intensity was normalized against β-actin.
NSC-34 and RD cells (105 cells) were seeded onto coverslips, incubated overnight and infected with EV71 at M.O.I. 10 and 1, respectively. At 48 h.p.i. (NSC-34) and 12 h.p.i. (RD), the cells were fixed with iced cold methanol for 15 minutes at -20°C. The cells were then washed extensively with PBS to remove residual methanol and blocked with 5% (w/v) Bovine Serum Albumin (BSA) (Sigma Aldrich) in PBS for 1 hour at 37°C, prior to immunostaining with relevant antibodies (S7 Table). The nucleus was revealed using NucBlue Live ReadyProbes Reagent (Molecular Probes). Fluorescence images were captured using Olympus IX81 microscope and further processed using ImageJ.
Limb muscles, spinal cord and brainstem from EV71-infected AG129 mice were harvested at day 4 p.i. after systemic perfusion. The organs were fixed in 4% PFA overnight prior immersing in 15% and 30% sucrose solution, and embedded in Tissue-Tek OCT (VWR) solution. The organ samples were then frozen at -80°, sectioned (10 μm) using a cryostat (Leica) and mounted on a glass slide prior to blocking and staining as described above.
Samples were fixed for 1h at RT with 2.5% glutaraldehyde containing 1% Tannic acid in 0.1M cacodylate buffer (pH 7.2), then washed three times for 5 min (each time) in 0.1M cacodylate buffer and post-fixed for 1h at RT with 1% osmium tetroxide in the same buffer. Samples were then dehydrated in a graded series of ethanol and embedded in Spurr. Thin sections were stained with 2% uranyl acetate and lead citrate and observations were performed by transmission electron microscopy using a FEI Tecnai Spirit G2 at 100Kv. Image were taken using a FEI Eagle 4K CCD camera.
NSC-34, SK-N-SH (8 × 106 cells) and RD (4 × 106 cells) cells were seeded onto 6-well plate overnight. The cells were dislodged by incubating them in 10 mM EDTA/PBS in 4°C for 10 mins and spun down to collect the cell pellet. Cells were then blocked with either human or murine Fc-blocker (564200 or 553142, BD Pharmigen; 1:400) for 30 min at RT, and subsequently stained with anti-PHB antibody (AB75766, Abcam; 1:200) and/or anti-rabbit AF488 antibody (A27034, ThermoScientific; 1:500) for 30 min at 4°C. Stained cells were then fixed with 4% PFA. Flow cytometry analysis was carried out using the Becton-Dickinson Fortessa flow cytometer and analysed using FlowJo v10.
Two-week old AG129 mice (B&K Universal) were bred and housed under specific pathogen-free conditions in individual ventilated cages. Infection was performed by injecting intraperitoneally (i.p.) 107 PFU of S41 (in 100 μL) per mouse. At day 1 and 3 post-infection (p.i.), mice were injected i.p. with 0.25 mg/kg of Roc-A (in 0.25% DMSO in sterile olive oil). The control group was inoculated with 0.25% DMSO in olive oil. Clinical manifestations were observed for a period of 20 days. Clinical score was graded as follows: 0, healthy; 1, ruffled hair and hunched back; 2, limb weakness; 3, one limb paralysis; 4, two limbs paralysis; and 5, death. Two limbs paralysis was used as criterion for early euthanasia. For virus titer determination, infected mice were euthanized at day 4 p.i. and perfused with 50 mL of sterile PBS systemically. The fore and hind limb muscles, spinal cord and brain were harvested and weighed before mechanical homogenization in 1 mL of serum-free DMEM. The homogenates were spun down at 10,000 rpm for 10 minutes at 4°C, and clarified using a 0.22 μm filter before serial dilution was carried out for plaque assay. Viral titers were expressed as PFU per gram of tissue.
|
10.1371/journal.pgen.1000348 | Heterochromatic Threads Connect Oscillating Chromosomes during Prometaphase I in Drosophila Oocytes | In Drosophila oocytes achiasmate homologs are faithfully segregated to opposite poles at meiosis I via a process referred to as achiasmate homologous segregation. We observed that achiasmate homologs display dynamic movements on the meiotic spindle during mid-prometaphase. An analysis of living prometaphase oocytes revealed both the rejoining of achiasmate X chromosomes initially located on opposite half-spindles and the separation toward opposite poles of two X chromosomes that were initially located on the same half spindle. When the two achiasmate X chromosomes were positioned on opposite halves of the spindle their kinetochores appeared to display proper co-orientation. However, when both Xs were located on the same half spindle their kinetochores appeared to be oriented in the same direction. Thus, the prometaphase movement of achiasmate chromosomes is a congression-like process in which the two homologs undergo both separation and rejoining events that result in the either loss or establishment of proper kinetochore co-orientation. During this period of dynamic chromosome movement, the achiasmate homologs were connected by heterochromatic threads that can span large distances relative to the length of the developing spindle. Additionally, the passenger complex proteins Incenp and Aurora B appeared to localize to these heterochromatic threads. We propose that these threads assist in the rejoining of homologs and the congression of the migrating achiasmate homologs back to the main chromosomal mass prior to metaphase arrest.
| Proper chromosome segregation is essential during the production of eggs and sperm. Chromosome missegregation during meiosis results in the lethality of the offspring or in children carrying extra copies of a given chromosome (for example, Down syndrome). Recombination results in homologous chromosomes becoming physically interlocked in a manner that is normally sufficient to ensure proper segregation. Chromosomes that fail to undergo recombination require additional mechanisms to ensure their proper segregation. In Drosophila melanogaster oocytes we show that chromosomes that fail to recombine undergo dynamic movements on the meiotic spindle prior to their proper segregation. Although previous studies had shown that non-recombinant chromosomes move to opposite sides of the developing meiotic spindle, we show that these chromosomes can cross the spindle and re-associate with their homologs to attempt reorientation. Additionally, we observed threads connecting separated non-recombinant chromosomes that contained heterochromatic DNA and passenger complex proteins. These threads could assist the non-recombinant chromosomes in locating their homologs during their dynamic movements on the spindle. These chromosome movements and the heterochromatic threads are likely part of the mechanism ensuring proper segregation of nonexchange chromosomes.
| The accurate segregation of homologs during meiosis is essential for the propagation of virtually all eukaryotes. In many organisms proper chromosome segregation is ensured by recombination and the formation of chiasmata. Chiasmata lock homologs together and constrain the centromeres to orient towards opposite poles of the meiotic spindle, thus ensuring the proper segregation of recombinant (chiasmate) chromosomes during meiosis I. However, in some instances homologs do not undergo recombination, and thus, fail to form chiasmata. For example, in Drosophila melanogaster oocytes the 4th chromosomes are always nonexchange (achiasmate) and X chromosome recombination can be completely suppressed when oocytes are heterozygous for an X chromosome balancer, such as FM7. In both cases the nonexchange homologs are segregated faithfully, despite the lack of chiasmata.
The mechanism that mediates achiasmate chromosome segregations is called the homologous achiasmate system [1] which takes advantage of a curious feature of the biology of heterochromatin in Drosophila oocytes. Although homologs repel each other at diplotene in most organisms, this separation is incomplete in Drosophila oocytes. While the pairing and synapsis of euchromatic regions ceases at the end of pachytene, heterochromatic pairings persist until after germinal vesicle breakdown (GVBD) and during the early stages of spindle assembly [2]. Previous studies have shown that heterochromatic homology is both necessary and sufficient to ensure the proper segregation of achiasmate homologs [1],[3] and that during early prometaphase the kinetochores of achiasmate homologs are oriented toward opposite poles of the developing spindle [2]. Based on these observations it had been inferred that this initial proper co-orientation of achiasmate homologs is sufficient to ensure their eventual proper segregation.
Indeed, cytological studies of fixed oocytes demonstrated that, following the completion of the chromosome-driven assembly of the anastral spindle, achiasmate homologs are often symmetrically located on opposite halves of the spindle, such that each homolog is positioned between the main chromosomal mass comprised of chiasmate bivalents and the nearest spindle pole [4],[5]. The symmetrical positioning of these homologs on opposite half-spindles supported a model in which the initial co-orientation of achiasmate homologs, which occurs as a simple result of the maintenance of heterochromatic pairing, facilitates the proper (and precocious) segregation of achiasmate chromosomes towards opposite poles of the spindle. The separated homologs were always observed to be located on the same arc of the meiotic spindle, and thus their position with respect to the poles was thought to reflect a balance between poleward forces exerted at the kinetochore and plate-ward forces exerted on the chromosome arms by the Nod chromokinesin-like protein [4]. Following the completion of spindle assembly, the oocyte enters a prolonged metaphase arrest until passage through the oviduct initiates the onset of anaphase I [6].
The observations described above suggested a three step model for homologous achiasmate chromosome segregation. First, prior to GVBD achiasmate homologs are connected only by heterochromatic pairings. Second, these heterochromatic pairings are sufficient to ensure the establishment of kinetochore co-orientation of homologous achiasmate chromosomes. Third, following the release of heterochromatic pairings between achiasmate chromosomes during early-prometaphase, achiasmate homologs begin moving precociously towards their respective poles, stopping between the poles and the kinetochores [4]. This was presumed to be the configuration of chromosomes at metaphase I. More recently, both the observations reported here and those of Gilliland et al. [7] have necessitated a significant revision of this model.
Gilliland et al. [7] showed that the symmetrical arrangement of achiasmate chromosomes positioned between the poles and the spindle equator does not define the metaphase I-arrested oocyte, but rather is the defining feature of mid-prometaphase. At the end of prometaphase the achiasmate chromosomes congress to the metaphase plate prior to metaphase I arrest. In doing so, they join the autosomes and the chromosomes appear to form a single mass with a distinctive ‘lemon-shaped’ DNA morphology (Figure 1) [7]. The achiasmate chromosomes are oriented toward opposite spindle poles at each end of this ‘lemon’ and the meiotic spindle contracts in length after this chromosome congression [7]. These observations have allowed us to develop a classification system describing the stages from GVBD to metaphase I arrest in Drosophila oocytes. We define early-prometaphase I as the period from GVBD to the completion of a bipolar spindle. Mid-prometaphase I defines that period during which achiasmate homologs are clearly separated from the main mass and positioned between the center of the spindle and the poles. Late prometaphase I describes a poorly studied stage in which achiasmate chromosomes are retracted to the main mass in a fashion that results in their proper orientation. Finally, the term metaphase I describes the stage at which all of the chromosomes are clustered into a lemon-like structure prior to passage through the oviduct and entry into anaphase I.
We show below that during mid-prometaphase the movement of achiasmate chromosomes towards the poles is neither unidirectional nor fully coordinated, but rather a dynamic process in which separated homologs may cross the spindle midzone, rejoin their homolog, and sometimes undergo another separation event. This suggests that the separation and re-association of achiasmate homologs is a natural part of the process of achiasmate chromosomes becoming aligned on opposite halves of the meiotic spindle, and thus oriented towards opposing spindle poles. These dynamic patterns of movement suggest that achiasmate chromosomes can lose their initial co-orientation from early prometaphase and may repeatedly re-orient.
Given that achiasmate chromosomes can move away from and congress back to the metaphase plate, the question arises as to how such movements are coordinated? Recently several labs have provided evidence for the existence of connections between chromosomes during both meiosis I and mitosis [8]–[11]. LaFountain et al. [8] demonstrated that during meiosis I in crane-fly spermatocytes severing a trailing chromosome arm sometimes resulted in the trailing arm retracting back to the metaphase plate and then re-associating with its homolog on the opposite half spindle. This re-association suggested that homologs are connected by some sort of tether during meiosis I. Physical connections between chromosomes have been observed in mitotic cells [9]–[11]. Baumann et al. [9] reported the existence of threads containing the protein PICH (Plk-1 interacting checkpoint helicase) and what appears to be centromeric DNA connecting sister chromatids during mitosis in cultured cells [9]. The PICH-containing threads progressively increase in length during metaphase and disappear during anaphase [9]. Baumann et al. [9] proposed that PICH threads connecting chromosomes in mitotic cells could be used to monitor tension between the separating sister chromatids and signal checkpoint proteins. The passenger complex protein Incenp appeared to co-localize with the PICH threads in metaphase I cells [11]. While during anaphase the RecQ helicase, BLM, and its complex partners Topo-IIIα and hRMI1, also localize to the PICH threads [10]. Careful staining with BrdU demonstrated that the BLM threads connecting chromosomes are composed of DNA even when DAPI staining is not evident [10]. These threads could potentially arise by several mechanisms including from the by-products of repair of stalled replication forks or catenated centromeric chromatin.
One can imagine that physical connections between achiasmate chromosomes in Drosophila oocytes could assist achiasmate chromosomes in associating with their homologs during their dynamic movements and facilitate the re-establishment of co-orientation. We demonstrate that during the period of oscillatory chromosome movement, migrating achiasmate chromosomes are connected by pericentric heterochromatic threads that can span large distances and that the chromosome passenger complex proteins Incenp and Aurora B appear to localize to these threads. We propose that these threads act to restrict the movement of achiasmate homologs and facilitate their rejoining during prometaphase.
The studies of achiasmate chromosome movement in living mid-prometaphase oocytes described below often revealed periods of time in which both achiasmate homologs were located on the same half-spindle. Since such figures had not previously been reported in fixed wild-type oocytes, we conducted an examination of a large number of fixed mid-prometaphase/metaphase oocytes to determine the position of the achiasmate chromosomes relative to the main chromosomal mass. Because heterozygosity for the multiply inverted X chromosome balancer FM7 suppresses exchange between the X chromosomes we analyzed fixed images from wild-type oocytes with chiasmate (X/X) or achiasmate Xs (FM7/X) [1]. The results of this analysis are presented in Figure 2.
In 703 fixed mid-prometaphase/metaphase oocytes derived from FM7/X females, 111 were identified in which both achiasmate Xs were separated from the main chromosomal mass. Although the X chromosomes were symmetrically positioned on opposite half-spindles in 102 of these 111 oocytes, in 9 cases we observed the X chromosomes on the same half of the meiotic spindle. In 8/9 cases the two achiasmate X chromosomes appeared to be physically associated (Figure 2). Moreover, both X chromosomes appeared to be oriented toward the nearby pole in all nine oocytes. Indeed, with a few exceptions, including one presented below, we always observed the centromeres of achiasmate chromosomes to be oriented towards the closest pole regardless of whether the homolog is on the same or opposite half-spindle. This was determined by the known locations of the centromere relative to the brightly-staining blocks of heterochromatin on both X and FM7 (Figure 2) [2].
Among 610 fixed X/X prometaphase/metaphase oocytes, 19 were found to have achiasmate Xs that were separated from the main chromosomal mass. (The presence of such oocytes in females carrying two normal sequence X chromosomes is not surprising given that the X chromosomes fail to recombine in 5–10% of normal meioses.) In a single X/X oocyte, both the X chromosomes were found to be on the same half spindle (Figure 2). These X chromosomes were physically associated with each other, but were oriented vertically along the Z axis so it is not clear whether they are pointed towards or away from the nearest spindle pole. However, the centromeres of both homologs were clearly oriented in the same direction. Thus, it is clear that even if oocytes carry structurally normal X chromosomes, achiasmate chromosomes do not always separate and proceed towards opposite poles in a symmetric fashion during prometaphase.
The appearance of oocytes in which both achiasmate homologs are found on the same half-spindle raises obvious questions about the event(s) that created them. For example, does a mechanism exist in which separated homologs are free to move back and forth along the same arc of the meiotic spindle? To address this question we used live imaging to study chromosome movements during prometaphase.
Live-imaging has been successfully used by other groups to examine early and mid-prometaphase I in wild-type Drosophila oocytes [12],[13]. However, oocytes with achiasmate X chromosomes were not observed. Additionally, these studies utilized only a fluorescent spindle marker, and any chromosome movement was inferred from the dark spots in the fluorescent spindle label 12,13. In our studies it was essential that we also label the chromosomes so the movements and identities of the relatively small achiasmate X chromosomes could be more easily discerned. Using Oli-green to label the chromosomes and rhodamine-conjugated tubulin to label the spindle microtubules allowed us to clearly visualize the chromosomes within the meiotic spindle. Our lab has reported the use of these methods to successfully examine metaphase I arrest in wild-type oocytes and prometaphase in several meiotic mutants [7],[14],[15].
We investigated whether oocytes with achiasmate X chromosomes would display any differences in the progression of early prometaphase (GVBD and spindle assembly) from X/X oocytes. (A brief description of early to mid-prometaphase for oocytes with chiasmate Xs is provided in the Supporting Information.) In 14 out of 17 FM7/ X oocytes undergoing GVBD, the chromosomes remained tightly associated as a single mass during spindle assembly and for the duration of imaging after the formation of a long tapered spindle (Video S1 and Figure S1A). For the remaining 3 oocytes, while the chiasmate bivalents remained together, one or both achiasmate Xs briefly moved away from the autosomes after the completion of spindle assembly. In one instance, the two achiasmate Xs were observed to move out toward opposite sides of the spindle midzone approximately 22 minutes after the completion of spindle assembly and then returned to the spindle midzone (Video S2 and Video S3). In another case, the Xs briefly moved together out from the spindle midzone and then rejoined the main chromosomal mass (data not shown). In the third case, one achiasmate X moved away from the spindle midzone and quickly returned (data not shown). Thus, for oocytes with achiasmate Xs, the achiasmate chromosomes have the ability to move away from and even rejoin the spindle midzone shortly after a bipolar spindle has been achieved. We speculate that since a majority of the oocytes in early prometaphase displayed no discernable chromosome movement after the completion of spindle assembly a constraint must be lifted or a signal given to initiate the movements of mid-prometaphase (see below).
Because our study of fixed images revealed instances of mid-prometaphase oocytes with achiasmate Xs on the same half spindle, we used live-imaging of FM7/X oocytes during mid-prometaphase to more carefully examine the movements of achiasmate homologs. Nineteen oocytes were identified as being in mid-prometaphase by virtue of having X chromosomes visibly separated from the chiasmate autosomes. In 16/19 oocytes, the achiasmate Xs were found associated on the same half of the meiotic spindle at some point during imaging and often were observed to display dynamic movements on the meiotic spindle during mid-prometaphase. In 10 out of these 16 oocytes, the two associated X chromosomes moved back towards the spindle midzone and approached the main chromosomal mass during the course of live-imaging. An example is shown in Figure 3 and Video S4. At the start of this movie the two achiasmate Xs are clearly visible on the same side of the spindle midzone. The two Xs associated within a few minutes and moved together back towards the spindle midzone. In a second oocyte (Video S5 and Video S6), following the return of the associated X chromosomes to the main mass, the Xs dissociated and then assumed positions on opposite sides of the autosomes at the spindle midzone. However, in perhaps the most informative case (Figure 4 and Video S7), we observed an oocyte in which a set of re-conjoined X chromosomes arose from well separated homologs—a process that included an X traversing the meiotic spindle. Figure 4 shows the X chromosomes of this oocyte starting out on opposite halves of the meiotic spindle followed by one X traversing the spindle to re-associate with its homolog (Figure 4 and Video S7). Eventually the rejoined homologs then move back towards the spindle midzone (Figure 4 and Video S7).
In 6/16 oocytes with associated Xs, the associated Xs did not move together back towards the spindle midzone during imaging. However, in two of the these oocytes X chromosomes were associated on the same side of the spindle midzone at the beginning of imaging and then dissociated such that one X moved to a position on the other side of the spindle midzone (Video S8 and Video S9). In the final 4/16 FM7/X mid-prometaphase oocytes, the 2 conjoined X chromosomes remained on the same side of the spindle midzone throughout the period of imaging. Since chromosomes are retracted for metaphase arrest, we speculate that if imaging could have been continued the achiasmate chromosomes would have eventually congressed back to the main chromosomal mass [7].
In 3/19 oocytes we observed the achiasmate Xs on opposite half spindles for all or part of imaging without their association on the same half spindle (Figure S1B and S1C). In one of these oocytes both of the Xs returned to the main chromosome mass. In the other two oocytes the achiasmate chromosomes remained on opposite half spindles for the duration of imaging.
In three oocytes, all the chromosomes were tightly associated at the spindle midzone of the bipolar spindle for the duration of observation. These oocytes were likely in metaphase I arrest [7].
Another striking example of oscillating achiasmate chromosomes was observed in the course of our studies of FM7 nodb17/noda oocytes, which have defects in the polar ejection force (PEF) generated by the Nod protein and have achiasmate X chromosomes [4],[16],[17]. The loss of the force pushing chromosomes back to the spindle midzone appeared to increase the movement of achiasmate chromosomes. Although in most cases homologs were ejected from the meiotic spindle in nod oocytes (KAC, SFH, JLC, and RSH, unpublished data), we observed an FM7 nodb17/noda oocyte in which an achiasmate X chromosome crossed the spindle midzone five times during imaging (Figure S2 and Video S10). This phenomenon is not restricted to this single oocyte, as achiasmate chromosomes were also observed to cross the spindle midzone in additional nod mutant oocytes (KAC and RSH, unpublished observations).
To be sure that the behavior of achiasmate X chromosomes described above was not idiosyncratic to chromosomes like FM7 that contain heterochromatic rearrangements, we performed similar studies of females of the genotype In(1)dl-49/X. In(1)dl-49 carries only a single euchromatic inversion that greatly reduces the frequency of X chromosomal exchange [18],[19]. As was the case for FM7/X females we observed the achiasmate X chromosomes in this genotype to show oscillatory behavior (Video S11 and Video S12). At the start of Video S11 the X chromosomes are on opposite halves of the spindle. The X on the top half of the spindle crosses the spindle midzone and associates with its homolog on the bottom half of the spindle. This movie demonstrates that the ability to cross the spindle midzone and associate with a homolog is not particular to the FM7 chromosome, but rather are characteristics of achiasmate chromosomes.
Using live-imaging we observed that achiasmate chromosomes undergo unexpected dynamic movements on the meiotic spindle. Besides having the ability to move towards opposite poles (as assumed from previous fixed oocyte studies) we have shown that achiasmate chromosomes can associate on the same half of the meiotic spindle. Additionally, we observed an achiasmate X crossing the spindle midzone to associate or dissociate with its homolog in 3/19 FM7/X oocytes. In several cases, the achiasmate 4th chromosomes were observed to undergo dynamic movements on the meiotic spindle (data not shown). (Due to their smaller size and decreased fluorescence it was more difficult to consistently observe the achiasmate 4th chromosomes using live-imaging).
The movements of the achiasmate chromosomes upon the meiotic spindle during prometaphase also affected the stability of the chiasmate autosomes on the spindle midzone. Unlike what is typically observed in X/X oocytes (see below), the two sets of autosomes were seen to “slip” in respect to each other, rather than remaining tightly aligned on the spindle midzone in 11 of the 16 (62%) mid-prometaphase FM7/X oocytes that displayed Xs on the same side of the spindle midzone during mid-prometaphase. In 10 of these 11 oocytes the slippage resulted in the two sets of autosomes being fully separated from each other at some point during imaging. Figure 4G shows an example of this phenomenon with the two major autosomes clearly separated from each other. In the remaining oocyte the autosomes are offset with respect to one another but still touching at their ends. Typically when the achiasmate Xs were positioned at the spindle midzone this slippage of the autosomes was not observed, supporting the idea that this autosome movement is a secondary effect of the movement of the achiasmate chromosomes.
The slippage of chiasmate chromosomes is also observed at a substantially lower frequency (9/53) in X/X oocytes, however, it was difficult to discern whether the chromosomes that had moved away from the spindle midzone were autosomes or X chromosomes in five of the nine oocytes. In 4/9 of the oocytes the identities of the chromosomes could be established, and the slippage of the two major autosomes was associated with prior movement of X chromosomes. Thus, these examples of autosomal slippage may represent the 5–10% of X/X oocytes in which the X chromosomes fail to undergo exchange (see Supplemental Data).
The slippage of chiasmate autosomes in response to the positioning of the achiasmate chromosomes could also be seen in our fixed images. Out of 176 oocytes from three day old virgin FM7/X females, the X and FM7 chromosomes were physically separated from the chiasmate chromosomes in 31 oocytes. Of those 31 oocytes, ten had the two major autosomes asymmetrically positioned relative to each other. Of those ten oocytes, seven had clear chromosome positioning defects (such as both X or 4 homologs on same side of spindle, or an X was closer to the adjacent pole than the nearby 4). In contrast, of the 21 oocytes that did not appear to have autosomal slippage, only two had chromosome positioning defects. Our sample size for comparably aged chiasmate X females was considerably smaller, but out of three oocytes with nonexchange chromosomes out on the spindle, two appeared to have chiasmate autosomal slippage as well. In fixed X/X oocytes we rarely observed chiasmate chromosomes that had moved away from the spindle midzone unless spontaneous achiasmate X chromosomes were also present (data not shown).
This low frequency of chromosome movement observed in X/X oocytes is consistent with Matthies et al. [13], who reported little movement of chiasmate chromosomes away from the spindle midzone in X/X oocytes. These results differ from live-imaging experiments reported by Skold et al. [12]. The authors interpreted from dark spots in fluorescent NCD-GFP in X/X oocytes that chromosomes often moved out from the spindle midzone during prometaphase [12].
A previous study of ald (the Drosophila homolog of mps1, a protein kinase required for the spindle assembly checkpoint) showed images of prometaphase oocytes from ald mutant females with DNA threads running between chromosomal masses [20]. The importance of these threads became more evident when unmistakable threads were also found between the obligately achiasmate 4th chromosomes in a strong ald mutant genotype (Figure 5A). It has been shown that blocks of homologous heterochromatin are necessary and sufficient for pairing of nonexchange chromosomes [1]–[3]. However the mechanism for maintaining this association is unknown. We reasoned that these blocks of heterochromatin may actually be paired by a mechanism mediated by DNA-DNA linkages of some sort, and that ald mutants enhance these threads in some manner that makes them easier to observe. If this idea is correct, then these threads should also be present in wild-type oocytes. Therefore, we searched for evidence of threads in wild-type oocytes.
While normal image processing methods revealed little evidence of threads, by processing images to emphasize faint details (at the expense of overexposing the main chromosome masses) we were able to find clear evidence of threads in oocytes from wild-type FM7/X (Figure 5B) and X/X (Figure 5C) oocytes. We observed DAPI threads running between 4th chromosomes (Figure 5B and 5C), as well as between FM7 (Figure 5D) and normal sequence X chromosomes (Figure 5D and 5E).
However, clear threads coming from achiasmate X chromosomes were found less frequently than threads coming from 4th chromosomes. We believe that this paucity of examples of X threads reflects a difficulty in observation, rather than an actual deficit in frequency, for two reasons. First, threads can only be visualized when a chromosome is sufficiently separated away from any adjacent chromosomes that there is enough space to see the thread. The 4th chromosomes move closer to the spindle pole than the Xs, and are therefore more likely to have enough space for threads to be visualized. Since X chromosomes remain closer to the autosomes, fewer figures are available where X threads could have been detected if present. Secondly, threads frequently taper off to the point where they can no longer be detected by DAPI staining in a few microns. As the entire 4th chromosome is only about 0.5–1 microns long, threads can escape occlusion by the originating chromosome over a short distance. However, the X itself is around two microns long, and most X chromosomes are oriented with the euchromatin pointed back towards the spindle midzone, along the arc of the spindle that a tethering thread would follow. Therefore, X threads coming from the centromeric heterochromatin would have to be longer than the rest of the chromosome arm to have a chance to be seen. We note that our figure with the most robust X chromosome thread (Figure 5D, left inset) was apparently fixed in the middle of reorienting one of its chromosomes, with its centromere pointed away from the adjacent spindle pole. In this figure, the thread is coming off the side of the chromosome, and tapers off to invisibility by ∼2.5 microns.
If these threads are the mechanism that allows heterochromatin to mediate pairing, then they should contain heterochromatic DNA sequences. To determine this we performed Fluorescent In-Situ Hybridization (FISH) using probes that preferentially highlighted either X or 4th chromosome heterochromatic repeats [2]. For the 4th probe we were able to find evidence of FISH probe running between homologs (Figure 6A and 6B). These data show that the DAPI containing threads are comprised, at least in part, by heterochromatin and are consistent with the threads being part of the mechanism by which heterochromatic pairings ensure the proper segregation of achiasmate chromosomes. Additionally, we have recently demonstrated that the metaphase I arrest configuration in Drosophila oocytes has all chromosomes retracted into what appears to be a single DNA mass [7]. By examining metaphase I arrested oocytes, we were able to observe the 4th chromosome probe hybridizing along threads in the compact mass (Figure 6C). This suggests that these threads are not fully resolved until meiotic anaphase I.
In addition to the heterochromatic connections between homologs, we were able to observe threads apparently running between a 4th chromosome and a nearby X (Figure 6D). The observation of X-4 threads could occur for multiple reasons. Frequently the chromosomes are aligned with all chromosome masses along a single arc of the spindle. Therefore, any thread running from a 4 to its homolog would by necessity first run into the adjacent X, and could be mistaken for an X-4 heterologous thread. Alternatively, these threads could represent genuine linkages between heterologous chromosomes. We note that Dernburg et al. [2] found that the 4th chromosome satellite probe often hybridizes to a small spot on the X near the centromere, which is also evident in our figures (Figure 6D, 6E, and 6F). There is also an unexplained genetic relationship between the X and 4th chromosomes that causes their rates of nondisjunction to be strongly correlated across many different Drosophila meiotic mutants [21]–[23]. Moreover, even in otherwise genetically normal females an extra copy of chromosome 4 increases the frequency of X chromosome nondisjunction [24], and small duplications of X heterochromatin can induce high levels of 4th chromosome nondisjunction [25]. These findings suggest that inter-heterolog threads may be a genuine phenomenon, and could potentially facilitate the proper segregation of a single X and a single 4th chromosome to each half of the meiotic spindle, which would provide a mechanism behind the correlation of X and 4th chromosome nondisjunction rates. This finding would also be consistent with those of a recent study in male Drosophila, where a condensin mutant exhibited DNA bridges between heterologous chromosomes [26]. While we do not attempt to differentiate these possibilities here, we note that both inter-homolog and/or inter-heterolog connections are consistent with the proposals that achiasmate chromosomes remain connected by DNA tethers during prometaphase and that these threads facilitate the proper segregation of achiasmate chromosomes.
We then set out to determine the frequency of oocytes with DAPI threads connecting their 4th chromosomes. In order to visualize these threads in the presence of the much brighter DAPI staining of the autosomes, achiasmate chromosomes must have moved out on the spindle to allow for adequate space between the chromosomes. Gilliland et al. [7] showed that the proportion of oocytes in this configuration declines with female age, as females store metaphase I arrested oocytes with chromosomes in a single DNA mass. Therefore, to quantify thread abundance, we dissected and fixed the ovaries from 2 day old X/X virgin females, and identified 45 oocytes where the threads connecting nonexchange chromosomes could potentially be visualized.
We then examined these oocytes for evidence of threads and scored them into three categories (Figure 7). Oocytes were scored as “−“ if they had no discernable evidence of threads (Figure 7A), scored as “+” if the chromosomes did not have clear threads but did have some evidence of threads such as short spurs or hooks of DAPI staining coming off the chromosome in the direction of its homolog (Figure 7B), and scored as “+++” if there were unmistakable DAPI threads coming off the nonexchange chromosomes towards their homologs; these threads were not necessarily complete, but were long and robust enough that they could be clearly seen without image enhancement (Figure 7C). We scored 9 oocytes as “−“, 22 as “+”, and 14 as “+++”. Therefore, 14/45 (31%) of oocytes had unmistakable threads, and 36/45 oocytes (80%) had at least some evidence of threads, suggesting that achiasmate chromosomes are still connected to their homolog by heterochromatic threads during mid-prometaphase, when dynamic movements on the meiotic spindle is observed.
It is not surprising that the largest class of oocytes is “+”, which shows evidence of threads but have no visible complete DAPI threads. In studies of PICH containing threads in mitotic cells [9]–[11], PICH failed to colocalize with DAPI, but it did colocalize with BrdU under carefully controlled conditions [11]. This suggests that while the threads are composed of DNA, DAPI is not sensitive enough to detect them once they become very thin, or that the DNA in threads may be in the wrong configuration for robust DAPI fluorescence. Therefore, we reason that detection by DAPI would underestimate the true abundance of threads, and conclude that these threads are, at the very least, relatively common in wild-type meioses.
If these threads are being used to guide the biorientation of nonexchange chromosomes, it is reasonable to expect that this process requires interactions with the spindle and spindle-associated proteins. In addition to PICH localizing to ultrafine anaphase bridges [9], the chromosomal passenger complex protein Incenp was found to localize to PICH- and BrdU-containing anaphase threads in mitotic cells [11]. While there is no identifiable PICH homolog in flies, we were able to localize Incenp in Drosophila oocytes, and we found that Incenp appeared to localize to the meiotic threads projecting from 4th chromosomes (Figure 8A). They also could be found running between blocks of heterochromatin on X chromosomes, even when chromosomes appeared to be chiasmate (Figure 8B) or when the chromosomes were not sufficiently separated for DAPI threads to be visualized (Figure 8B and 8C). We also note that these proteins localized strongly along the thickened tubulin arc of the spindle (Figure 8B) as had been seen for DAPI threads in other figures (Figure 5C). Additionally, while Wang et al. [11] failed to see the chromosomal passenger complex protein Aurora B colocalize to mitotic PICH threads, we found that Aurora B also localized to DAPI threads in Drosophila oocytes (Figure 8D). The identification of Incenp to both meiotic and mitotic DNA threads suggests that DNA threads may be playing conserved roles in mitosis and meiosis.
To quantify how often this association was seen, we dissected wild-type females, treated their ovaries with anti-Incenp antibodies and scored figures with appropriate chromosome configurations as described above. Out of 37 figures with chromosomes in positions where threads could be seen, 35 had staining along where DNA threads would be located, even when visible DAPI threads were not evident. However, we must emphasize that Incenp and Aurora B are not primarily localizing to threads, as both localize along chromosome arms and other parts of the spindle. Furthermore, we must also emphasize that the tubulin spindle is also in this region, and therefore we cannot discriminate between an enrichment of Incenp on the microtubules running between nonexchange chromosomes, or Incenp localization to a DNA thread in the same region of the spindle.
In this paper we demonstrate that the positions of achiasmate chromosomes during mid-prometaphase are much more dynamic than fixed images had previously suggested [4],[5]. Both our fixed and live imaging studies demonstrate that achiasmate chromosomes can undergo movements during which a given achiasmate chromosome crosses the spindle midzone to rejoin its homolog, followed by either movement of both homologs back to the main chromosomal mass or movement of one homolog back to the other ‘empty’ half spindle. Additionally, centromeres were pointed towards the closest pole in 9/9 fixed images of FM7/X oocytes with Xs on the same half spindle (see Figure 2). These results are consistent with a view of mid-prometaphase in which the kinetochores of achiasmate homologs often lose their connections to the closest pole, re-associate with the more distant spindle pole, and then traverse the meiotic spindle as they move towards that pole. These results suggest that the initial co-orientation at early prometaphase by heterochromatic pairings is not the sole mechanism by which heterochromatic pairings ensure the proper segregation of achiasmate homologs during meiosis I.
We speculate that the dynamic movement of the achiasmate X chromosomes on the spindle disrupts the position of the autosomes on the spindle midzone. These movements could disrupt the balanced forces acting from opposing spindle poles on the chiasmate chromosomes and push them temporarily closer to a spindle pole. Upon congression of the achiasmate chromosomes to the spindle midzone, the autosomes can maintain their balanced positions on the spindle midzone.
What mechanisms then eventually ensure the partitioning of achiasmate chromosomes to opposite half-spindles, and the stable co-orientation of achiasmate homologs? One could imagine a crowded pole model in which some mechanism exists to limit the number of homologs per half-spindle such that the presence of an additional X or 4 is poorly tolerated [27]. Interestingly, in 4/16 movies in which both Xs were on the same half-spindle at some point during imaging, they remained on that same half-spindle throughout the length of each movie. Because the average length of these movies is rather short (8.6 minutes), it is possible, and perhaps even likely, that congression towards the spindle midzone of one or both Xs would have occurred if image could have continued. Nonetheless, the longest of these movies was 17.5 minutes in length, and thus if such a “crowded pole” model is correct it must tolerate the presence of two nonexchange X chromosomes on the same half spindle for at least that period of time.
Following Nicklas [28],[29] we propose that the inability of achiasmate homologs to maintain a stable association with a given pole (and to occasionally move to the other) reflects a requirement for sustained tension to maintain kinetochore associations with a given pole [28],[29]. We speculate three forces act upon achiasmate homologs: 1) poleward forces exerted at the kinetochore; 2) the ‘polar ejection force’ (PEF) that acts to push chromosomes towards the center of the spindle, and which is mediated in Drosophila by the Nod kinesin-like protein [4],[17]; and 3) the heterochromatic threads described above. We propose that although properly oriented achiasmate homologs initially move toward opposite spindle poles, at some frequency a migrating achiasmate homolog loses its kinetochore attachment and then is pushed away from the closest pole and towards the main mass by the PEF. This movement is eventually blocked by the PEF generated by the other pole, which increases in strength as a chromosome approaches a pole [30]. However, if the migrating chromosome achieves attachment to the other pole, this may be sufficient to allow it to move toward that pole, thus rejoining its homolog on the same side of the spindle. Indeed, the role of the PEF emanating from the opposite pole in blocking frequent crossings of the mid-spindle may well explain the high frequencies of such movements in a nod mutant oocyte (Video S10 and Figure S2). For reasons described below, we infer that because kinetochore-pole attachments are not fully stable at this stage, such mid-positioning of achiasmate homologs are ‘corrected’ by the very same process that created them, namely loss of attachment of one or both kinetochores to the nearby pole and subsequent re-attachment of one or both homologs to the opposite pole.
Thus, in a model akin to the one proposed by Nicklas [28] for mal-oriented chiasmate bivalents in male meiosis, we propose that during mid-prometaphase achiasmate homologs undergo cycles of attachment and reattachment until stable orientation mediated by kinetochore attachment, polar ejection forces, and perhaps the tension generated by heterochromatic threads during prometaphase is achieved [28]. However, this period of instability is limited because achiasmate homologs eventually retract into the main chromosome mass prior to metaphase arrest [7]. These observations raise questions regarding the mechanisms involved: what terminates the oscillatory behavior of achiasmate chromosomes and allows them to maintain stable co-orientation? One possible explanation for this transition is provided by the seminal work of Brunet et al. [31]. These authors observed that during early prometaphase in mouse oocytes chromosomes do not possess stable ‘ends-on’ kinetochore microtubule interactions, but rather oscillate about the spindle midzone - possibly as a consequence of both lateral associations between the kinetochore and spindle microtubules and interactions of the chromosome arms with microtubules. Evidence that lateral kinetochore associations are sufficient to mediate long range chromosome movements during congression has been provided by Kapoor at al. [32]. Brunet et al. [31] further showed that the transition from such lateral associations of kinetochores with the microtubules to more canonical kinetochore-microtubule interaction correlates with the removal of CLIP-170 from the kinetochores—perhaps as a component of allowing meiotic kinetochores to become ‘competent’ to form proper kinetochore microtubule associations [31]. (Consistent with this view, we found evidence that the fly homolog of CLIP-170, known as CLIP-190 [33] is present on prometaphase kinetochores but absent on metaphase chromosomes. [WDG and RSH, unpublished data]).
Our findings and previous work lead us to a model in which during prometaphase, meiotic chromosomes lack the ‘strong’ ends-on kinetochore associations characteristic of metaphase chromosomes. Chiasmate chromosomes, which possess bi-oriented (and linked) kinetochores, oscillate to some degree around the spindle midzone, perhaps limited by their larger size. However, achiasmate X and 4th chromosomes are more likely ‘buffeted’ about the developing spindle due to a lack of chiasmata, lateral movements mediated by the kinetochore, and the PEFs emanating from the two poles. We imagine that stable positioning of these achiasmate chromosomes is not achieved until kinetochore maturation at the end of prometaphase, both as a consequence of the formation of canonical kinetochore-microtubule interactions and perhaps also by the tension generated by heterochromatic threads.
This model is consistent with our previous studies of meiotic progression in oocytes homozygous for loss-of-function mutations in the ald/mps1 gene. In these oocytes, prometaphase is greatly shortened and oocytes appear to enter anaphase I (as defined by loss of sister chromatin cohesion along the euchromatic arms) upon the completion of spindle assembly [14],[20]. In such oocytes, achiasmate homologs that have not ‘had time’ to establish stable bi-orientations are often caught on the same half spindle connected by DNA threads.
The process of resolving the heterochromatic threads may contribute to stabilizing these achiasmate oscillations, and also eventually to their retraction into the main mass by metaphase I by progressively limiting the distance between two achiasmate homologs [7]. As described below, that such threads can generate force is suggested by the studies of potentially similar inter-homolog connections by LaFountain and his collaborators [8]. Such threads may also serve to prevent chromosomes from moving too close to the spindle poles, (thus decreasing the PEF) and function like chiasmata in terms of balancing the kinetochore forces on a given pair of achiasmate homologs. Based on the observation of FISH-hybridizing threads in metaphase-arrested oocytes (Figure 5H), and by analogy to the progressive resolution of ultrafine bridges during mitotic anaphase [9], we expect that heterochromatic threads connecting homologs are resolved during meiotic anaphase I. However, we cannot rule out that threads are resolved during late-prometaphase or metaphase arrest. Failure to resolve the threads by anaphase I would likely prevent the proper segregation of chromosomes or the heterochromatic threads would be severed leading to DNA loss. The presence of heterochromatic sequences other than those recognized by the FISH probes or euchromatic sequences within the DAPI threads will require further investigation. Additionally, whether the threads are composed of chromatin will also require further research. Finally, data from ald mutant oocytes suggest DNA threads exist between autosomes (data not shown), but since the chiasmata lock homologs in close proximity it will be more difficult to definitively confirm their existence in wild-type oocytes [20].
LaFountain et al [8] suggested the existence of threads connecting homologs at their telomeres during meiosis in crane-fly spermatocytes. Chromosomes moving towards the spindle poles during meiosis I were often observed to have chromosome arms trailing towards the metaphase plate. When these trailing chromosome arms were severed from their kinetochores they were sometimes observed to move towards the metaphase plate, cross the metaphase plate and even rejoin their homologous chromosome arms while their kinetochores continued progressing towards their original pole [8]. This work suggested the existence of a thread between homologs, and that a tension existed on the thread that could bring homologs together if the opposing poleward force was released [8]. The force on the homologs decreased as meiosis I progressed [8]. These results suggest that DNA threads connecting homologs may be a conserved aspect of meiosis I. Furthermore, a recent paper reported that a condensin mutant resulted in DNA threads connecting nonhomologous chromosomes in Drosophila male meiosis [26]. That study did not determine if bridges are still present in wild-type males. But by analogy with our initial discovery of threads in ald, the DNA bridges may be present but less prominent in wild-type. If this proves to be the case, it would be consistent with this mechanism being utilized to ensure proper segregation of chromosomes in meiosis I of both sexes.
The tight physical association of homologous heterochromatin in early meiosis suggests that the heterochromatic connections are established during DNA replication [2]. One potential mechanism for establishing linkages would be as a by-product of the repair of stalled replication forks [34]. This utilizes a strand invasion mechanism, which after resolution of the three-strand intermediate can result in the catenation of DNA strands from homologous chromosomes. These catenations would then hold homologs together by their heterochromatin until Topo-II or other enzymes resolve them. While we currently have no direct evidence for how threads are established, this model has the advantage that the formation of threads would be a spontaneous side effect of ancestral DNA repair machinery, which could then be co-opted by evolution to facilitate the segregation of nonexchange homologs.
Our studies have led to a revised model of meiosis I in Drosophila oocytes (Figure 9). The chiasmate and achiasmate homologs are paired due to chiasmata and heterochromatic pairings, respectively, at the end of prophase. During prometaphase and metaphase I chiasmate chromosomes are properly co-orientated for segregation due to the chiasmata locking chromosomes on the spindle midzone. Achiasmate chromosomes can move towards opposite spindle poles (Figure 9, top) or towards the same spindle poles while being connected by heterochromatic threads (Figure 9, bottom). Achiasmate chromosomes can lose their initial co-orientation during mid-prometaphase, and the heterochromatic threads could assist in the achiasmate homologs rejoining and re-attempting proper co-orientation. Eventually the achiasmate chromosomes again successfully acquire co-orientation and are retracted to the metaphase plate to form the ‘lemon-shaped’ DNA configuration of metaphase I arrest. During metaphase or anaphase I the heterochromatic threads would then be progressively dissolved to allow proper segregation of the chromosomes.
Future work is needed with respect to the mechanism by which heterochromatic threads are formed and resolved during meiosis I, into how the threads function to bring mal-oriented achiasmate chromosomes together, and how they facilitate the eventual co-orientation and segregation of achiasmate chromosomes. The discovery of heterochromatic threads provides potentially new insights into the mechanism of how heterochromatic pairing ensures proper achiasmate chromosome segregation.
Chiasmate wild-type (X/X) oocytes were from females that were yw;pol; w1118 , Oregon R or an F1 cross between w1118 and Oregon R. Oocytes with achiasmate X chromosomes (FM7/X) were obtained by crossing w1118 flies to FM7; pol flies and examining the oocytes from the resulting FM7/ w1118; pol/+ female progeny. FM7 is a balancer chromosome that completely suppresses recombination with a normal X chromosome (and carries the marker y, w1 and B). The genotype of the ald flies were FM7/X; ald{P:GS13084-excision23}/Df(3R)ED5780 as described in [14],[20]. The genotype of the nod mutant flies was FM7 nodb17/noda.; pol [35],[36]. The dl-49/X flies were obtained by crossing dl-49 v f; pol flies to w1118 flies [18].
Live-imaging was performed similar to Matthies et al. [37] and Davis [38] with a few modifications. Briefly, approximately stage 13 oocytes were dissected from ovaries of 2–3 day old, well-fed adult females and the oocytes were aligned in halocarbon oil 700 (Sigma) on a no. 1 ½ coverslip in which a small well had been made within electrical tape. Oocytes were injected using standard micro-injection procedures with an approximately 1∶1 ratio of rhodamine-conjugated tubulin minus glycerol (Cytoskeleton) and Quant-iT OliGreen ssDNA Reagent (Invitrogen) diluted 0.7–1 fold with water. After injection oocytes were covered with a piece of YSI membrane.
The well slides were placed on a temperature-controlled bionomic controller (Technology, Inc) set at 23–24°C. Oocytes were imaged using a LSM-510 META confocal microscope (Zeiss) at 40× with a zoom of 2–2.5 or 100× with a 1.5 zoom. Images were acquired using the AIM software v 4.2 by taking a 10 series Z-stack at 1 micron intervals with 20 seconds between acquisitions which resulted in a set of images approximately every 40 seconds.
For the determination of when achiasmate Xs moved off the spindle midzone multiple oocytes were examined at approximately 15 minute intervals from GVBD until clear images of a meiotic spindle were no longer possible or by observing a single oocyte for a long period of time at approximately 40 second intervals. The time of GVBD and when achiasmate Xs had separated from the autosomes were recorded.
For image processing Z-stacks were made into 2-D projections using maximum projection and concatenated using the AIM v 4.2 software.
For in situ hybridization the 1.686 satellite sequences (also known as the 359 bp repeats) on the X chromosome and the AATAT repeats on the 4th chromosome were chosen as probes [2],[19]. The Alexa Fluor 488 and 647 dyes were used. The details of probe generation and labeling, egg chamber dissection and fixation, fluorescent in situ hybridization and microscopy observation were described in Xiang and Hawley [19] with one notable exception: the egg chambers were denatured with the probes at 82°C rather than 94°C for 2 min before incubation for hybridization at 30°C overnight [19]. Immunofluorescent preps were performed as previously described in Gilliland et al. [14] with rat anti-alpha-tubulin (1∶250), anti-Incenp (1∶500), anti-Aurora B (1∶500), and anti-CLIP-190 (1∶200). All secondary antibodies (Alexa 488 and Alexa 555) were at 1∶250. DAPI-only preps were performed as described in Gilliland et al. [7].
Microscopy of fixed oocytes was conducted using a DeltaVision microscopy system (Applied Precision, Issaquah, WA) equipped with an Olympus IX70 inverted microscope and high-resolution CCD camera. All images were acquired with the 100× objective, some with 1.5× auxiliary magnification. The image data were deconvolved using the SoftWoRx v.25 software (Applied Precision) and projected with multiple stacks as described in Xiang and Hawley [19]. For image enhancement of threads, threads were first identified by examination of the unprojected image stack. Only the image slices that contained the threads were used for making 2D projections (instead of the entire stack), and were projected using SoftWoRx's Sum projection method instead of the Maximum Intensity projection. Coloration was adjusted to increase visibility of the threads over the background, allowing the main chromosome masses to be saturated.
|
10.1371/journal.ppat.1002900 | Telomere Length Affects the Frequency and Mechanism of Antigenic Variation in Trypanosoma brucei | Trypanosoma brucei is a master of antigenic variation and immune response evasion. Utilizing a genomic repertoire of more than 1000 Variant Surface Glycoprotein-encoding genes (VSGs), T. brucei can change its protein coat by “switching” from the expression of one VSG to another. Each active VSG is monoallelically expressed from only one of approximately 15 subtelomeric sites. Switching VSG expression occurs by three predominant mechanisms, arguably the most significant of which is the non-reciprocal exchange of VSG containing DNA by duplicative gene conversion (GC). How T. brucei orchestrates its complex switching mechanisms remains to be elucidated. Recent work has demonstrated that an exogenous DNA break in the active site could initiate a GC based switch, yet the source of the switch-initiating DNA lesion under natural conditions is still unknown. Here we investigated the hypothesis that telomere length directly affects VSG switching. We demonstrate that telomerase deficient strains with short telomeres switch more frequently than genetically identical strains with long telomeres and that, when the telomere is short, switching preferentially occurs by GC. Our data supports the hypothesis that a short telomere at the active VSG expression site results in an increase in subtelomeric DNA breaks, which can initiate GC based switching. In addition to their significance for T. brucei and telomere biology, the findings presented here have implications for the many diverse pathogens that organize their antigenic genes in subtelomeric regions.
| A broad array of human pathogens (including bacteria, fungi and parasites) vary the proteins on their cell surface to escape the immune response of their hosts. This process, called antigenic variation, relies on a repertoire of variant protein encoding genes in the genome and the organism's ability to accurately switch from the expression of one variant gene to another. A common theme in both the diversification of these variant genes and the mechanisms required for their expression is that they are often located near the ends of chromosomes. The ends of chromosomes are protected by structures called telomeres. Regions near the telomere are referred to as subtelomeric and are commonly thought to be comparatively unstable DNA sites. It is therefore intriguing that organisms that rely on antigenic variation for survival would organize their critical survival genes in these sites. Trypanosoma brucei is a model organism for the study of antigenic variation. The causative agent of African sleeping sickness, this unicellular parasite possesses an antigenic repertoire of unparalleled diversity, which can only be expressed from specific subtelomeric sites. Here we use the power of the T. brucei model to investigate the effect of telomere length on antigenic variation.
| Trypanosoma brucei is an extracellular human pathogen with an unparalleled capacity to evade host humoral immunity. The causative agent of African sleeping sickness in humans and nagana in cattle, T. brucei is transmitted into the bloodstream of its host by a tsetse vector and can grow to densities as high as 109 cells per milliliter of blood. The parasitemia is cyclically diminished to nearly undetectable levels, to be followed by another wave of immense growth [1]. These rounds of parasitemia reflect the battle between the host immune system and the pathogen's elegant mechanisms of immune evasion.
At the forefront of this battle is the T. brucei cell surface, which is primarily composed of about 107 copies of a single, densely packed Variant Surface Glycoprotein (VSG) [2]. VSG is highly immunogenic, yet T. brucei escapes immune recognition by switching the monoallelic expression of one VSG-encoding gene to another [1], [3]. This process of surface antigen variation is made possible by a genomic repertoire of more than 1000 highly divergent VSG-encoding genes and pseudogenes [4], [5], [6]. Furthermore, existing VSGs can recombine to form novel mosaic VSGs, making the depth of the repertoire potentially limitless [7].
VSGs are encoded throughout the T. brucei genome, which consists of 11 megabase chromosomes, numerous intermediate chromosomes, and ∼100 minichromosomes [6], [8]. Although most VSGs are found in VSG arrays within megabase chromosomes or singly on minichromosomes, they can only be transcribed by bloodstream-form parasites from one of the ∼15 Bloodstream Expression Sites (BES) at a time. Proper expression of VSG on the cell surface is essential for survival [1]. Each BES is composed of a collection of Expression Site Associated Genes (ESAGs), a long repetitive element (70-bp repeats), and a terminal VSG encoding gene, which are transcribed from one upstream promoter and spliced into separate mRNAs before translation [4], [9]. All known BESs are located within 60 kb from the end of the chromosome positioning the expressed VSG within >2 kb of repetitive telomeric DNA [4]. The organization of surface antigen encoding genes in subtelomeric regions is a common theme among pathogens that employ antigenic variation to evade host defenses [10], [11].
There are three predominant mechanisms by which T. brucei VSG switching can occur: In Situ (IS) transcriptional VSG switching — the inactivation of one BES coupled with activation of transcription from a new BES [12], [13], [14], [15], Reciprocal Telomeric Exchange (TE) — a homologous recombination event between two chromosome ends resulting in the balanced transfer of a new VSG to the active BES and the previously active VSG to a silent BES [16], [17], and Duplicative Gene Conversion (GC) — a non-reciprocal transfer of a VSG containing DNA to the active BES that results in loss of the previously active VSG from the genome [18], [19], [20], [21]. GC is predicted to account for the majority of VSG switching under natural conditions because it is the only mechanism that permits the expression of non-telomerically-encoded VSG genes. How each of these mechanisms is orchestrated is largely unknown.
Although the mechanistic basis of each type of VSG switch is unknown, it had long been predicted that GC would be initiated by a DNA break. Recent studies have shown that the artificial induction of a DNA double-stranded break (DSB) proximal to the BES repetitive region upstream of the active VSG increases the frequency of VSG switching by as much as 250-fold, recapitulating the rate estimated in natural isolates [22], [23]. As predicted, VSG switching under these conditions occurred mainly by GC [22]. Furthermore it was shown that DNA breaks accumulate in the repetitive region of the active BES [22]. However, the natural source of DNA breaks that precipitate GC remains a mystery.
A proposed source of GC initiating DSBs is related to the proximity of the BES encoded VSGs to the telomere (usually within >2 Kb) [4], [24]. The actively transcribed BES frequently experiences large stochastic terminal deletions, which are hypothesized to result from the very high levels of transcription at the end of the chromosome [25], [26]. Thus it has been suggested that when the telomere of the chromosome harboring the active BES is short the DNA breaks precipitating telomeric deletions would occur upstream of the VSG, resulting in an antigenic switch [24]. This claim was further correlated to the fact that strains of T. brucei that have been recently isolated from nature, who switch at a rate of approximately 10−2–10−3, have shorter telomeres than laboratory-adapted strains, whose rate of switching can be are 100–10,000-times lower [27], [28], [29], [30]. In all, this suggested an inverse correlation between telomere length and the amount antigenic switching in T. brucei [24], yet there were no data in direct support of this hypothesis.
T. brucei strains with the protein component of telomerase deleted (TERT−/−) are unable to repair telomeric breaks (such as those that occur frequently at the actively transcribed BES) and undergo progressive shortening of all the telomeres in the genome by 3–6 bp/Population Doubling (PD) [31]. Previously it was shown that when the telomere of a TERT−/− isolate is short, the actively expressed VSG is lost over the course of several weeks and replaced by a new VSG gene, which suggested an increase in VSG switching [32]. However, because of the time span in which those experiments took place, that study could not differentiate between the equivalent possibilities of an increase in switching versus the death of short-telomere clones being replaced by a subpopulation of switchers arising at the normal in vitro frequency [32].
In this study we directly tested the proposed correlation between VSG switching and telomere length. Using updated techniques, we compared the frequency of VSG switching between strains with wild-type length or shortened telomeres at the active BES. Large populations of switched clones were analyzed to identify their mechanism of switching and determine if those with short telomeres switch by way of a preferred mechanism. The findings presented here provide experimental support for the hypothesis that telomere length directly affects the frequency of VSG switching, and answer long-standing questions about the relationship between VSG switching, telomere length, and gene conversion.
To address the effect of telomere length on T. brucei antigenic switching, we first isolated strains with various telomere lengths at their active BES (BES1 expressing VSG427-2 [221]). The telomere of BES1 in a population wild-type isolates (WT) can range from ∼10–15 kb (FIG. 1. A). TERT−/− clones with short (∼1.5 kb), medium (∼5.0 kb), and long (>10 kb) BES1 telomeres were isolated and characterized by Southern blot (FIG. 1. B). The active BES telomere in a TERT−/− strain is not only prone to progressive shortening (3–6 bp/PD) but also massive truncations [31]. Thus, medium- and long-telomere clones can only be handled for a minimal number of passages before they shorten (as evidenced by the smear under the primary band in FIG. 1. B – “Long”). In contrast, critically short telomeres are stabilized by an unknown, telomerase-independent, mechanism that appears to be unique to T. brucei [33]. Thus resulting in the short-telomere clone used here (FIG. 1. B – “Short”), which can be stably maintained at a length of ∼1.5 kb for numerous passages.
Populations of trypanosomes, as with any organism, are heterogeneous and this can affect both the expressed VSG and telomere length. Furthermore, accurate determination of the frequency of VSG switching requires that the populations being compared undergo a comparable number of population doublings (PD) during the experiment. Therefore, WT, TERT−/− short- and long-telomere clones with similar growth rates (FIG. S1) were grown from single-cells to ∼5×107 in vitro, thereby performing a modified Luria-Delbrück fluctuation analysis [34] (FIG. 1C). The VSG switching frequency of the resulting populations of trypanosomes was determined using the previously published magnetic-activated cell sorting (MACS) depletion of the initiating VSG followed by flow cytometry quantification [22].
TERT−/− short clone populations switched their expressed VSG at a significantly (P<0.0001) higher frequency (11.3×10−5±4.6) than both WT (1.8×10−5±1.1) and TERT−/− long-telomere clones (2.5×10−5±1.9), which were not significantly different from each other (P = 0.3136) (FIG. 1D). The TERT−/− long-telomere clones serve as a proxy for TERT complementation in this study because a previous study showed that ectopic TERT expression results in rapid elongation of the active site telomere (∼160 bp/PD) [33], which prevents the analysis of a short-telomere TERT complemented clone by this method. The switching frequencies of the TERT−/− short-telomere clones covered a broad range of values (3.2–22×10−5), which correspond to a 6- to 36-fold increase in switching compared to WT. These data might be explained by a stochastic increase in subtelomeric DNA breakage that occurs when the telomere of the active site is short. Alternatively, in the absence of TERT complementation data, the increase in switching could arise from a combinatorial effect of having a short telomere in the context of a telomerase mutant. In either case these data support the notion that subtelomeric breaks promote antigenic switching.
The output of the VSG switching assay represents the proportion of the input population that is no longer expressing the starting VSG type, but this value does not account for replication of progeny, the number of VSG switching events intervening before the final measurement, or potential genotypic diversity of phenotypically identical cells in the resulting population (FIG. 1C). We therefore adopted the nomenclature Observed Switching Frequency (henceforth referred to as switching frequency or OSF) to indicate the limitations of this measured value (FIG. 1D).
Hypothetically, populations that switch more often will contain a greater diversity of expressed VSGs. Thus we predicted that populations with higher OSF values would contain a larger diversity of VSG transcripts. To investigate this prediction, the MACS eluates from 12 of the 18 TERT−/− short-telomere switched populations (FIG. 1D) were grown to a sufficient extent for RNA extraction to identify the expressed VSGs, which were then compared with the determined OSF for that population (FIG. 2A, y-axis values correspond directly with FIG. 1D TERT−/− short data). This type of analysis is only possible due to the extensive nucleotide sequence diversity of VSG-encoding genes, which allows each VSG to be accurately distinguished from another [35]. Although there was a subtle trend of populations with OSF>10 to express a greater diversity of VSGs (∼3–6) than those with an OSF<10 (1–2 VSGs) (FIG. 2B), we did not observe a linear relationship between the switching frequency of these populations and the depth of their VSG diversity (FIG. 2). This is highlighted by the fact that the highest OSF (22×10−5) of this set contained only one expressed VSG (VSG427-3 [224]) (FIG. 2A & 2B).
It is worth noting that the diversity of VSGs identified in TERT−/− short-telomere VSG switching assays was similar to those seen in other studies [1]. VSG427-3 (224) was the most commonly observed VSG, which was present in 10 of the 12 populations analyzed, and the sole VSG expressed in 3 of those populations (FIG. 2A & 2C). The second most commonly expressed was VSG427-1 (060) in 7 populations, followed by VSG427-8 (OD1) in 4 populations (FIG. 2C). These data support previous in vivo and in vitro studies suggesting that certain VSGs are favored donors and that VSG switching follows a semi-predictable order [36]. In addition to the predictably expressed VSGs, we isolated two VSGs that were recently demonstrated to reside on minichromosomes (427-23 & 427-24) [22] and a VSG that had not previously been annotated (FIG. 3, “NOVEL”).
The lack of a clear linear relationship between the switching frequency and the depth VSG diversity (FIG. 2) could arise from two alternative biological situations: extensive propagation of an early switch event (FIG. 1C, top) or multiple switch events that result in expression of the same VSG (FIG. 1C, bottom). As noted, the population of TERT−/− short telomere with the highest OSF (22×10−5) contained only VSG427-3 (224) (FIG. 2A). Did this population arise from one or multiple VSG switch events? To address this question, we isolated TERT−/− short-telomere trypanosomes that had switched from VSG427-2 (expressed from BES1) to VSG427-3 (originating from BES7) from three separate fluctuation analysis experiments. The genotypic differences among isolates from the same starting clone were then compared by PCR and Southern blot analysis (FIG. 3).
Eighteen switchers expressing VSG427-3 were isolated from three populations (3 clones from two populations [#2 & #3] and 12 from the third [#5]). All 18 clones had switched to VSG427-3 by GC (expressed from BES1), as evidenced by the loss of the VSG pseudogene from BES1 (FIG. 3A; BES1 map VSGΨ green box). Using PCR primers unique to ESAG1 [4] in BES1, we determined that VSG427-3 switchers from populations #2 and #5 contained a mixture of clones that had either lost or retained BES1 ESAG1 during GC (2 of 3 in population 2 and 2 of 12 in population 5 had lost ESAG1). Loss of BES1 ESAG1 indicates that resolution of GC occurred upstream of ESAG1. Therefore, clones expressing VSG427-3 in these populations arose from at least two distinct VSG switching events.
To further analyze the genotypes of the VSG427-3 switchers, we used a unique region in BES7 to probe a HindIII-digested large-fragment separation gel and Southern blot (FIG. 3A). There are two HindIII sites in BES1 between the repeat region and ESAG4 that distinguish it from BES7 (FIG. 3A). Thus, GC resolution downstream of ESAG1 results in the formation of a >10 kb fragment, upstream of ESAG1 but downstram of ESAG2 results in a >14 kb fragment, and upstream of ESAG2 but downstream of ESAG4 results in a fragment >24 kb (FIG. 3C). Because BES7 is intact in the genome regardless of switching, all strains produce the >24 kb and ∼2.4 kb fragments. Exact prediction of product sizes for fragments resolved beyond the BES7 repeat region (FIG. 3A, yellow boxes) is not possible due to missing sequence data [4].
Southern blot data largely agreed with PCR data (FIG. 3B), such that the VSG427-3 switchers that retained BES1 ESAG1 by PCR produced the restriction fragment associated with resection downstream of ESAG1 (>10 kb fragment present for #2 = 2/3, #3 = 2/3, & #5 = 3/12 [only partial data shown]). However, one isolate from clone 3 retained ESAG1 but produced a fragment that was predicted to result in the loss of ESAG1 (#3 D2, >14 kb). This combination of data could arise from a recombination event of unpredicted complexity. These results further demonstrate that each population of phenotypically identical VSG427-3 switchers analyzed are composed of at least 2 genotypes, and therefore must arise from multiple genetic events. Fine mapping or sequencing the point of resection for these strains could show further switching diversity, but is not possible due to the >90% sequence similarity between BES1 and BES7 [4]. Therefore, the OSF values displayed in figure 1 under-represent genetic switching frequency by not accounting for the number of VSGs expressed in the population (FIG. 2) and the genetic diversity of the switched population (FIG. 3).
To determine if telomere length affects the mechanism of switching as well as frequency, populations arising from single cells clones of wild-type and TERT−/− short-telomere strains were MACs sorted, their OSF determined (FIG. 4B – using the same methodology used to produced the data in FIG. 1 with new starting populations), and plated to limiting dilution. The secondary clones were then analyzed for the expression of VSG427-2 (221) by high-throughput screening (HTS) flow cytometry, and all cultures not expressing VSG427-2 were deemed “switched clones.” From 12 populations of WT, we screened 1617 secondary clones, identifying 189 WT switched clones (12% of the post-MACS screened population), whereas screening of 255 secondary clones from 6 populations of TERT−/− short-telomere clones resulted in a similar number (188) of switched clones to be further analyzed (74% of the post-MACS screened population). The mechanism of switching for all secondary clones was then determined, based on three criteria: (1) resistance or sensitivity to a BES1 promoter-proximal antibiotic marker (WT marked with hygromycin [HYG] & TERT−/− marked with blasticidin [BSD]), (2) the presence or absence of VSG427-2 (221) in the genome, and (3) the presence or absence of the promoter-proximal resistance marker in the genome. Thus the switch type can be counted in the following way: IS events correspond to MarkerS, 427-2(221)+, Marker+; TE events are MarkerR, 427-2(221)+, Marker+; GC results in MarkerR, 427-2(221)−, Marker+; ES GC (Expression Site Gene Conversion), a subtype of GC in which the entire active BES is replaced by the donor BES, are MarkerS, 427-2(221)−, Marker−; and UD, for Undetermined, when switched clones did not fit the other criteria (FIG. 4A & TABLE S3). These data were also used to establish a minimal number of independent switch events. By creating a matrix based on the 12 known combinations of phenotype and genotype analyzed (VSG427-3 [224] expression as determined flow cytometry [data not shown]), switch type [IS, TE, GC, ES GC or UD], & BES1 VSG Ψ+/−) against the source populations (12 for WT & 6 for Short), we determined that the 189 WT switched clones arose from at least 47 independent switchers and the 188 Short switched clones from at least 25 (TABLE S3). The higher number for WT is an artifact of the increased complexity of its matrix (i.e. more originating populations). Determining the mechanism of switching in TERT−/− long-telomere clones was not possible because of their propensity to break at the active BES, which results in the rapid formation of a heterogenic population of telomere lengths.
WT VSG switched clones produced similar average levels of IS (33%), GC (24%), and TE (37%). This was in contrast to the TERT−/− short-telomere VSG switchers, which showed a clear preference for GC-based switching (88%), only a small amount of IS (7%) and negligible TE (1%) (FIG. 4C [“GC” shown is the sum of both GC and ES GC] & TABLE S1, S2, S3). Switching by GC for a subset of TERT−/− short-telomere clones was further confirmed by pulsed-field gel separation of chromosomes followed by Southern blot analysis using specific VSG probes, which showed both the loss of the initiating VSG (427-2) and the duplication of the newly expressed VSG to the active BES (FIG. 4D). These data, in conjunction with published data that induction of a DNA break in the active site initiates a GC based switch [22], suggest that a short telomere at the active site can increase switch initiating subtelomeric DNA breaks. Further mathematical analysis of switched clone data suggested that the heightened level of GC could account for the overall increase in switching frequency observed when the telomere is short (mathematical validation shown in Experimental Procedures & TABLE S3). The same analytical process showed that the frequency of IS switching (FIS = average OSF×% IS, for both WT and TERT−/− short-telomere clones) was not significantly different between WT and TERT−/− short-telomere clones (TABLE S3). To further analyze the nature of the GC events in WT and TERT−/− short, all isolates identified as GC were assayed for the presence of BES1 VSG pseudogene (PSD or Ψ) in the genome, a parameter that indicates the location of GC resolution. BES1 VSG PSD was lost from TERT−/− short-telomere switchers (90%) ∼30% more often than WT (58%) (FIG. 4C, lined bars “Ψ− GC+”). This indicates that the resolution of GC events in TERT−/− switchers often occurs farther upstream than in WT VSG switchers and suggests that the initiating lesion itself may occur farther upstream when the telomere is short.
The body of isolated switcher data was further analyzed to determine if individual populations, OSF values, or isolation dates, affected the propensity to switch by a given mechanism (TABLE S3, FIG. S2 & S3). Each initiating clone results in a population of secondary switched clones with a somewhat variable percentage of each switching mechanism. Generally, population variations did not affect the results, but there were two WT populations (#2-5 & #2-6) with a large number of switchers (34 & 25, respectively) that had a high percentage of TE (76% & 80%, respectively) (FIG. S2 & TABLE S3), which gave the average WT TE value of 37%. If reanalyzed, the median values for WT were IS = 37%, GC = 20%, and TE = 19%, thus reducing the overall significance of TE. Analysis of the percent switch mechanism in comparison with the switching frequency for each population showed a very subtle trend for populations with higher OSF values to contain a higher level of diversity in the mechanisms represented, which was similarly true for both WT and short-telomere isolates (TABLE S3). Due to different growth rates following limiting dilution, VSG switch isolates were selected 7, 9, or 11 days after initial plating. Analysis of the VSG switch mechanism by isolation date for WT and TERT−/− short-telomere clones displayed a trend of increased IS and decreased GC with time (this pattern was more apparent in WT), while TE did not (FIG. S3). This suggests that isolates that switch by IS initially grow more slowly than those that switch by GC; the basis of this difference is unknown. None of the alternative analysis of the data in this section detract from the central result, namely that TERT−/− short-telomere clones switch preferentially by GC.
Telomeres are structures of DNA and protein that protect the ends of chromosomes from DNA loss and damage. Yet, somewhat counterintuitively, T. brucei is only one of the many pathogens whose critical antigenic diversity genes are organized in potentially fragile subtelomeric regions [10], [37]. In 2007 Dreesen et al. proposed a model in which telomere length (at the active BES) had an inverse correlation with the frequency of VSG switching in T. brucei. The model also predicted that switching at short telomeres would occur by GC, resulting from an increase in BES internal DSBs when the telomere is short [24]. The 2007 proposal was based on a correlation between two main bodies of data: 1) the telomeres of laboratory-adapted strains are longer than those of strains recently isolated from nature [30] & 2) populations of TERT−/− short-telomere strains progressively loose the initially expressed VSG, which suggested, but did not demonstrate an increase in the frequency of VSG switching [33]. Although this model gained popular support, its predictions were unsubstantiated.
Here we have used updated techniques to rigorously test the proposed correlation between telomere length and VSG switching. Our data demonstrate, for the first time, that antigenic switching in T. brucei increases in a TERT−/− mutant when the telomere of the active BES is short (in direct contrast to TERT−/− strains with long a telomere whose switching is like WT) (FIG. 1). Furthermore, we show that the increase in switching can be accounted for by a significant increase in GC (in comparison with other measured switching mechanisms), which is likely due to an increase in DNA breaks in the region upstream of the VSG in the active/short-telomere BES (evidenced by an increased loss of VSG pseudogene [Ψ]) (FIG. 4).
There are two equally valid mechanistic interpretations of these data. The first is that telomere shortening per se leads to an increase in the frequency of DNA breakage in locations that surround the VSG, thus precipitating a switching event, as hypothesized originally by Cross and colleagues [24]. Alternatively, the phenotype we observe might be due to a telomere-capping defect, which would be the result of the combined effect of shortened telomeres in the context of a telomerase-null mutant. Indeed, such a capping defect has been shown to increase gene conversion in the budding yeast Kluyveromyces lactis [38]. To distinguish between these two mechanistic possibilities, we have made concerted efforts to complement telomerase expression in TERT−/− short-telomere T. brucei so that their switching frequency could be compared to un-complemented TERT−/− short-telomere strains. However, we have been unable to set up a strictly regulated inducible system to reconstitute TERT expression while retaining a short telomere at the active expression site (even at exceedingly low expression levels). This is due to the inherent leakiness of inducible systems available for T. brucei, together with the impressive efficiency of catalytically active telomerase to rapidly elongate a short telomere at the active expression site [30], [31]. Also, the impact of telomerase mutations, if any, on t-loop formation and telomere capping in T. brucei is not known at this time. Thus, at the moment, we cannot distinguish between the possibility that the phenotype we observe in the TERT−/− short-telomere strains is due to a direct effect of telomere shortening, as originally proposed, or a capping defect (combined with telomere shortening) due to lack of telomerase function, along the lines of K. lactis [38]. Nevertheless, the general appropriation of conserved mechanisms of chromosome end protection (breakage, lengthening and capping) for the purposes of increased antigenic variation is an intriguing possibility as an aspect of antigenic variation in T. brucei.
T. brucei strains that have been recently isolated from nature are distinct from their laboratory propagated cousins in that they switch more frequently (approximately 10−2–10−3) [27], have shorter telomeres [30], and preferentially switch by GC [39]. Thus, it would appear that by producing a T. brucei strain with an artificially shortened telomere we have, at least partially, recapitulated these characteristics of a natural isolate. The 10- to 100-fold increase OSF measured in the TERT−/− short-telomere strain does not cover the approximately 100- to 10,000-fold difference in the rate of switching published between recent isolates (approximately 10−2–10−3) and laboratory-adapted strains in vitro (approximately 10−5–10−6) [28], [29]. Comparative analysis of switch rate approximations has been a central challenge in the field due to significant variability among the methodologies used and the very small sets of VSG switchers (often less than 10) from which these rate approximations were often derived [27], [28], [29]. The OSF value is simply a measurement of the proportion of cells in the population that are no longer coated in the parental VSG, which does not account for the phenotypic diversity of VSGs in the population (which was included in previous approximate rate derivations) (considered in FIG. 2) or the genotypic diversity (for which no method has ever accounted) (considered in FIG. 3). The findings presented here suggest that the shorter telomeres of natural isolates could contribute to their comparatively high level of VSG switching.
The question that persists is, how naturally occurring populations keep their telomeres short? In addition, T. brucei telomeres grow by 6–8 bp/PD during growth under laboratory conditions (a process that has only been observed in this organism) [25], [30]. What is distinct between the natural T. brucei lifecycle and their laboratory propagation that could account for the observed differences in telomere lengths? Although telomerase regulation in T. brucei is not understood, in other organisms telomerase is activity is regulated by the cell cycle such that telomeres only lengthen in the transition from S phase to G2 [40], [41], [42]. If this were also the case for trypanosomes, the organisms increased in vivo growth rate (which is more than 2 times higher than in vitro) could result in a reduced duration of telomerase activity and thus progressively shorter telomeres. An alternative possibility is that a regulated component of telomere structure or stability, or possibly telomerase itself, is affected when trypanosomes are not permitted to undergo their natural life cycle, which includes passage through the tsetse. In support of this hypothesis, laboratory adapted T. brucei can recover its in vivo switch rate following passage through the tsetse [29]. Although little is known about the regulation of telomerase and other telomere-associated proteins in trypanosomes, perhaps this is a missing connection between the switching behavior of natural isolates and extensively adapted T. brucei strains.
Cell lines were generated from Lister427 bloodstream-form trypanosomes derived from the ‘single marker’ line [43], with a hygromycin resistance marker at the BES1 promoter [22] (“wild-type” [WT]), or homozygous telomerase (Gene ID: 3664223 & protein accession: XP_829083) deletion mutant with blasticidin resistance marker at the BES1 promoter (TERT−/−) [31], [32]. TERT−/− short and long active-site telomere BES clones were isolated from single cell cultures of TERT−/− and telomere lengths were determined by Southern Blot analysis (below) [31]. Trypanosomes were cultured in vitro in HMI-9 medium at 37°C [44].
DNA restriction fragments were separated by standard agarose gel electrophoresis (1–15 kb), Field Inversion Gel Electrophoresis (FIGE) [45] (1–25 kb, BIORAD “Program 1”), Rotating Agarose Gel Electrophoresis [46] (“Classical Program” for separation of megabase, intermediate, and minichromosomes), using published methods. Southern blots were produced using established methods of capillary blotting by neutral transfer (GE Scientific). DNA probes were made by PCR amplification using previously published primer sequences [4], [22], 32P radiolabeled using Prime-It II Random Labeling Kit (Stratagene), and purified over G-50 microcolumns (GE Healthcare). Blots were probe-hybridized, washed, and visualized by phosphorimaging as described (GE Healthcare).
Strains expressing VSG427-2 were single cell cloned by limiting dilution and expanded to a total of ∼5×107 cells prior to MACS isolation and flow cytometry-based quantification of VSG switching frequency as previously described [22]. The VSG switching frequency, here termed the “Observed Switching Frequency (OSF)”, is a direct measure of the proportion of living (measured by propidium iodide staining) trypanosomes in a population that no longer express VSG427-2 on their surface, compared to the total input population, both of which are normalized to a control sample (CountBright Beads purchased from and used according to Invitrogen instructions).
Following MACS depletion of VSG472-2 expressing cells, half of the effluent was removed prior to OSF determination by flow cytomertry and grown in 50 mL HMI-9 with pen-strep until confluent. RNA was extracted, cDNA synthesized, VSG RT-PCR amplified and subcloned (pGEM-T Easy, Promega) using established methods (tryps.rockefeller.edu/trypsru2_protocols_index.html, “VSG cloning for mRNA”), prior to sequencing and expressed VSG determination by NCBI BLAST.
Following MACS depletion of VSG427-2 expressing cells (from single cell cultures grown for the specific purpose of VSG427-3 switcher isolation and not OSF determination), the effluent was labeled with anti-224 antibody and bound to a second MACS column. Following standard MACS binding and wash steps the bound trypanosomes (anticipated VSG427-3+ switchers) were plunged from the column, cloned and screened for VSG427-3 expression by flow cytometry. VSG427-3 expressing clones were grown for genomic DNA isolation [46], which was analyzed by PCR using BES1 VSG pseudogene primers (pseudo5-F: 5′- GCGCCGAATTTAATGCAATATGCAACG & pseudo5-R: 5′- GCAGGCCGTCTTTTGAGTTGTAGTAAG) & BES1 ESAG1 primers (ESAG1 Fw1: 5′-GAGCAAACTGATAGGTTGGAAAAG & ESAG1 Rv1 5′-GCACTGGCGGCCACTCCATTGTC) and HindIII FIGE Southern blot analysis using a BES7 specific probe (FIG. 4A – red bar) produced by PCR using unique primers (BES7 UR1 Fw: 5′-GCAACTAACTACTGTAATTCCC & BES7 UR Rv: 5′-GCTACTAATGTGTTTCAATATGCG).
Following expansion of VSG427-2 (221) expressing WT and TERT−/− short-telomere cells from single cells to ∼5×107 total cells and MACS depletion of VSG472-2 expressing cells as described above [22], the effluent was split into two samples. Half of the resulting cells were used to determine the OSF and the other half was cloned by limiting dilution in 96-well tissue-culture plates. Clone growth was observed for two weeks following initial plating. At 7, 9, and 11 (WT only) days after plating, confluent cultures were split into new 96-well plates and allowed to become confluent. Final identification of VSG427-2− switchers was performed by high-throughput-screening (HTS) flow cytometry using anti-VSG221 antibody on an LSRII with HTS adaptor for 96 well plate analysis (BD Biosciences). VSG427-2 negative clones were then analyzed for BES1 promoter activity by antibiotic selection of the BES1 promoter proximal marker (WT: hygromycin & TERT−/−: blasticidin) and the Southern dot blots for the presence of VSG427-2 (221), the promoter proximal antibiotic resistance gene and BES1 VSG pseudogene in the genome (using radiolabeled probes to VSG221, hygromycin, blasticidin, pseudogene). Southern dot blot was adapted for trypanosomes from manufacturers protocols (GE Healthcare) by adding ∼1×106 cells/well to the membrane, cell lysis with denaturation buffer, neutralization, and fixation with UV prior to radiolabeled probe hybridization, washing, and visualization as described above for Southern blot analysis. Primers for PCR production of Southern dot blot probes: 221 Prb Fw (5′- GTAACAGCTACTGCAACAGCGAGC)& 221 Prb Rv(5′- GCTTCTTCAACAAGCTTGGTAACG), HYGRO Prb Fw (5′-GCTCTCGATGAGCTGATGCTTTGG) & HYGRO Prb Rv (5′-GATAGAGTTGGTCAAGACCAATGC), BES1 PSEUDO Fw (5′- CATTAAATTCAAGCGTCTAGACCGCAGC) & BES1 PSUEDO Rv (5′- GCGCGTTGTTCCGTATCTGCTGAGC)
If ΣFShort≈ΣFWT+ΔFGC, then GC accounts for the increase in short-telomere OSF. Where FMech = Average OSF×% Mechanism and ΔFGC = FGC,Short−FGC,WT. First the F for each mechanism for both WT and Short were determined, the sum of which is equal to the average OSF for each strain (ΣFWT = 1.38 & ΣFShort = 8.84). Then the ΔFGC was determined and used to show that ΣFWT+ΔFGC = 8.86 which is fundamentally identically to the ΣFShort (8.84), thus making ΣFShort≈ΣFWT+ΔFGC a valid statement (TABLE S3).
|
10.1371/journal.pgen.1001199 | Localization of a Guanylyl Cyclase to Chemosensory Cilia Requires the Novel Ciliary MYND Domain Protein DAF-25 | In harsh conditions, Caenorhabditis elegans arrests development to enter a non-aging, resistant diapause state called the dauer larva. Olfactory sensation modulates the TGF-β and insulin signaling pathways to control this developmental decision. Four mutant alleles of daf-25 (abnormal DAuer Formation) were isolated from screens for mutants exhibiting constitutive dauer formation and found to be defective in olfaction. The daf-25 dauer phenotype is suppressed by daf-10/IFT122 mutations (which disrupt ciliogenesis), but not by daf-6/PTCHD3 mutations (which prevent environmental exposure of sensory cilia), implying that DAF-25 functions in the cilia themselves. daf-25 encodes the C. elegans ortholog of mammalian Ankmy2, a MYND domain protein of unknown function. Disruption of DAF-25, which localizes to sensory cilia, produces no apparent cilia structure anomalies, as determined by light and electron microscopy. Hinting at its potential function, the dauer phenotype, epistatic order, and expression profile of daf-25 are similar to daf-11, which encodes a cilium-localized guanylyl cyclase. Indeed, we demonstrate that DAF-25 is required for proper DAF-11 ciliary localization. Furthermore, the functional interaction is evolutionarily conserved, as mouse Ankmy2 interacts with guanylyl cyclase GC1 from ciliary photoreceptors. The interaction may be specific because daf-25 mutants have normally-localized OSM-9/TRPV4, TAX-4/CNGA1, CHE-2/IFT80, CHE-11/IFT140, CHE-13/IFT57, BBS-8, OSM-5/IFT88, and XBX-1/D2LIC in the cilia. Intraflagellar transport (IFT) (required to build cilia) is not defective in daf-25 mutants, although the ciliary localization of DAF-25 itself is influenced in che-11 mutants, which are defective in retrograde IFT. In summary, we have discovered a novel ciliary protein that plays an important role in cGMP signaling by localizing a guanylyl cyclase to the sensory organelle.
| C. elegans mutants that either fail to form or arrest development as dauer larvae, a stress-resistant lifestage, usually have defects in genes involved in evolutionarily conserved signaling pathways. In this study, we identified the gene mutated in daf-25 mutant strains, which inappropriately arrest as dauer larvae and are also defective in the sense of smell. The mammalian counterpart of DAF-25 is Ankmy2, a protein of unknown function that contains three ankyrin repeats and a zinc finger MYND domain, both of which are predicted to bind other protein(s). We show that DAF-25/Ankmy2 is required for the proper localization of a membrane-bound guanylyl cyclase—a class of protein that functions in cyclic GMP signaling—to cilia, which are conserved sensory organelles. We further demonstrate that mammalian Ankmy2 binds the retinal guanylyl cyclase GC1, suggesting a role for Ankmy2 in vision—which critically depends on cyclic GMP signal transduction—suggesting the potential involvement of Ankmy2 in human retinal disease, as well as other cilia-related diseases such as obesity.
| The dauer larva of Caenorhabditis elegans is an alternate third larval stage where a stress resistant, non-aging life plan is adopted in harsh environmental conditions [1]. Dauer larvae disperse and will resume development when conditions improve. The study of dauer formation has elucidated a complex gene network used to control the decision to go into diapause [2]. The dauer pathway includes well-recognized members in the canonical TGF-β (Transforming Growth Factor-Beta) and Insulin/Insulin-like signaling (IIS) pathways, as well as proteins affecting olfactory reception, neuron depolarization and peptide hormone secretion. Many mutants isolated as dauer formation defective (Daf-d) or constitutive (Daf-c) have revealed the key signaling components [2]. Here we identify DAF-25, a novel member of the olfactory signaling pathway that is associated with cGMP signaling—a signal transduction pathway with established links to cilia [3]. We show that the mammalian ortholog, Ankmy2, is expressed in ciliary photoreceptors and interacts with a guanylate cyclase (GC1), as predicted from the C. elegans results.
The olfactory signaling cascade has been well characterized in the two C. elegans amphids, organs consisting of a set of twelve bilaterally symmetric pairs of ciliated sensory neurons [4], [5]. While similar to mammalian olfactory signaling, at least some proteins involved are also homologous to those implicated in mammalian phototransduction [6]. Chemicals are sensed at the afferent, ciliated ends of sensory neurons where they contact the environment through pores in the cuticle. The cilia are required for chemosensation of chemical attractants and repellants, as well as for dauer entry and exit [7]. For many odorants the specific neurons that detect the odor are known [4]. For example, the AWA, AWB and AWC neuron pairs sense volatile odorants such as pyrazine, benzaldehyde, trimethyl thiazole and isoamyl alcohol. The ASH pair of ciliated olfactory neurons can detect changes in osmotic pressure.
The connection between dauer formation, chemosensory behavior and cilia is well known [2], [8]. C. elegans hermaphrodites only possess non-motile (primary) cilia which are found at the dendritic ends of 60 sensory neurons in the head and tail [5], [8]. Intraflagellar transport (IFT) proteins, normally required for building cilia, are well conserved in C. elegans and several have been discovered in this organism through the identification of sensory mutants [9]. Indeed, dauer formation is a sensory behavior dependent on the balanced inputs of dauer pheromone, temperature and food signals [4].
Proteins in the olfactory component of the dauer pathway include SRBC-64 and SRBC-66 (dauer pheromone receptors), DAF-11, a guanylyl cyclase, G-proteins (gpa-2 and gpa-3), the Hsp90 molecular chaperone DAF-21, the IFT protein DAF-10, and the DAF-19 RFX-type transcription factor [10]–[14]. DAF-19 is strictly required for cilium formation as it regulates the expression of many cilia-related genes through a consensus sequence dubbed ‘x-box’ [15]. daf-11, daf-19 and daf-21 are Daf-c, whereas daf-6 and daf-10 are Daf-d [16]. daf-19, daf-6 and daf-10 are all dye-filling defective, indicating that their cilia (if present) are not exposed to the environment [17], [10]. By contrast, daf-11 and daf-21 mutants show wild-type dye filling [18]. All five mentioned daf genes are defective in recovery from the dauer diapause, presumably because they cannot detect the bacterial food stimulus [17]. Dauer recovery defects are present for mutants with broad chemosensory defects caused by abnormal ciliogenesis or signaling, and for many Unc genes, such as unc-31, which encodes a dense core vesicle secretion protein [17], [19], [20]. Our genetic screen for C. elegans Daf mutants has uncovered a novel ciliary protein, DAF-25, which participates in cGMP-associated signaling by modulating the ciliary localization of a guanylyl cyclase, DAF-11. The mammalian ortholog of DAF-25, Ankmy2, interacts with ciliary photoreceptor guanylyl cyclase 1 (GC1), indicating that the role of the MYND domain protein in cilia function is likely to be conserved and potentially relevant to human retinal disease or other ciliopathies.
To identify genes potentially implicated in sensory transduction, we uncovered four alleles of daf-25 in various screens for new mutants exhibiting a temperature-sensitive Daf-c phenotype. Three alleles (m98, m137, and m362) were isolated from ethyl methanesulfonate (EMS) mutagenesis screens and the fourth, m488, was isolated in a screen for Daf-c mutants with transposon insertions [21], [22].
Epistasis tests with the Daf-d mutants daf-12, daf-16, daf-3, daf-6 and daf-10 were used to position daf-25 into the existing genetic pathway. Mutations in the daf-12 nuclear hormone receptor gene suppress most Daf-c mutants [16], [23] including daf-25 (0% dauer larva formation, n>200 for daf-25(m362); daf-12(m20) compared to 97.5%, n = 281 for daf-25(m362) at 25°C). DAF-16/FOXO is the major downstream effector for Insulin/IGF1 signaling [24] as is DAF-3/Co-Smad for the TGF-β pathway [25]. Mutations in daf-16 and daf-3 only partially suppress the Daf-c phenotype of daf-25 (37.6% dauer larvae, n = 407 for daf-25(m362); daf-16(m26) and 60.0%, n = 167 for daf-25(m362); daf-3(mgDf90) at 25°C), indicating that DAF-25 likely functions upstream of both pathways.
Importantly, daf-10, which encodes an IFT protein (DAF-10/IFT122) required for ciliogenesis [11], suppresses daf-25 (0% dauer larvae, n>200 for daf-25(m362); daf-6(e1387) compared to 97.5%, n = 281 for daf-25(m362) at 25°C), suggesting a function for DAF-25 within sensory cilia. daf-6 mutants have closed amphid channels and cannot smell chemoattractants or form dauer larvae even though their cilia are present [5]. Interestingly, daf-6 mutations do not suppress the daf-25 Daf-c phenotype (97.4% dauer larvae, n = 312 for daf-25(m362); daf-6(e1377) at 25°C), indicating that DAF-25 acts downstream of DAF-6, and that environmental (ciliary) input is not required for the Daf-c phenotype. DAF-6/PTCHD3 is expressed in the glial (sheath) cell that forms the amphid sensory channel, allowing contact of the sensory cilia to the environment through pores in the cuticle [26]. 8-bromo-cGMP rescues the dauer phenotype of daf-25 (0% dauer larva formation for daf-25(m362) on 8-bromo-cGMP, n = 72 compared to 32% dauer larva formation on the control, n = 65, both at 20°C), similar to that previously reported for daf-11 [12] indicating that DAF-25 functions upstream of the cGMP pathway in the cilia. Indeed the Daf-c phenotype of daf-25(m362) is very similar to that of daf-11(m84) at all temperatures tested (Table S1). The epistasis results are also similar to those for daf-11, indicating that both genes function at the same point in the genetic pathway—upstream of cilia formation and cGMP signaling in the cilia, and downstream of environmental input.
daf-25 mutants are temperature-sensitive Daf-c and defective in dauer recovery. They constitutively form virtually 100% dauer larvae at 25°C, which do not recover upon transfer to 15°C. The Daf-c phenotype is rescued by maternally contributed daf-25 as seen in the progeny of daf-25(m362) heterozygous hermaphrodites which form zero percent dauer larvae at 25°C (n>200). Moreover, daf-25 animals exhibit defective responses to various chemosensory stimuli as well as a moderate defect in response to osmotic stress (37 of 45 daf-25(m362) adults crossed the sucrose hyperosmotic boundary compared to 1 of 45 for N2, χ2-p-value = <0.00001, while 0 of 30 daf-25(m362) and N2 adults crossed a glycerol boundary). Adults are also defective in egg laying. Despite the fact that Daf-c genes in the IIS pathway (like daf-2 and age-1) extend adult lifespan [27], daf-25 mutants show no significant difference in lifespan from N2 (Figure S1).
daf-25 mutants are defective in chemotaxis to at least four volatile odorants (Figure 1). Wild-type N2 adults were attracted to the compounds tested, the chemotaxis-defective mutant daf-11 was partially attracted, whereas the two daf-25 mutants tested were nearly unresponsive (Figure 1). DAF-11 and the cGMP pathway are known to regulate responses to the AWC neuron-mediated odors isoamyl alcohol, trimethyl thiazole and benzaldehyde, and our results indicate that DAF-25 is also required in this pathway [28]. The AWA-detected scent, pyrazine, is not reported to be detected by the cGMP pathway, suggesting that DAF-25 participates in another signaling pathway in AWA neurons. Interestingly, alhough it has been shown that the cGMP pathway does not participate in AWA-mediated olfaction, the particular tested allele daf-11(m47) was previously shown to have reduced affinity for pyrazine [28], as we have seen here.
To establish if the olfactory phenotypes are associated with ciliary defects, mixed-stage populations of daf-25 mutants and N2 were stained with the lipophillic dye, DiI. Mutants with cilia structure anomalies have abrogated dye filling of the olfactory neurons [29], whereas daf-25 mutants take up the dye normally at all ages, suggesting that they have structurally intact cilia (Figure S2). To confirm this possibility, we further examined the integrity of ciliary structures by transmission electron microscopy. Ciliary ultrastructures in two daf-25(m362) L2 larvae—including transition zones, middle segments consisting of doublet microtubules, and distal segments composed of singlet microtubules—was indistinguishable from the two N2 controls (Figure S3). We conclude that daf-25 animals have no obvious defects in ciliogenesis or cilia ultrastructure.
To identify the daf-25 genetic locus, we first used three-factor genetic crosses to map the m362 allele to the left arm of Chromosome I. Then, we employed a modified SNP mapping procedure [30], in which we selected for recombinants in the unc-11-daf-25 interval to map daf-25 to the left-most 1 Mbp of Chromosome I. Finally, we used a custom-made high-density array for the left-most ∼2.5 Mbp for comparative genomic hybridization (CGH). Two molecular lesions in daf-25(m362) were identified in exon 4 of Y48G1A.3 (Figure 2), including a 31 bp deletion and a G>A change 72 bp to the right of the deletion. Subsequent sequencing of PCR products from mutant genomic DNA uncovered the lesions in the remaining alleles. The m98 mutant has a 996 bp deletion that removes the first two exons, m137 has an ochre stop in the fourth exon, and m488 has a Tc1 transposon insertion in the third exon (Figure 2).
Y48G1A.3 encodes the C. elegans ortholog of mammalian Ankmy2 (by reciprocal BLAST), a protein with three ankyrin repeats and a MYND-type zinc finger domain. The C. elegans protein shares throughout its length (388 residues) 52% similarity and 32% identity with mouse Ankmy2 (440 residues). The C. elegans ankyrin repeat domain is 40% identical and the MYND domain is 55% identical to the murine ortholog. Ankmy2 is very well conserved among chordates, with identity percentages compared to human Ankmy2 of 99% for macaque, 93% for cow, 88% for mouse, and 76% for zebrafish (Figure S4). Although the protein is highly conserved, there is no reported functional data for this gene from any organism. The MYND domain is thought to function in protein-protein interactions, although only a small number of MYND domain-containing proteins have been characterized, including the AML1/ETO protein, which binds SMRT/N-CoR through its MYND domain [31].
To analyze the transcript(s) generated by the daf-25 gene, we employed a PCR-based approach. Using primers for the SL1 transplice sequence or poly-T in combination with gene-specific primers, we were able to amplify only one isoform (Figure S5). This result is consistent with the RNA-Seq and trans-splice data found on Wormbase, which shows a daf-25 transcript sequence identical to that presented in Figure S5, including the 5′ and 3′ UTRs [32]–[34]. We were unable to amplify an SL2 trans-spliced product using multiple gene specific primers and an SL2 primer under any conditions tested.
To determine the sub-cellular localization of DAF-25, a GFP-tagged protein was constructed. The daf-25 upstream promoter (approx. 2.0 kb 5′ of the ATG) was fused to the daf-25 cDNA in frame into the pPD95.77 vector (gift from Dr. Andrew Fire) containing GFP (without a nuclear localization signal) and the unc-54 3′UTR. This construct was found to be expressed in many ciliated sensory neurons, including the following pairs of anterior neurons: AFD, ASK, ASI, ASH, ASJ, ASG, ASE, ADF, AWA, AWB, AWC and IL2 (Figure 3). It is also expressed in the PQR ciliated neuron and one ventral interneuron. We also show expression of the DAF-25::GFP construct in the 7A ciliated neuron in the male tail though we did not fully examine male expression due to the limited number examined and the mosaic expression associated with extra-chromosomal arrays. Most importantly, the fluorescence of the GFP-tagged protein was localized to the cilia of all these cells. The GFP-fusion construct was judged to be functional because it fully rescued the Daf-c phenotype of daf-25(m362) at 25°C while non-transgenic siblings arrested as dauers (n>200).
To investigate whether the ciliary localization of DAF-25 might depend on the intraflagellar transport (IFT) machinery, the DAF-25::GFP construct was crossed into che-11, which is required for retrograde transport in the cilia. In che-11 mutants, IFT-associated proteins accumulate in the cilia [35]. The DAF-25::GFP translational fusion protein accumulated within the cilia and basal body (base of cilia) despite a reduction in total GFP fluorescence (mean DAF-25::GFP fluorescence in che-11 (8.7E12) compared to N2 (1.4E13), p<0.00001, n = 9 for both), suggesting that the protein is associated with IFT (Figure 4). To test for a possible role for DAF-25 in the core IFT complex, GFP translational fusion constructs of two IFT proteins, CHE-2 and CHE-11 [30], were crossed into the daf-25(m362) mutant background and analyzed by time-lapse microscopy. The velocities of IFT transport of CHE-2 and CHE-11, as determined by kymograph analysis, were unchanged in daf-25 compared to that of wild type animals (Figure 4). Specifically, transport velocities in the middle segment were ∼0.7 µm/s, and in the distal segments ∼1.2 µm/s, exactly as reported for all studied IFT proteins [36]. Collectively, our data show that DAF-25 is not essential for IFT, and is therefore unlikely to be a core component of IFT transport particles—consistent with the findings that the ciliary ultrastructure of the daf-25 mutant is intact (Figure S3). However, its accumulation within cilia in the retrograde IFT mutant does suggest that it is associated with (i.e., transported by) the IFT machinery.
The phenotype of daf-25 is most similar to that of daf-11, and our epistasis results placed daf-25 at the same position in the genetic pathway previously reported for daf-11 [37]. To test for possible functional interactions, a strain harboring DAF-11::GFP (gift from Dr. James Thomas), which is known to localize to cilia [12], was crossed with two daf-25 mutants (m98 and m362). In wild-type animals, the DAF-11::GFP protein localized to the sensory cilia of the olfactory neuron pairs ASI, ASJ, ASK, AWB and AWC (Figure 5A), all of which express DAF-25::GFP (Figure 3). In both daf-25 mutants, the DAF-11::GFP protein was observed only in a region near the base of cilia, rather than along their length (Figure 5B). To assess more precisely where the DAF-11::GFP protein is mislocalized, we introduced into the same strain a ciliary (IFT) marker, namely tdTomato-tagged XBX-1 (a gift from Dr. B. Yoder), which localizes at basal bodies and along the ciliary axoneme [38]. Visualization of the two fluorescently-tagged proteins in the daf-25 mutant revealed that DAF-11::GFP accumulates at the very distal end of dendrites, with little or no localization to the basal body-ciliary structures (Figure 5E). This indicates that DAF-25 is required for the proper localization of DAF-11 to the cilia, providing a likely explanation for the similarities between the daf-11 and daf-25 mutant phenotypes. To test if the DAF-25-DAF-11 functional interaction is specific, GFP-tagged ciliary channel proteins (OSM-9/TRPV4 and TAX-4/CNGA1) and IFT-associated proteins (CHE-2/IFT80, CHE-11/IFT140, CHE-13/IFT57, BBS-8/TTC8, OSM-5/IFT88 and XBX-1/D2LIC) were also crossed into the daf-25(m362) mutant background. All eight reporters showed normal localization to the olfactory cilia in the wild-type N2 and daf-25(m362) strains, indicating the possible specificity of DAF-25 for guanylyl cyclases (OSM-9::GFP localization in the daf-25 mutant shown in Figure 5C and 5D; the remaining constructs are presented in Figure S6). The mislocalization of DAF-11::GFP in daf-25(m362) was not suppressed by daf-12(sa204) (Figure S7). This indicates that it is the abrogation of DAF-25 rather than entry into dauer that controls the ciliary localization of DAF-11.
The GFP reporter results suggest a potentially specific function for DAF-25 in cilia. This finding is consistent with the reported regulation of daf-25 by the ciliogenic DAF-19 RFX-type transcription factor [39]. Taken together, DAF-25 appears to be an adaptor protein required for the transport or tethering of the guanylyl cyclase DAF-11 within sensory cilia.
To ascertain if a functional association between DAF-25/Ankmy2 and guanylyl cyclase is evolutionarily conserved, we used a pull-down experiment to test whether mouse Ankmy2 interacts with the retinal-specific guanylyl cyclase GC1, a mammalian homolog of DAF-11 present within ciliary photoreceptors. We amplified Ankmy2 cDNA from a mouse retinal cDNA preparation (gift from Simon Kaja), and constructed a cDNA clone with the rhodopsin 1D4 epitope to use for co-IP experiments with anti-1D4 monoclonal antibody [40]. We co-expressed both in HEK293 cells to test for GC1 co-immunoprecipitation with the 1D4 epitope-tagged Ankmy2 (HEK293 cells do not express rhodopsin). Pull-down of Ankmy2 co-precipitated GC1, but not another control protein (retinal membrane protein ABCA4; Figure 6). This indicates that the functional interaction between DAF-25/Ankmy2 and guanylyl cyclase observed in ciliated sensory cells may be conserved between mouse and worm.
In this study, we have identified in a genetic screen for Dauer formation mutants a novel MYND domain-containing ciliary protein, DAF-25, that is required for the proper localization of a guanylyl cyclase (DAF-11) to sensory cilia. Disruption of DAF-25 does not interfere with intraflagellar transport (IFT) or ciliary ultrastructure, but the protein accumulates in a che-11 retrograde IFT mutant. We therefore propose that DAF-25 is associated with IFT not as a ‘core’ protein but instead as an adaptor for transporting ciliary cargo. In our model, abrogation of DAF-25 would thereby not allow transport of DAF-11, which explains the improper localization of DAF-11 in daf-25 mutants at the very base of cilia and the similarity in phenotype between daf-11 and daf-25 mutants.
The amino acid sequence and domain structure similarity between DAF-25 and Ankmy2 suggests an important function for the latter mammalian protein that may be similar to DAF-25 in C. elegans. We attempted to co-immunoprecipitate DAF-25 and DAF-11 in C. elegans but were unable to satisfactorily remove a sufficient amount of background proteins to avoid confounding any identified interaction (data not shown). We also showed that the retinal guanylyl cyclase GC1 binds to Ankmy2, and we propose that the functional relationship between DAF-25 and DAF-11 is conserved between Ankmy2 and GC1 in ciliated photoreceptor cells. Indeed, Ankmy2 may be required for the transport of not only GC1 but perhaps other cilia-targeted guanylyl cyclases as well as other cilia-targeted proteins in mammals. Further studies will be required to experimentally confirm whether Ankmy2 is required for transport of GC1 to the rod outer segment, and to test if Ankmy2 lesions result in retinal disease or a ciliopathy syndrome that includes retinopathies. Mutations in ciliogenesis and cilia related genes cause human disease phenotypes including Bardet-Biedl syndrome, retinopathies, obesity, situs inversus and polycystic kidney disease, among others [41], [42]. Interestingly, GC1 and the nuclear hormone receptor Nr2e3 shown to regulate Ankmy2 expression in mouse retina both harbor mutations in patients with retinal disease [43], [44].
While this research was being conducted we became aware of another group that cloned and characterized chb-3 (Y48G1A.3/daf-25/Ankmy2) as a suppressor of the che-2 body size phenotype [45]. Fujiwara et al., (in press) describe the cloning of chb-3/daf-25 and its essential role in GCY-12 cilia localization. They show that DAF-25 is required in a subset of sensory neurons to rescue the phenotypes they assayed (dauer formation and body size) using a tax-4 promoter. This indicates that DAF-25 function is required in the neurons where cGMP signaling takes place (TAX-4 is a subunit of cGMP-gated calcium channel). They also show expression of DAF-25 in the ASJ neurons (one pair of neurons where DAF-11 is expressed) is required for rescue of the dauer phenotype, also indicating a cell autonomous role for DAF-25. It is interesting that screens for the Daf-c and Chb (che-2 body size suppressor) phenotypes both resulted in the identification of daf-25/chb-3 and separately identified its apparent ciliary cargos daf-11 and gcy-12, guanylyl cyclases that specifically work in dauer formation and body size, respectively. This indicates that DAF-25/CHB-3/Ankmy2 may interact with cilia-targeted guanylyl cyclases in a general manner and that much of the phenotype of daf-25/chb-3 mutants reflects a global defect in cGMP signaling, potentially along with other unidentified cargo proteins.
In conclusion, our findings uncover a novel ciliary protein that plays an important role in modulating the localization/function of cGMP signaling components, which are known to play a critical role in the function of ciliary photoreceptors [46]. DAF-25/Ankmy2 may also play a role in the ciliary targeting of other as of yet identified proteins. As such, Ankmy2 could participate in phototransduction and be associated with retinopathies, and more generally, could be implicated in other ciliary diseases (ciliopathies).
daf-25 mutations were created by treatment of N2 with 0.25 M EMS, or by mut-2 transposon mobility, and selection for constitutive dauer formation as previously described [22]. For 3-factor mapping, fog-1(e2121) unc-11(e47) was crossed with daf-25(m362) and daf-25(m362) unc-35(e259) was crossed with dpy-5(e61). Scoring the genotypes of the F2 progeny required the phenotyping of F3 progeny (due to the maternal effect of the daf-25 dauer phenotype). Pooled SNP mapping was completed as previously described [30] with some changes. In the Po generation, CB4856 males were crossed to daf-25;unc-11 double mutant hermaphrodites. The F1 males were crossed with CB4856 hermaphrodites. F2 hermaphrodites were selected by absence of Unc progeny. F3 hermaphrodites were placed one to a plate and were selected into wild type or mutant pools based on absence or presence of dauers in the F4. Wild type and mutant pools of F3 hermaphrodites were subject to SNP analysis as previously described [30]. ArrayCGH was done as previously described [47] for the leftmost 2.4 Mbp of Chromosome I with 50 base probes spaced every four base pairs.
Epistasis analysis was performed by crossing daf-25(m362) into daf-12(m20), daf-16(m26), daf-3(mgDf90), daf-10(e1387) and daf-6(e1377). Once the double mutants were isolated, the dauer phenotype was assayed to determine if daf-25 was suppressed fully (no constitutive dauer larvae formed at 25°C), partially (fewer dauer larvae than daf-25(m362) control) or no suppression. Treatment with cGMP was performed as previously described [12] with 5 mM 8-bromo-cGMP (Sigma). Neuronal dye-filling was assayed by incubating a mixed-stage population of each genotype in Vibrant DiI (Molecular Probes) 1000-fold diluted in M9 buffer for one hour followed by washing in M9 and one hour destaining on plates. Chemotaxis assays were performed synchronized day-1 adults as previously described with the volatile attractants trimethyl-thiazole, pyrazine, benzaldehye and isoamyl alcohol [48].
The DAF-25::GFP construct was created by inserting the 2.0 kb promoter region 5′ of the AUG followed by daf-25 cDNA the into the pPD95.77 vector (gift from Dr. Andrew Fire). After microinjection into N2 adults [49] with 10 ug/ml of pRF4 (contains rol-6(su1006)), and 90 ug/ml of DAF-25::GFP plasmid (described above), transgenics lines were established based on the roller phenotype. The extra-chromosomal array mEX179(pdaf-25::DAF-25::GFP, rol-6(su1006)) was crossed into daf-25(m362) and rescue of the Daf-c phenotype was detected by normal non-dauer development in the F3 progeny grown at 25°C. GFP fluorescence was visualized on a Zeiss Axioskop with a Qimaging Retiga 2000R camera.
To measure the integrity of IFT within the daf-25(m362) mutant, kymograph analyses were performed using GFP-tagged CHE-11 and CHE-2 IFT markers. Time-lapse movies were obtained for the different strains, including N2, and kymographs were generated from the resulting stacked tiff images using Metamorph software (Universal Imaging, West Chester, PA). Rates of fluorescent IFT particle motility along middle and distal segments were measured as described previously [35], [50]. To assess how disrupting IFT affects the ciliary localization of DAF-25::GFP, mEX179 was crossed into che-11 mutants and visualized by microscopy essentially as described [35]. Fluorescence intensity was measured by analyzing images in ImageJ by highlighting the entire head region for each animal, then measuring pixel density minus the pixel density for an equal sized adjacent region. The localization of several GFP–tagged proteins in daf-25(m362) animals, namely DAF-11, OSM-9, TAX-4, CHE-2, CHE-11, CHE-13, BBS-8, OSM-5 and XBX-1, were ascertained by crossing the reporter into the mutant, followed by visualization using standard microscopy. Co-localization was carried out by injecting the osm-5p::XBX-1::tdTomato into daf-25(m362);daf-12(sa204) and crossing it into TJ9386 which carries the DAF-11::GFP reporter [12].
Staged N2 and daf-25 L2 larvae were produced by harvesting eggs from gravid adults by alkaline hypochlorite treatment, followed by overnight hatching in M9 buffer, and subsequent incubation of hatched L1 larvae on seeded NGM plates for 26 hours at 16°C. Worms were then washed directly into a primary fixative of 2.5% glutaraldehyde in 0.1 M Sorensen phosphate buffer. To facilitate rapid ingress of fixative, worms were cut in half using a razor blade under a dissecting microscope, transferred to 1.5 ml Ependorf tubes and fixed for one hour at room temperature. Samples were then centrifuged at 3,000 rpm for two minutes, the supernatant removed and the pellet washed for ten minutes in 0.1 M Sorensen phosphate buffer. The worms were then post-fixed in 1% osmium tetroxide in 0.1 M Sorensen phosphate buffer for one hour at room temperature. Following washing in Sorensen phosphate buffer, specimens were processed for electron microscopy by standard methods. Briefly, they were dehydrated in ascending grades of alcohol to 100%, infiltrated with Epon and placed in aluminum planchetes orientated in a longitudinal aspect and polymerized at 60°C for 24 hours.
Using a Leica UC6 ultramicrotome individual worms were sectioned in cross section from anterior tip, at 1 µm until the area of interest was located as judged by examining the sections stained with toluidine blue by light microscopy. Thereafter, serial ultra-thin sections of 80 nm were taken for electron microscopical examination. These were picked up onto 100 mesh copper grids and stained with uranyl acetate and lead citrate. Using a Tecnai Twin (FEI) electron microscope, sections were examined to locate, in the first instance, the most distal (anterior) region of the cilia, then to the more proximal regions of the ciliary apparatus. At each strategic point, distal segment, middle segment and transition zone/fiber regions were tilted using the Compustage of the Tecnai to ensure that the axonemal microtubules were imaged in an exact geometrical normalcy to the imaging system. All images were recorded, at an accelerating voltage (120 kV) and objective aperture of 10 µm, using a MegaView 3 digital recording system.
Mouse ankmy2 cDNA, amplified from retinal RNA, was engineered to contain a sequence encoding a 9 amino acid 1D4 C-terminal epitope as previously described [40]. Ankmy2-1D4 and either human GC1 or the retinal ABC transporter ABCA4 as a control were co-expressed in HEK 293 cell. HEK 293 cell extracts were solubilized in 18 mM CHAPS in TBS (20 mM Tris, 150 mM NaCl, 1 mM EDTA, 1 mM MgCl2 and Complete inhibitor). The solution was stirred at 4°C for 20 minutes and subsequently centrifuged in an Optima TLA100.4 rotor (Beckman) for 10 minutes at 40,000 rpm to remove any residual unsolubilized material. The solubilized extract was applied to an immunoaffinity resin consisting of the Rho 1D4 antibody conjugated to Sepharose 2B [31]. After incubation at 4°C for one hour, the resin was extensively washed with TBS to remove unbound protein, and the bound proteins were eluted with 0.2 mg/ml of the 1D4 competing peptide in TBS for analysis by Western blot labeled with Rho 1D4 antibody for the detection Ankmy2-1D4 and antibodies to GC1 or ABCA4.
|
10.1371/journal.pgen.1002070 | Finished Genome of the Fungal Wheat Pathogen Mycosphaerella graminicola Reveals Dispensome Structure, Chromosome Plasticity, and Stealth Pathogenesis | The plant-pathogenic fungus Mycosphaerella graminicola (asexual stage: Septoria tritici) causes septoria tritici blotch, a disease that greatly reduces the yield and quality of wheat. This disease is economically important in most wheat-growing areas worldwide and threatens global food production. Control of the disease has been hampered by a limited understanding of the genetic and biochemical bases of pathogenicity, including mechanisms of infection and of resistance in the host. Unlike most other plant pathogens, M. graminicola has a long latent period during which it evades host defenses. Although this type of stealth pathogenicity occurs commonly in Mycosphaerella and other Dothideomycetes, the largest class of plant-pathogenic fungi, its genetic basis is not known. To address this problem, the genome of M. graminicola was sequenced completely. The finished genome contains 21 chromosomes, eight of which could be lost with no visible effect on the fungus and thus are dispensable. This eight-chromosome dispensome is dynamic in field and progeny isolates, is different from the core genome in gene and repeat content, and appears to have originated by ancient horizontal transfer from an unknown donor. Synteny plots of the M. graminicola chromosomes versus those of the only other sequenced Dothideomycete, Stagonospora nodorum, revealed conservation of gene content but not order or orientation, suggesting a high rate of intra-chromosomal rearrangement in one or both species. This observed “mesosynteny” is very different from synteny seen between other organisms. A surprising feature of the M. graminicola genome compared to other sequenced plant pathogens was that it contained very few genes for enzymes that break down plant cell walls, which was more similar to endophytes than to pathogens. The stealth pathogenesis of M. graminicola probably involves degradation of proteins rather than carbohydrates to evade host defenses during the biotrophic stage of infection and may have evolved from endophytic ancestors.
| The plant-pathogenic fungus Mycosphaerella graminicola causes septoria tritici blotch, one of the most economically important diseases of wheat worldwide and a potential threat to global food production. Unlike most other plant pathogens, M. graminicola has a long latent period during which it seems able to evade host defenses, and its genome appears to be unstable with many chromosomes that can change size or be lost during sexual reproduction. To understand its unusual mechanism of pathogenicity and high genomic plasticity, the genome of M. graminicola was sequenced more completely than that of any other filamentous fungus. The finished sequence contains 21 chromosomes, eight of which were different from those in the core genome and appear to have originated by ancient horizontal transfer from an unknown donor. The dispensable chromosomes collectively comprise the dispensome and showed extreme plasticity during sexual reproduction. A surprising feature of the M. graminicola genome was a low number of genes for enzymes that break down plant cell walls; this may represent an evolutionary response to evade detection by plant defense mechanisms. The stealth pathogenicity of M. graminicola may involve degradation of proteins rather than carbohydrates and could have evolved from an endophytic ancestor.
| The ascomycete fungus Mycosphaerella graminicola (Figure S1) causes septoria tritici blotch (STB), a foliar disease of wheat that poses a significant threat to global food production. Losses to STB can reduce yields of wheat by 30 to 50% with a huge economic impact [1]; global expenditures for fungicides to manage STB total hundreds of millions of dollars each year [2]–[3]. This fungus is difficult to control because populations contain extremely high levels of genetic variability [4] and it has very unusual biology for a pathogen. Unlike most other plant pathogens [5]–[7], M. graminicola infects through stomata rather than by direct penetration and there is a long latent period of up to two weeks following infection before symptoms develop. The fungus evades host defenses [8] during the latent phase, followed by a rapid switch to necrotrophy immediately prior to symptom expression 12–20 days after penetration [5], [9]–[10]. Such a switch from biotrophic to necrotrophic growth at the end of a long latent period is an unusual characteristic shared by most fungi in the genus Mycosphaerella. Very little is known about the cause or mechanism of this lifestyle switch [9]–[10] even though Mycosphaerella is one of the largest and most economically important genera of plant-pathogenic fungi.
A striking aspect of M. graminicola genetics is the presence of many dispensable chromosomes [11]. These can be lost readily in sexual progeny with no apparent effect on fitness. However, the structure and function of dispensable chromosomes are not known. Here we report the first genome of a filamentous fungus to be finished according to current standards [12]. The 21-chromosome, 39.7-Mb genome of M. graminicola revealed an apparently novel origin for dispensable chromosomes by horizontal transfer followed by extensive recombination, a possible mechanism of stealth pathogenicity and exciting new aspects of genome structure. The genome provides a finished reference for the Dothideomycetes, the largest class of ascomycete fungi, which also includes the apple scab pathogen Venturia inaequalis, the southern corn leaf blight pathogen Cochliobolus heterostrophus, the black Sigatoka pathogen of banana, M. fijiensis, and numerous other pathogens of almost every crop.
The finished genome of M. graminicola isolate IPO323 consists of 21 complete chromosomes, telomere to telomere (Figure S2), with the exceptions of one telomere of chromosome 21 and two internal gaps of unclonable DNA that are missing from chromosome 18 (Table 1). Alignments between the 21 chromosomes and two genetic linkage maps yielded an excellent correspondence (Figure 1 and Figure S3), representing the most complete and the first finished sequence of a filamentous fungus. The next most complete genome of a filamentous fungus is that of Aspergillus fumigatus, which did not include centromere sequences and contained 11 gaps in total [13]. The complete 43,960-bp mitochondrial genome also was obtained and has been described elsewhere [14].
Comparative genome hybridizations using a whole-genome tiling array made from the genome sequence of IPO323 demonstrated striking sexually activated chromosomal plasticity in progeny isolates (Figure 2) and chromosome number polymorphisms in field isolates. For example, isolate IPO94269, a field strain from bread wheat in the Netherlands, was missing two chromosomes that were present in IPO323 (Figure 2A).
Sexual-driven genome plasticity was particularly evident among progeny isolates in the two mapping populations, including losses of chromosomes that were present in both parents and disomy for others [11]. For example, progeny isolate #51 of the cross between IPO323 and IPO94269 lost chromosomes 14 and 21 (Figure 2B) even though they were present in both parents. This isolate also was missing chromosome 20, which was polymorphic for presence between the parents of the cross. More surprisingly, this isolate was disomic for chromosomes 4 and 18 (Figure 2B), indicating that chromosomes can be both gained and lost during meiosis. For chromosome 18, both copies must have originated from IPO323 because no homolog was present in IPO94269. Molecular markers for chromosome 4 appeared to be heterozygous indicating that both parents contributed a copy to progeny isolate #51 (data not shown). Progeny isolate #2133 of the cross between isolates IPO323 and IPO95052 showed loss of three dispensable chromosomes (15, 18 and 21) that were present in both parents (Figure 2C), most likely due to non-disjunction during meiosis. Thus, extreme genome plasticity was manifested as chromosome number and size polymorphisms [11] that were generated during meiosis and extended to core as well as dispensable chromosomes.
The whole-genome hybridizations also indicated that the core and dispensable chromosomes can be remarkably uniform for gene content, given the high capacity of the latter for change. Comparative genome hybridizations between IPO323 and IPO95052, an isolate from a field of durum wheat in Algeria, showed that they had the same complement of core and dispensable chromosomes (Figure 2D). This was surprising, because populations of the pathogen from durum wheat (a tetraploid) usually are adapted to that host and not to hexaploid bread wheat, yet the chromosomal complements of isolates from these hosts on different continents were the same.
Evidence for repeat-induced point mutation (RIP), a mechanism in fungi that inactivates transposons by introducing C to T transitions in repeated sequences [15]–[16], was seen in genome-wide analyses of transition∶transversion ratios in long terminal repeat (LTR) pairs from 20 retrotransposon insertions which had 255 transitions and 6 transversions for a ratio of 42.5∶1. Similarly high transition∶transversion ratios were found in all repetitive sequences analyzed and extended to the coding regions in addition to the LTRs [17]. The reverse transcriptase coding regions from transposon families RT11 and RT15 had transition∶transversion ratios of 27.8∶1 and 25.3∶1, respectively, instead of the 1∶1 ratio expected among 6,939 mutations analyzed. This high incidence of transitions most likely reflects changes caused by RIP. The coding regions of all transposons with more than 10 copies included stop codons that prevent proper translation, indicating that they were inactivated.
There were significant differences in structure and gene content between the 13 core and eight dispensable chromosomes (Table 1 and Table 2); the latter are referred to collectively as the dispensome. The dispensome constituted about 12% of the genomic DNA but contained only 6% of the genes. In contrast, the 13 core chromosomes had twice as many genes per Mb of DNA, about half as much repetitive DNA, a significantly higher G+C content, and much higher numbers of unique genes (Table 1 and Table 2). Genes in the dispensome were significantly shorter, usually were truncated relative to those on the core chromosomes (Table 2) and had dramatic differences in codon usage (Figure S4).
About 59% of the genes on core chromosomes could be annotated compared to only 10% of those on the dispensome (Table 2). Some unique genes in the dispensome with intact, presumably functional reading frames, had possible paralogs on the core chromosomes (Figure S5) that appeared to be inactivated by mutations (Figure S6). A majority of the annotated dispensome genes coded for putative transcription factors or otherwise may function in gene regulation or signal transduction (Table S1). Most of the redundant genes on the dispensome were copies of genes present on core chromosomes, yet no syntenic relationships could be identified. Instead, each dispensable chromosome contained genes and repetitive sequences from all or most of the core chromosomes (Figure 3 and Figure S7) with additional unique genes of unknown origin. Sharing of genetic material applied to core chromosomes as well as the dispensome, consistent with a high level of recombination (Figure S8). Whether the primary direction of transfer is from core to dispensable chromosomes or vice versa is not known.
The dispensome contained fewer genes encoding secreted proteins such as effectors and other possible pathogenicity factors compared to the core set. Signal peptides showed no enrichment on the dispensome (Table S1) except for a few clusters overlapping with transposon-related repeats. Although mature microRNAs have not been demonstrated in fungi, they may be important regulatory molecules. In the M. graminicola genome, 418 non-overlapping loci potentially encoding pre-microRNA-like small RNA (pre-milRNA) were predicted computationally based on the RFAM database [18]. This number was similar to the 434 loci predicted in the 41-Mb genome of Neurospora crassa using the same approach. Of the 418 putative pre-milRNA loci predicted in the genome of M. graminicola, 88 (21%) are located on the 11% of the genome present as dispensome. This is about twice as much as is expected on the basis of a random distribution. Therefore, the dispensome is enriched for pre-milRNA loci.
The 418 pre-milRNA loci code for 385 non-redundant pre-milRNA sequences that can give rise to distinguishable mature milRNAs. The occurrence of mature milRNAs derived from the predicted set was analyzed in a small-RNA data consisting of almost 6 million reads (Illumina platform) generated from germinated spores of M. graminicola isolate IPO323 (Table S2).
Many of the non-redundant predicted milRNA sequences were represented in the RNA reads, at widely different amounts per sequence. In total, 65 of the 385 non-redundant sequences were observed 10 times or more. Two predicted sequences occurred more than a thousand times each, experimentally confirming the presence of putatively mature milRNAs derived from computationally predicted pre-milRNA sequences. In N. crassa, computationally predicted putative milRNA sequences also were confirmed experimentally [19], supporting the likelihood of their existence in M. graminicola.
The origin of the dispensome of M. graminicola is not clear. The two most likely origins would be degeneration of copies of the core chromosomes or by horizontal transfer. Disomy for core chromosomes, as seen in one of the progeny isolates, could provide the origin for a dispensable chromosome. If one of the two chromosome copies became preferentially subject to RIP followed by breakage or interstitial deletions this could result in a degenerated copy of that core chromosome. However, in that case we would expect the dispensome to share large regions of synteny with specific core chromosomes, and this was not observed, which renders this explanation less likely.
The large differences in codon usage between core and dispensable chromosomes could be explained by horizontal transfer or possibly by RIP. To discriminate between these hypotheses, RIP was simulated on the genes of the core chromosomes. Principal components analysis (PCA) of the simulated data set did not reduce the differences in codon bias (Figure S9A); if anything, it made them farther apart. This result was consistent whether it included only putative functional, truncated copies or entire pseudogenes after RIPping (data not shown). DeRIPping of genes on the dispensable chromosomes also did not affect the results (Figure S9B), so RIP could not explain the observed differences in codon usage between core and dispensable chromosomes. PCA of a sample of genes shared between core and dispensable chromosomes showed few differences in codon bias (Figure S9C) or amino acid composition (Figure S9D), consistent with an origin by duplication and exchange among chromosomes. This conclusion was supported when the analysis was expanded to include all genes with putative homologs on core and dispensable chromosomes (Figure S9E) even though these genes had a very different codon usage compared to the entire sets of genes on the core chromosomes (Figure S9F).
To test the horizontal transfer hypothesis, additional PCAs were performed on simulated horizontal transfer data sets made by combining the genome of M. graminicola with those of two other fungi. Best non-self BLAST hits for genes on the M. graminicola dispensome most often were to fungi in the Pleosporales or Eurotiales (Table S3), so published genomes from species representing those orders were chosen for analysis. PCA of the combined genomes of M. graminicola and Stagonospora nodorum (representing the Pleosporales) gave separate, tight clusters for the core chromosomes of M. graminicola versus most of those from S. nodorum (Figure S10A). Dispensable chromosomes of M. graminicola formed a looser, distinct cluster, and a fourth cluster was comprised of M. graminicola chromosome 14 plus scaffolds 44 and 45 of S. nodorum (Figure S10A); this may indicate the existence of dispensable chromosomes in the latter species. Analysis of the combined genomes of M. graminicola plus Aspergillus fumigatus (Eurotiales) gave a similar result (Figure S10B). The separate clustering by PCA of the M. graminicola dispensome and core chromosomes is consistent with an origin by horizontal transfer, but not from either of the two species tested. PCA on the frequencies of repetitive elements also indicated a separation between core and dispensable chromosomes (Figure S11), consistent with the horizontal transfer hypothesis.
A more refined test of the RIP hypothesis was performed by using the observed rates of all mutations in families of transposons with 10 or more elements to simulate mutational changes on replicated samples drawn from the core chromosomes. Observed mutation rates were calculated from aligned sequences; multicopy transposons were chosen for this analysis because they are the most likely to have been processed through the RIP machinery so will reflect the actual biases that occur in M. graminicola. Codon bias and other parameters in the mutated samples were then compared to those in the dispensome and in the original, non-mutated samples. Application of the mutational changes moved the samples drawn from the core chromosomes closer to the value observed for the dispensome, but the dispensome remained distinct except for a few of the analyses that are least likely to be affected by selection (Figure 4). This confirmed that the dispensome has been subject to RIP but that this alone was not sufficient to explain the observed pattern of codon usage.
Pairwise sequence comparisons between the chromosomes of M. graminicola and scaffolds of Stagonospora nodorum, another wheat pathogen in the Dothideomycetes but in a different order from Mycosphaerella, revealed multiple regions with approximately 70–90% similarity (Figure 5). However, the similarity did not extend to the dispensome, which generally was different from all of the S. nodorum scaffolds. Detailed examination showed that each region of similarity generally represents only one or a few genes in both organisms. Comparisons between the initial draft genome (version 1.0) of M. graminicola (Figure 5A) and the finished sequence (Figure 5B) revealed some misassemblies and also indicated scaffolds that ultimately were joined in the final assembly.
A surprising result was that the dot-plot patterns were very different from those that characterize the macro- or microsynteny seen in other organisms when viewed at a whole-scaffold/chromosome scale. Instead of the expected diagonal lines indicating chromosomal regions with content in the same order and orientation, the dots are scattered quasi-randomly within ‘blocks’ defined by scaffold/chromosome boundaries (Figure 5). For many S. nodorum scaffolds the vast majority of dots related are shared exclusively with one or a small number of M. graminicola chromosomes. For example, there are predominant one-to-one relationships between M. graminicola version 3 chromosomes 11 and 12 with S. nodorum scaffolds 21 and 7 (Figure 5B, circle V), respectively. Similarly, M. graminicola chromosomes 5–10 each had strong relationships with 2 to 4 scaffolds of S. nodorum. We refer to this conservation of gene content but not order or orientation among chromosomes as ‘mesosynteny’. Analyses of additional genomes has shown that mesosynteny as defined here occurs among all Dothideomycetes tested and may be unique to that class of fungi (data not presented).
Macrosyntenic relationships are used commonly to assist the assembly and finishing of fragmented genome sequences [20]–[23], particularly in prokaryotes. Sequences that are macrosyntenic to a long segment of a closely related genome are highly likely to be joined physically. If mesosynteny between a new genome assembly and a reference genome also may be used to suggest scaffolds that should be juxtaposed it could significantly reduce the cost and complexity of assembling and finishing genomes. To test whether mesosynteny could be used to predict scaffold or contig joins in a genomic sequence, versions 1 and 2 of the M. graminicola genome assembly were analyzed to determine whether any of the improvements in the finished genome could have been predicted bioinformatically by mesosynteny (Dataset S1).
The first version of the M. graminicola genome consisted of 129 scaffolds (http://genome.jgi-psf.org/Mycgr1/Mycgr1.home.html). Comparison of M. graminicola version 1 scaffolds with those of the P. nodorum genome predicted all scaffold joins made in version 2 (Figure 5, Dataset S1). Version 1 scaffolds 10 and 14 (Figure 5: group I), 7 and 17 (groups II, VII and IX), and 12 and 22 (groups III and VIII) were joined into chromosomes 7, 5 and 10, respectively. Mesosynteny also indicated both instances where version 1 scaffolds were assembled incorrectly and subsequently were split in version 2. Compared to the scaffolds of P. nodorum, M. graminicola version 1 scaffold 4 exhibited regions of mesosynteny adjacent to regions of no synteny. Corrections to the assembly made in version 2 separated these two distinct regions into separate chromosomes. Version 1 scaffolds 4 and 9 (Figure 5: groups IV/VI and V) were corrected to version 2 chromosomes 6 and 16 (Figure 5: group IV/VI) and chromosomes 12 and 21 (Figure 5: group V) respectively. Mesosynteny was remarkably successful and has great potential to assist the assembly and finishing of fungal genomes.
Generally, gene families involved in cell wall degradation are expanded in fungal plant pathogens [24]–[25]. However, in M. graminicola, gene families characterized by the Carbohydrate-Active Enzyme database (CAZy) [26] as plant cell wall polysaccharidases were severely reduced in size (Figure 6). According to the CAZy analysis, the genome of M. graminicola contains fewer genes for cellulose degradation than those of six other fungi with sequenced genomes including both grass pathogens and saprophytes (Table 3), and only about one-third as many genes for cell wall degradation in total compared to the other plant pathogens (Table S4). This reduction in CAZymes in M. graminicola was very visible when the putative genes were divided based on polysaccharide substrate (Table S4). In addition, genes involved in appressorium formation, which are required for pathogenesis of many plant pathogens including Magnaporthe oryzae [27], were absent or reduced in the Mycosphaerella graminicola genome, reflecting its alternative host-penetration strategy.
To further analyze the mechanism of stealth pathogenesis, we profiled the growth on polysaccharides of M. graminicola compared to Stagonospora nodorum and Magnaporthe oryzae, two pathogens of the cereals wheat and rice, respectively, with sequenced genomes (Figure S12). Growth of M. graminicola corresponded with the CAZy annotation for a strongly reduced number of genes encoding putative xylan-degrading enzymes. Furthermore, the CAZy annotation demonstrated that M. graminicola contains a much smaller set of glycoside hydrolases, carbohydrate esterases, and carbohydrate binding modules (CBMs) compared to the other two cereal pathogens (Table S5). The strong reduction of CBMs in M. graminicola suggests a different strategy in the degradation of plant cell walls compared to the other two species. The M. graminicola genome is particularly depauperate for enzymes degrading cellulose, xylan and xyloglucan compared to the other two species, so is very atypical for a cereal pathogen.
A possible mechanism of stealth pathogenesis was indicated by gene families that were expanded in the genome of M. graminicola. In comparative analyses of gene families and PFAM domains with several other fungi, the most striking expansions were observed for peptidases (M3, S28, pro-kuma, M24, metalloendopeptidase, metalloproteinase) and alpha amylases (glycoside hydrolase family 13) (Tables S6 and S7). This suggests that alternative nutrition sources during the biotrophic phase of infection may be proteins which are available in the apoplast, or possibly starch from chloroplasts that are released early in the infection process [5]. Overall, these analyses revealed that the genome of M. graminicola differs significantly from those of other cereal pathogens with respect to genes involved in plant penetration as well as polysaccharide and protein degradation (Figure 6, Table 3), which most likely reflects its stealthy mode of pathogenesis.
Differences in gene expression during the different stages of infection were evident from an analysis of EST sequences [9] from wheat leaves 5, 10 and 16 days after inoculation (DAI) with M. graminicola. Most genes were present at only one sampling time with little overlap, particularly between the library from the biotrophic stage of infection (5 DAI) compared to the other two (Figure S13A). Lack of overlap extended to a library from minimal medium minus nitrogen to simulate the nitrogen starvation thought to occur during infection (Figure S13B). Expression of genes for cell wall-degrading enzymes also was reduced during the biotrophic stage of infection [9], consistent with the stealth-pathogenicity hypothesis.
The dispensome as defined here includes all parts of the genome that can be missing in field or progeny isolates with no obvious effects on fitness in axenic culture, on a susceptible host or during mating. For M. graminicola, this includes the eight known dispensable chromosomes in isolate IPO323 plus any others that may be discovered in the future. The core genome consists of all chromosomes that are always present in field and progeny isolates, presumably because they contain genes that are vital for survival so cannot be lost. Both core and dispensable chromosomes may be present in two or possibly more copies, but core chromosomes are never absent.
The dispensome of M. graminicola is very different from the supernumerary or B chromosomes in plants and some animals. The B chromosomes of plants contain few if any genes and are composed mostly of repetitive elements assembled from the A chromosomes. They may have a negative effect on fitness [28] and appear to be maintained primarily by meiotic drive [29]. In contrast, the dispensome of M. graminicola contains many unique and redundant genes and is not maintained by meiotic drive, as individual chromosomes are lost readily during meiosis [11].
Dispensable chromosomes have been reported in other fungi but they are significantly fewer and larger (from 0.7 to 4.9 Mb with an average of about 1.5 to 2.0 Mb) than those in M. graminicola (from 0.42 to 0.77 Mb) and mostly are composed of repetitive DNA with few known genes [30]. Unlike the dispensome of M. graminicola, the few genes on dispensable chromosomes in other fungi often are pathogenicity factors [31]–[33] and whole chromosomes may be transferred asexually [34]. Dispensable chromosomes in other fungi are different from the dispensome of M. graminicola except for the conditionally dispensable or lineage-specific chromosomes reported recently in Nectria haematococca (asexual stage: Fusarium solani) and other species of Fusarium [35]–[36], which also were different from core chromosomes in structure and gene content and contained numerous unique genes. However, unlike those in M. graminicola, dispensable chromosomes of Fusarium species had clear functions in ecological adaptation, were transferred more or less intact among closely related species [35] and did not show extensive recombination with core chromosomes.
The high instability of the M. graminicola dispensome during meiosis and mitosis would cause it to be eliminated unless it provided a selective advantage to the pathogen at least under some conditions. The unique genes with annotations indicated possible functions in transcription or signal transduction. There also was an enrichment for predicted pre-milRNAs, which may indicate that parts of the dispensome are involved in gene regulation. Based on dispensable chromosomes in other plant pathogens, genes on the dispensome were expected to be involved with host adaptation or pathogenicity, yet so far no genes for pathogenicity or fitness of M. graminicola have been mapped to the dispensome [37]. A more interesting possibility is that the dispensome facilitates high recombination among chromosomes and could provide a repository of genes that may be advantageous under certain environmental conditions. This hypothesis should be tested by additional experimentation.
A recent comparison of the M. graminicola genome with that of its closest known relative, the unnamed species S1 from wild grasses in Iran, identified probable homologs for all of the dispensome chromosomes in the sibling species except for chromosome 18 [38]. These putative homologs presumably are dispensable also in species S1, but this has not been proven and only one isolate has been sequenced. Species S1 and M. graminicola are thought to have diverged approximately 10,500 years ago [39], concomitant with the domestication of wheat as a crop. Therefore, unlike dispensable chromosomes in other fungi, the dispensome of M. graminicola appears to be relatively ancient and has survived at least one speciation event. Analyses of two recently sequenced Dothideomycetes with Mycosphaerella sexual stages, M. pini (asexual stage: Dothistroma septosporum) and M. populorum (asexual stage: Septoria musiva), showed that they contained clear homologs of all of the core chromosomes of M. graminicola, but none of their chromosomes corresponded to the dispensome (B. Dhillon and S. B. Goodwin, unpublished). Taken together, these observations indicate that the dispensome of M. graminicola most likely was acquired prior to its divergence from a common ancestor with species S1 more than 10,000 years ago, but after the split of the M.graminicola-S1 lineage from that which gave rise to M. pini and M. populorum. The mechanism for the longevity of this dispensome with no obvious effects on fitness is not known.
More than half of the genes on the dispensome and almost all of the transposons also were present on core chromosomes. Moreover, there was no increase in gene numbers so a simple transfer of chromosomes from another species does not explain all of the observations. Instead, we propose a new model for the origin of dispensable chromosomes in M. graminicola by horizontal transfer followed by degeneration and extensive recombination with core chromosomes. The tight clustering of the dispensable chromosomes in the PCAs, with the possible exception of chromosome 14, indicates that they probably came from the same donor species. However, it is difficult to explain why they are so numerous. The most likely mechanism of horizontal transfer is via a sexual or somatic fusion with another species that had eight or more chromosomes, in which only a few genes were maintained on each donated chromosome. Chromosome segments that were redundant with the core set could be eliminated, leaving only those that are unique or that could confer some sort of selective advantage to the individual or to the dispensome. The fitness advantage could be transitory or occur only under certain conditions to allow those chromosomes to be dispensable, at least on an individual or population basis. Another possibility is that the numerous dispensable chromosomes are fragments from one or two larger chromosomes that were broken, acquired additional telomeres and lost content to result in their current, reduced complements of genes. High recombination within chromosomes and transfer of content between the donor and host chromosomes must have occurred to explain the observed pattern of shared genes.
The recombination hypothesis is supported by degenerated copies of some unique genes that were found on core chromosomes. These most likely represent genes that were copied from core to dispensable chromosomes, after which the copy on the core chromosome became inactivated, probably by RIP. Duplication, diversification and differential gene loss were proposed recently as the origin of lineage-specific gene islands in Aspergillus fumigatus [40], but that process seems to be very different from what occurred in M. graminicola. In A. fumigatus, large blocks of genes with synteny to other chromosomes were found, the opposite of what was seen for M. graminicola. The origin and evolution of the dispensome in M. graminicola seems to be very different from those reported for dispensable chromosomes in other fungi [35]. Unlike other fungi in which single chromosomes seem to have been transferred recently, the dispensome of M. graminicola most likely originated by ancient horizontal transfer of many chromosomes thousands of years ago. So far it is not known to be conditionally dispensable, unlike dispensable chromosomes in other fungi, which have clear roles in ecological adaptation.
The mesosyntenic analyses provided a new approach that complements the use of genetic linkage maps to support whole-genome assembly. Gene content was highly conserved on syntenic chromosomes in the two distantly related species, but there was little or no conservation of gene order or orientation. The comparison of the version 1 assembly of M. graminicola with the related S. nodorum genome sequence indicated scaffolds that should be merged and others that were erroneously assembled into one scaffold. Hence, mesosynteny validated the high-density genetic analyses and may provide an additional tool for whole-genome assembly for fungi where linkage maps do not exist or cannot be generated. Groups of genes in S. nodorum that corresponded to more than one group in M. graminicola may indicate scaffolds that should be joined in S. nodorum or, more likely, may reflect chromosomal rearrangements that have occurred since the divergence of S. nodorum and M. graminicola from an ancient common ancestor.
Considering their early divergence [41] relative to species within the same genus, the degree of mesosyntenic conservation between M. graminicola and S. nodorum is striking. However, it is very surprising that the synteny only applied to gene content but not order or orientation. In comparisons between other organisms, synteny plots usually yield diagonal lines even between unrelated species such as humans and cats [42]. The lack of diagonal lines in the comparisons of S. nodorum with M. graminicola indicate a high rate of shuffling of genes on chromosomal blocks that have remained constant over long periods of evolutionary time. The mechanism by which these small chromosomal rearrangements occur is not known.
The greatly reduced number of cell wall-degrading enzymes (CWDEs) in the genome of M. graminicola compared with other sequenced fungal genomes might be an evolutionary adaptation to avoid detection by the host during its extended, biotrophic latent phase and thus evade plant defenses long enough to cause disease. Similar loss of CWDEs in the ectomycorrhizal fungus Laccaria bicolor was thought to represent an adaptation to a symbiotic lifestyle [43]. Based on these results we propose a novel, biphasic mechanism of stealth pathogenesis. During penetration and early colonization, M. graminicola produces a reduced set of proteins that facilitate pathogenicity and function as effectors in other fungi. Instead of the usual carbohydrate metabolism, nutrition during the extended biotrophic phase may be by degradation of proteins rather than carbohydrates in the apoplastic fluid and intercellular spaces. The large number of proteases expressed during the early stages of the infection process supports this hypothesis. The biotrophic phase terminates by a switch to necrotrophic growth, production of specific cell wall-degrading enzymes and possibly by triggering programmed cell death [5], [9]–[10].
Stealth biotrophy raises the intriguing possibility that M. graminicola and possibly other Dothideomycetes may have evolved originally as endophytes or could be evolving towards an endophytic lifestyle. The finished genome of M. graminicola provides a gold standard [12] for this class of fungi, which is the largest and most ecologically diverse group of Ascomycetes with approximately 20,000 species, classified in 11 orders and 90 families, and provides a huge advantage for comparative genomics to identify the genetic basis of highly divergent lifestyles.
Mycosphaerella graminicola isolates IPO323 and IPO94269 are Dutch field strains that were isolated in 1984 and 1994 from the wheat cultivar Arminda and an unknown cultivar, respectively. Isolate IPO95052 was isolated from a durum (tetraploid) wheat sample from Algeria. All isolates are maintained at the CBS-KNAW Fungal Biodiversity Centre of the Royal Netherlands Academy of Arts and Sciences (Utrecht, the Netherlands) under accession numbers CBS 115943 (IPO323), CBS 115941 (IPO94269) and CBS 115942 (IPO95052). Mycelia of each isolate were used to inoculate 200 mL of YG broth (10 g of yeast extract and 30 g of glucose per L) and were cultured until cloudy by shaking at 120 rpm at 18°C, after which the spores were lyophilized, 50 mg of lyophilised spores were placed in a 2-mL tube and ground with a Hybaid Ribolyser (model n° FP120HY-230) for 10 s at 2500 rpm with a tungsten carbide bead. DNA was extracted using the Promega Wizard Magnetic DNA Purification system for food according to instructions provided by the manufacturer.
Whole-genome shotgun (WGS) sequencing of the genome of M. graminicola used three libraries with insert sizes of 2–3, 6–8, and 35–40 kb. The sequenced reads were screened for vector using cross_match, trimmed for vector and quality, and filtered to remove reads shorter than 100 bases. WGS assembly was done using Jazz, a tool developed at the JGI [44]. After excluding redundant and short scaffolds, the assembly v1.0 contained 41.2 Mb of sequence in 129 scaffolds, of which 4.0 Mb (7.5%) was in gaps (Table S8). The sequence depth derived from the assembly was 8.88±0.04.
To perform finishing, the M. graminicola WGS assembly was broken down into scaffold-size pieces and each piece was reassembled with phrap. These scaffold pieces were then finished using a Phred/Phrap/Consed pipeline. Initially, all low-quality regions and gaps were targeted with computationally selected sequencing reactions completed with 4∶1 BigDye terminator: dGTP chemistry (Applied Biosystems). These automated rounds included resequencing plasmid subclones and walking on plasmid subclones or fosmids using custom primers.
Following completion of the automated rounds, a trained finisher manually inspected each assembly. Further reactions were then manually selected to complete the genome. These included additional resequencing reactions and custom primer walks on plasmid subclones or fosmids as described above guided by a genetic map of more than 2,031 sequenced markers plus paired-end reads from a library of Bacterial Artificial Chromosome clones. Smaller repeats in the sequence were resolved by transposon-hopping 8-kb plasmid clones. Fosmid and BAC clones were shotgun sequenced and finished to fill large gaps, resolve larger repeats and to extend into the telomere regions.
Each assembly was then validated by an independent quality assessment. This included a visual examination of subclone paired ends using Orchid (http://www-hagsc.org), and visual inspection of high-quality discrepancies and all remaining low-quality areas. All available EST resources were also placed on the assembly to ensure completeness. The finished genome consists of 39,686,251 base pairs of finished sequence with an estimated error rate of less than 1 in 100,000 base pairs. Genome contiguity is very high with a total of 21 chromosomes represented, 19 of which are complete and 20 of which are sequenced from telomere to telomere.
Both draft (v1.0) and finished (v2.0) assemblies of M. graminicola were processed using the JGI annotation pipeline, which combines several gene predictors:1) putative full-length genes from EST cluster consensus sequences; 2) homology-based gene models were predicted using FGENESH+ [45] and Genewise [46] seeded by Blastx alignments against sequences from the NCBI non-redundant protein set; 3) ab initio gene predictor FGENESH [45] was trained on the set of putative full-length genes and reliable homology-based models. Genewise models were completed using scaffold data to find start and stop codons. ESTs were used to extend, verify, and complete the predicted gene models. Because multiple gene models per locus were often generated, a single representative gene model for each locus was chosen based on homology and EST support and used for further analysis. Those comprised a filtered set of gene models supported by different lines of evidence. These were further curated manually during community annotation and used for analysis.
All predicted gene models were annotated using InterProScan [47] and hardware-accelerated double-affine Smith-Waterman alignments (www.timelogic.com) against the SwissProt (www.expasy.org/sprot) and other specialized databases such as KEGG [48]. Finally, KEGG hits were used to map EC numbers (http://www.expasy.org/enzyme/), and Interpro hits were used to map GO terms [49]. Predicted proteins also were annotated according to KOG [50], [51] classification.
Following the machine annotation, manual validation and correction of selected gene sets was performed by more than 30 annotators through a jamboree held at the JGI facilities in Walnut Creek, California, USA. Annotators were trained by JGI staff and continue to make modifications as necessary.
Potential microRNA-like small RNA (milRNAs) loci were annotated using the INFERNAL software tool and based on 454 microRNA families (covarion models) from the RFAM database version 9.1 [52]. milRNAs were predicted if their scores were higher than thresholds, defined individually for each family, in the same way as PFAM domains are predicted.
Experimental validation of the predicted milRNAs was done by sequencing of an RNA library Total RNA was isolated from spores germinated on water agar of M. graminicola isolate IPO323. A small RNA library was prepared according to the protocol for Illumina sequencing; small RNAs from 16–∼50 nt were isolated from gels, sequenced with an Illumina/Solexa single read DNA 50 cycles Genome Analyzer II, and compared by BLAST search against the list of 535 predicted pre-milRNAs from the genome sequence.
Assembly and annotations of the M. graminicola finished genome are available from the JGI Genome Portal at http://www.jgi.doe.gov/Mgraminicola and were deposited at DDBJ/EMBL/GenBank under the project accession ACPE00000000.
Whole-genome tiling microarrays were designed by choosing one 50-mer primer every 100 bases spanning the entire finished genome. The arrays were manufactured and hybridized by the Nimblegen Corporation with total DNA extracted from each field isolate.
The CodonW package (http://codonw.sourceforge.net/) was used for correspondence analysis of codon usage, which mathematically is identical to principal component analysis. CodonW requires as an input a set of coding sequences, usually of individual genes. For chromosome-level analyses coding sequences from the frozen gene catalog models for each chromosomes were concatenated, forming 21 ‘superORFs’, one for each chromosome. Because partial models may introduce some potential frameshifts with internal stop codons they were removed from the analysis; this did not affect the results as their total number is low. CodonW has no graphical outputs, so they were used as inputs for scatter plots in R (http://www.r-project.org/).
For M. graminicola only a similar analysis was done for repeats. RepeatScout was run on the genome to produce a set of ab initio-identified repeat sequences. From that set 81 distinct repeat sequences, each with an occurrence exceeding 20 times in the genome, were extracted. For each chromosome a vector of length 81 was calculated with the relative frequency of each repeat. A PC analysis was run on the resulting vectors using the standard principal component function pcomp in R. Separation at the repeat level means that these chromosomes have distinct evolutionary profiles not only on the protein-coding level, but also on other parts of the chromosomes, suggesting that entire chromosomes may be transferred horizontally.
Dot plots were generated via MUMMER 3.0 [53] with data derived from default PROmer comparisons between the M. graminicola genome assembly versions 1 and 2 (http://genome.jgi-psf.org/Mycgr3/Mycgr3.home.html) and S. nodorum SN15 assembly version 2 [54], available under GenBank accessions CH445325–CH445384, CH445386–CH445394 and CH959328–CH959365, or AAGI00000000. Additional comparisons and statistical analyses were made with custom-designed perl scripts.
Data from the M. graminicola version 2 comparison with S. nodorum were used to test the efficacy of mesosyntenic comparisons to assist the completion of fungal genomes. The mesosynteny-based prediction of scaffold joining involved 3 stages: determining the percent coverage of scaffolds/chromosomes for each scaffold/chromosome pair (i.e., a function of the number of ‘dots’ per ‘block’); determining which scaffold/chromosome pairs were significantly related and forming groups of joined scaffolds; and filtering out background levels of similarity due to sequence redundancy and incomplete genome assemblies.
Coordinates of homologous regions were obtained from PROmer coordinate outputs (MUMMER 3.0) and used to determine the percent of sequence covered by matches to a sequence from the alternate genome for each sequence pair. Where match coordinates overlapped on the sequence of interest, those matches were merged into a single feature to avoid redundancy. A perl script for conversion of PROmer coordinate outputs to a table of percent coverage is available on request.
Coverage values for each M. graminicola-S. nodorum sequence pair were subject to a binomial test for significance. The threshold for significance (Psig)≥0.95 was:where x is the percent coverage, n equals 100, and p is the probability of chromosome homology.
The probability of chromosome homology (p) was equal to 1/(21×19), which was derived from the number of M. graminicola chromosomes (21) and the approximate PFGE estimate of S. nodorum chromosomes (19) [55]. This is the likelihood that a given sequence pair represents related chromosomes. This model assumes that no whole-genome/chromosome duplication events have occurred previously between either fungal genome since divergence from their last common ancestor.
The significance of percent coverage (Psig) was tested bidirectionally for each sequence pair (i.e., for sequence pair A–B, both coverage of sequence A by B and coverage of sequence B by A were tested). Sequence pairs were significantly related if a test in either direction was successful. A minimum length threshold of 1 kb was also imposed for both sequences. Where multiple scaffolds of M. graminicola were significantly related to the same S. nodorum scaffold, those M. graminicola scaffolds formed a ‘joined group’ of candidates for representation of the same chromosome.
All possible paired combinations of M. graminicola scaffolds present within predicted joined groups were subject to filtering for high levels of background similarity as follows:The retention score is a measure of the reliability of scaffold join relationships. Joins between M. graminicola scaffold pairs with retention scores <0.25 were discarded.
Annotation of carbohydrate-related enzymes was performed using the Carbohydrate-Active Enzyme database (CAZy) annotation pipeline [26]. BLAST was used to compare the predicted proteins of M. graminicola to a collection of catalytic and carbohydrate-binding modules derived from CAZy. Significant hits were compared individually by BLAST to assign them to one or more CAZy families. Ambiguous family attributions were processed manually along with all identified models that presented defects (deletions, insertions, splicing issues, etc.).
Growth profiling of S. nodorum and M. graminicola was on Aspergillus niger minimal medium [56]. Cultures were grown at 25 degrees for seven days after which pictures were taken for growth comparison. Carbon sources used were: glucose (Sigma); soluble starch (Difco); alpha-cellulose (Sigma); Guar Gum (Sigma, galactomannan); Oat spelt xylan (Sigma); and Apple Pectin (Sigma).
Comparisons of sequence content between core and dispensable chromosomes was with Circos [57]. This tool draws ribbons connecting sequences that align in different data sets.
|
10.1371/journal.pgen.1005756 | Exonic Splicing Mutations Are More Prevalent than Currently Estimated and Can Be Predicted by Using In Silico Tools | The identification of a causal mutation is essential for molecular diagnosis and clinical management of many genetic disorders. However, even if next-generation exome sequencing has greatly improved the detection of nucleotide changes, the biological interpretation of most exonic variants remains challenging. Moreover, particular attention is typically given to protein-coding changes often neglecting the potential impact of exonic variants on RNA splicing. Here, we used the exon 10 of MLH1, a gene implicated in hereditary cancer, as a model system to assess the prevalence of RNA splicing mutations among all single-nucleotide variants identified in a given exon. We performed comprehensive minigene assays and analyzed patient’s RNA when available. Our study revealed a staggering number of splicing mutations in MLH1 exon 10 (77% of the 22 analyzed variants), including mutations directly affecting splice sites and, particularly, mutations altering potential splicing regulatory elements (ESRs). We then used this thoroughly characterized dataset, together with experimental data derived from previous studies on BRCA1, BRCA2, CFTR and NF1, to evaluate the predictive power of 3 in silico approaches recently described as promising tools for pinpointing ESR-mutations. Our results indicate that ΔtESRseq and ΔHZEI-based approaches not only discriminate which variants affect splicing, but also predict the direction and severity of the induced splicing defects. In contrast, the ΔΨ-based approach did not show a compelling predictive power. Our data indicates that exonic splicing mutations are more prevalent than currently appreciated and that they can now be predicted by using bioinformatics methods. These findings have implications for all genetically-caused diseases.
| The biological interpretation of most disease-associated variants has become a real challenge, especially with the implementation of next-generation sequencing. Particular attention is typically given to protein-coding changes often neglecting the potential impact of exonic variants on RNA splicing. Here, we used the exon 10 of MLH1, a gene implicated in hereditary cancer, as a model system to assess the prevalence of RNA splicing mutations among all single-nucleotide variants identified in a given exon by using minigene-based assays. Our study revealed an unexpected high proportion of splicing mutations in MLH1 exon 10 mostly affecting potential exonic splicing regulatory elements (ESRs), which are typically difficult to predict by using in silico tools. We then used five experimental datasets (MLH1, BRCA1, BRCA2, CFTR and NF1) to evaluate the predictive power of 3 in silico approaches recently described for pinpointing ESR-mutations. What’s more, besides predicting which exonic variants affect splicing, ΔtESRseq and ΔHZEI values also indicated the direction and severity of the induced splicing defects. In contrast, the ΔΨ-based approach did not show a compelling predictive power. Our data indicates that exonic splicing mutations are more prevalent than currently appreciated and that they can now be predicted by using bioinformatics methods. These findings have implications for all genetically-caused diseases.
| Tremendous progress has been made in recent years in high-throughput technologies enabling fast and affordable massive parallel DNA sequencing. These methods are now being implemented both in molecular diagnostic settings and in basic research laboratories and hold great promise for discovering the genetic bases of rare and complex diseases [1]. However, even if next-generation sequencing has greatly improved the detection of nucleotide changes in the genome of each individual, the biological and clinical interpretation of most variants remains challenging, representing one of the major hurdles in current medical genetics [2,3]. Several reasons account for the difficulty in distinguishing which variants may cause or contribute to disease [4,5]: (i) the number of single-nucleotide variants (SNVs) detected in each genome is very high, (ii) many disorders are genetically heterogeneous and often caused by rare variants, (iii) access to clinical and family data, such as pedigrees, is frequently limited, (iv) in silico tools aiming at identifying deleterious changes are still imperfect, and (v) biological samples from variant carriers are rarely available for functional analysis. There is obviously a great need for overcoming these limitations, and considerable efforts are now deployed for developing strategies allowing prioritization of variants for functional testing [4,6].
In general, scrupulous attention is given to exonic SNVs mapping to protein coding regions, especially to those producing missense changes. Indeed, one of the most widely used strategies for narrowing down the number of variants susceptible of causing disease is to use a combination of in silico tools that focus on protein features (such as PolyPhen-2, SIFT and Mutation Assessor, among others) [7–9]. Yet, such protein-centric view of the exome landscape merely represents a fraction of the “expression code” underlying each gene sequence. Current knowledge clearly indicates that, besides their protein coding potential, exonic sequences can play an important role in RNA splicing (reviewed in [10,11]). Notably, (i) the first and the last 3 exonic positions are an integral part of 3’ and 5’splice site (3’ss and 5’ss) consensus sequences, and (ii) exons may also contain splicing regulatory elements (ESRs), such as exonic splicing enhancers (ESEs) and exonic splicing silencers (ESSs). ESEs and ESSs usually correspond to 6–8 nucleotide stretches that serve as landing pads for splicing activator or splicing repressor proteins, respectively, thereby influencing the recruitment and activity of the splicing machinery. Whereas 3’ss and 5’ss consensus sequences have been extensively characterized leading to the development of in silico tools that reliably predict alterations in splice site strength (such as MaxEntScan, SpliceSiteFinder-like and Human Splicing Finder, among others), ESRs are still poorly understood and generally regarded as difficult to predict by using bioinformatics approaches [12–15].
Lately, three new in silico approaches were described as promising tools to predict variant-induced ESR alterations. The first approach relies on ESRseq scores established through experimental assessment of the ESR properties of all possible 6-nucleotide motifs [16]. Calculation of total ESRseq score changes (ΔtESRseq), taking into account overlapping hexamers and variant-induced changes, was then implemented by our group as a tool to predict the impact on splicing produced by exonic variants [17]. The second approach is based on ZEI scores derived from a RESCUE-type analysis that computed the relative distribution of hexamer motifs in exons and introns [18]. According to this study, the value of total HZEI score changes (ΔHZEI, also corresponding to overlapping hexamers) can be applied for predicting the impact on splicing of any exonic variant. The third and last approach relies on ΔΨ values (Ψ, percent spliced in) that were bioinformatically established upon compilation of RNAseq data obtained from different tissues, and integration of a large set of pre-established sequence features [19]. The ΔΨ-based approach was developed to predict splicing aberrations induced by any sequence change, either intronic or exonic, including variants affecting splice sites or splicing regulatory signals.
Today, it is estimated that ~15% of all point mutations causing human inherited disorders disrupt splice-site consensus sequences, particularly at intronic positions [20]. Yet, it is now speculated, based on in silico data, that disease-causing aberrant RNA splicing may be more widespread than currently appreciated, with up to 25% of exonic disease-associated variants being expected to disturb ESRs [11,21]. Here, we decided to use the exon 10 of the MLH1 gene as a paradigm to experimentally evaluate these assumptions. MLH1 exon 10 was selected as model system for 3 main reasons: (i) MLH1 is the major gene implicated in Lynch syndrome, one of the most frequent forms of hereditary cancer worldwide, formerly known as hereditary nonpolyposis colorectal cancer (HNPCC), (ii) this gene exhibits a large mutational spectrum, with at least 30% of variants being currently classified as variants of unknown significance and for which large national and international efforts persist in bringing clarification [22,23], and (iii) alterations of potential ESRs were already reported for 3 SNVs in this exon [13,24].
We retrieved all SNVs identified in MLH1 exon 10 and analyzed their impact on splicing by resorting to both minigene-based assays and analysis of patient’s RNA when available. Moreover, we used our experimental data to perform a comparative analysis of the 3 newly developed in silico tools aiming at predicting variant-induced ESR alterations. Our results revealed an unexpected high number of splicing mutations in MLH1 exon 10, most of which affecting potential ESRs, thus corroborating our initial hypotheses. Moreover, we confirmed the predictive power of ΔtESRseq- and ΔHZEI- based approaches, but not that of ΔΨ, for pinpointing this type of mutations.
We have recently shown that a high number of variants (15 out of 36, i.e. 42%) identified in the exon 7 of the BRCA2 gene have a negative impact on splicing [17], a finding strongly suggesting that either BRCA2 exon 7 is exceptionally sensitive to exonic splicing mutations, or that this type of mutations is more frequent than currently estimated. To test these hypotheses, we decided to study the impact on splicing of SNVs identified in another gene, more specifically in MLH1, a gene implicated in Lynch syndrome. Our approach was to use the exon 10 of MLH1 as a model system. We began by interrogating National and International public databases in order to retrieve all single substitutions reported in this exon (Table 1 and Fig 1A). As a result, we found a total of 22 SNVs, most of which identified in cancer patients suspected of Lynch syndrome, including 15 missense, 3 nonsense and 4 synonymous variants. Only 9 of these SNVs are currently classified as clearly pathogenic, and 1 as clearly not pathogenic (Table 1). To assess the impact of all 22 variants on MLH1 exon 10 splicing, we performed an ex vivo splicing assay with pSPL3m-M1e10-derived minigenes (Fig 1B). As shown on Figs 1C and S1, the wild-type pSPL3m-M1e10 minigene predominantly generated transcripts containing exon 10 (79% exon inclusion), and a minority of transcripts without exon 10. These results are in agreement with those previously reported by using the same minigene system [13], and indicate that wild-type pSPL3m-M1e10 mimics, at least in part, the alternative splicing pattern of endogenous MLH1 transcripts [25–27] (~35% to 67% exon 10 inclusion in normal blood samples, according to a study from Charbonnier and collaborators [25]). Importantly, the minigene assay results revealed that 17 out of the 22 SNVs (77%) altered the splicing pattern of exon 10 relative to wild-type (Figs 1C and S1). More precisely, 13 variants increased exon skipping, 4 variants increased exon inclusion, and only 5 variants showed no effect on splicing. Of note, 8 of the 13 exon-skipping mutations (5 missense, 1 synonymous, and 2 nonsense) are currently classified as pathogenic (Table 1). The 13 variants increasing exon 10 skipping can be separated into three categories according to the severity of the splicing defect observed in the pSPL3m-M1e10 minigene assay: a first category consisting of 6 variants causing near-total exon skipping (7% to 9% exon inclusion: c.793C>A, c.882C>T, c.883A>C, c.883A>G, c.884G>A and c.884G>C; S1 Table and S3B Fig), a second category consisting of five variants inducing moderate skipping (26% to 67% exon inclusion: c.793C>T, c.794G>A, c.845C>G, c.851T>A and c.882C>G, S1 Table) and a third category consisting of 2 variants (c.840T>A and c.842C>T) that led not only to increased exon 10 skipping (51% and 18% exon inclusion, respectively, S1 Table) but also to deletion of the last 48 nucleotides of the exon in a small fraction of the minigene transcripts (S1B Fig, right panel). A bioionformatics analysis using splice site-dedicated algorithms revealed that both variants, c.840T>A and c.842C>T, create a 5’splice site at exonic position +46 (MLH1 c.836, S1C Fig), which explains the 48-nucleotide deletion at the end of the exon but not the predominant exclusion of the entire exon.
To gain further insight into the severity of the splicing defects of the 13 mutants that induced exon 10 skipping in pSPL3m-M1e10, we tested these variants in the context of another splicing reporter minigene, pCAS2-M1e10 (S2 Fig), previously shown to be less sensitive to splicing mutations than pSPL3m-M1e10 [13]. Our results indicate that, in contrast to pSPL3m-M1e10 and as expected, wild-type pCAS2-M1e10 exclusively generated transcripts containing exon 10. We also observed that 5 out of 6 variants from the first category exhibited a behavior similar to the one observed with the pSPL3m minigenes, i.e. near-total exon 10 skipping; the exception being c.882C>T that induced partial, though strong, exon 10 skipping in the pCAS2 system. Four out of 5 variants from the second category showed splicing defects less severe than the ones observed with pSPL3m-M1e10 and, in one case, c.851T>A, no splicing anomaly was detected. One should note that the level of exon skipping induced by this last variant was borderline noticeable in the pSPL3m-derived minigene (Fig 1C and S1 Table). As for the third category group, we observed that the splicing defects detected in the pSPL3m minigene were faithfully reproduced in pCAS2 for one of the variants (c.842C>T, which predominantly yielded exon-skipped products) but not for the other (c.840T>A). The fact that, in the pCAS2 system, c.840T>A predominantly produced transcripts containing exon 10 deleted of its last 48 nucleotides and almost no exon-skipped products (S2 Fig) suggests that the type and severity of the splicing defect caused by this variant depends on surrounding nucleotide context.
We surmise, given their position at the termini of the exon, that 7 out of the 17 splicing mutations detected in exon 10 directly affect the definition of the reference splice sites, either at the level of the 3’splice site (YAG│G, first exonic position underlined) or of the 5’splice site (CAG│GURAGU, 3 last exonic positions underlined) [10]. Accordingly, the effects produced by these 7 variants (c.791A>G, c.882C>G, c.882C>T, c.883A>C, c.883A>G, c.884G>A and c.884G>C) could have been predicted by algorithms commonly used to predict the strength of splice sites, such as Splice Site Finder, MaxEntScan, and Human Splice Finder (HSF-ss). As shown on S3 Fig, in silico data derived from these programs strongly suggest that c.791A>G improves exon inclusion by directly increasing the strength of the 3’splice site, and that the variants at the last three positions of the exon (c.882 to c.884), induce exon skipping by decreasing, to different extents, the strength of the 5’splice site. Moreover, according to this bioinformatics analysis, especially with MaxEntScan (MES), one could have correctly predicted that the splicing defect produced by c.882C>G is less severe than the one induced by c.882C>T. Our data further indicates that a decrease in 5’ss MES score of ≥19% predicts very drastic exon 10 skipping.
Because the 10 remaining splicing mutations detected in MLH1 exon 10 are located outside the positions directly defining the splice sites, we strongly suspect that they interfere with exon recognition by altering ESRs. This type of splicing mutations includes variants that map within the first two thirds of exon 10 and are responsible for inducing either exon skipping (c.793C>A, c.793C>T, c.794G>A c.840T>A, c.842C>T, c.845C>G and c.851T>A) or exon inclusion (c.814T>G, c.815T>C and c.855C>T).
To better understand how MLH1 exon 10 splicing is regulated and how certain SNVs disturb this process, we decided to functionally characterize different regions of the exon in the context of pcDNA-Dup. This three-exon minigene contains a middle exon particularly sensitive to alterations of splicing regulatory elements [13,17,28]. For this purpose, four partially overlapping segments covering the entire exon 10 of MLH1 (R1 to R4, ~30 bp-long each) were individually inserted into the middle exon of this construct and then tested in the context of an ex vivo splicing assay (Fig 2A and 2B). As shown in Fig 2C, our results revealed that region R1 (MLH1 c.791-819) strongly contributed to the recognition of the middle exon. Regions R2 (c.809-c.840), R3 (c.828-c.859) and R4 (c.856-c.884) had a positive effect on exon inclusion as well, although more moderate for R2 and R3 (in comparison to R1), and weak for R4. Next, we tested all MLH1 exon 10 splicing mutations suspected of altering splicing regulatory elements (n = 10, Fig 2A). Importantly, we found that 8 out of these 10 variations altered the splicing efficiency of the middle exon relative to wild-type (Fig 2D). Variants c.793C>A and c.793C>T (both tested in the context of the R1 segment) induced exon skipping, with c.793C>A having an effect stronger than c.793C>T and thus faithfully recapitulating the defects initially detected in pSPL3m-M1e10 and pCAS2-M1e10 minigene assays. The impact on splicing of variants c.814T>G and c.815T>C was also reproduced in the context of pcDNA-Dup as both variants led to an increase in middle exon inclusion (R2 segment). Finally, variants c.840T>A, c.842C>T, c.845C>G and c.851T>A induced middle exon skipping (R3 segment), with c.842C>T having the more pronounced effect, reminiscent of the effects observed in the pSPL3m-M1e10 and pCAS2-M1e10 assays. Results obtained with c.793C>T and c.842C>T agree with those previously reported by using the pcDNA-Dup system [13]. In contrast, we found that the splicing alterations induced by c.794G>A and c.855C>T in the context of pSPL3m-M1e10 minigene could not be reproduced in the pcDNA-Dup assay. As shown on Fig 2D, c.794G>A did not induce middle exon skipping (R1 segment), and c.855C>T did not lead to an increase in exon inclusion (R3 segment), on the contrary, it slightly increased middle exon skipping. We hypothesize that the effect on splicing of these particular variants depends on surrounding nucleotide context. In conclusion, these experiments point to an apparent asymmetry in the distribution of ESRs in MLH1 exon 10, with a potential ESE-enrichment at the 5’ portion, corresponding to about the first 2/3 of the exon, and/or an ESS-enrichment towards the 3’end. Moreover, the evidence of portability, i.e. that the splicing defects caused by most of the analyzed SNVs can be reproduced in a completely heterologous system, further supports the hypothesis that these variants modify bona fide splicing regulatory elements.
Three independent in silico approaches based on the calculation of either ΔtESRseq [17], ΔHZEI [18], or ΔΨ [19] values were recently described as promising tools for predicting variants that potentially modify splicing regulatory elements. Because it remained to be determined if these tools could be applied to new cases and how they compare to each other, we next decided to evaluate their performance by using the output of the pSPL3m-M1e10 minigene assay as a new dataset. We started by separating the fifteen MLH1 exon 10 SNVs located at distance from the splice sites into three groups based on the minigene results (S1 Table). The first group included mutations that increased exon 10 skipping (n = 7), the second group included variations with no effect on exon 10 splicing (n = 5) and the third group included those that increased exon 10 inclusion (n = 3). Then, score differences (ΔtESRseq, ΔHZEI, and ΔΨ) were calculated for each variant and plotted in a parallel fashion in order to easily confront the discriminating power of the three bioinformatics approaches (Fig 3). Visual inspection of the plots revealed a striking distribution of ΔtESRseq clearly distinguishing the 3 groups of variants. More specifically, most variants with no effect on exon 10 splicing seemed to display ΔtESRseq values higher than those increasing exon skipping and lower than those increasing exon inclusion. This feature was not so clearly noticeable for ΔHZEI or ΔΨ. A statistical analysis further confirmed that ΔtESRseq values were overall concordant with the separation of variants in 3 groups according to minigene data (ANOVA test, p-value of 0.03), whereas ΔHZEI and ΔΨ were not (ANOVA test, p-values of 0.15 and 0.59, respectively) (Fig 3 and S1 Table).
To better assess the performances of the three in silico approaches, we set preliminary upper and lower thresholds for predicting variant-induced exon skipping and variant-induced exon inclusion: -0.5 and +0.5 for ΔtESRseq, -20 and +20 for ΔHZEI, and -0.05 and +0.05 for ΔΨ [19], respectively (Fig 3). As a consequence, score differences (Δ) smaller than the pre-established negative thresholds were considered indicative of increased exon skipping, whereas those higher than the positive thresholds were considered indicative of increased exon inclusion. We then determined the relative number of true calls produced by each in silico approach giving a particular attention to predictions of exon skipping, the most dreaded variant-induced splicing defect. As shown in Fig 3 and S1 Table, we observed that the ΔtESRseq approach produced the highest number of true calls, outperforming ΔHZEI and ΔΨ in discriminating variants that induce exon skipping from those that do not. More specifically, one can reckon 13 (87%) true calls for ΔtESRseq against 9 (60%) true calls for both ΔHZEI and ΔΨ. Overall, our results revealed a better sensitivity (86% versus 57%) and specificity (88% versus 63%) for ΔtESRseq than for ΔHZEI (S1 Table). In spite of displaying a level of specificity similar to ΔtESRseq (88%), ΔΨ showed a very high false negative rate (29% sensitivity). Finally, a statistical analysis further confirmed that ΔtESRseq values could discriminate variants that increased MLH1 exon 10 skipping from those that did not, whereas ΔHZEI and ΔΨ could not (t-test, p-values of 0.01, 0.11 and 0.42, respectively).
Results discordant with experimental data, such as the ΔtESRseq values obtained for MLH1 c.845C>G and c.855C>T (outliers in Fig 3) may be due to features not taken into account by the in silico approach such as the RNA secondary structure, the chromatin structure, the presence of ESRs longer than six nucleotides, or to a crosstalk with other splicing regulatory elements located nearby.
Next, we wondered if these bioinformatics methods could predict the severity of the splicing defects, i.e. if there was a correlation between the level of exon inclusion detected in the pSPL3m-M1e10 assay and the score differences produced by the ΔtESRseq, ΔHZEI and ΔΨ approaches. To answer this question, we plotted the minigene data against the score differences generated by the in silico approaches, and performed a regression analysis with both variables. As shown on Fig 4 and S1 Table, our results revealed a linear distribution of the in silico score differences as a function of exon inclusion levels for ΔtESRseq and ΔHZEI but not for the ΔΨ approach (Pearson correlation, p-values of 0.001, 0.004 and 0.93, respectively).
Then, we decided to compare the performance of these three new ESR-dedicated in silico tools with that of three previously used methods, more precisely EX-SKIP [29], ESEfinder [30,31] and HSF-SR [32]. As shown in S2 Table, EX-SKIP displayed relatively high specificity (75%) but low sensitivity (43%) in predicting exon-skipping mutations. Statistical analyses further revealed that EX-SKIP could not significantly distinguish MLH1 exon 10 variants that had an effect on splicing from those that did not (t-test and ANOVA test, p-values of 0.22 and 0.26, respectively). Still, we found that there was a correlation between the level of exon 10 inclusion and EX-SKIP values (Pearson correlation, p-value of 0.02). Data derived from ESEfinder, and especially HSF-SR, were difficult to interpret because of the presence of conflicting calls (S2 Table). Moreover, given their lack of comprehensiveness and global quantitation, the performance of these two in silico tools could not be statistically analyzed, nor properly compared to that of ΔtESRseq, ΔΔHZEI, ΔΨ or EX-SKIP.
To further evaluate the performance of the recently developed ΔtESRseq, ΔHZEI and ΔΨ-based approaches, we extended our study to the dataset that first revealed the predictive potential of ΔtESRseq [17]. This dataset includes a total of 32 BRCA2 exon 7 variants located outside the reference splice sites, some of which cause exon skipping (n = 11, suspected of altering ESRs) and some do not (n = 21), as determined experimentally. As shown in S5 Fig and S3 Table, we found again by using this dataset that ΔtESRseq displayed a slightly better sensitivity and specificity than ΔHZEI (in this case, 100% versus 91%, and 86% versus 76%, respectively) whereas ΔΨ showed very low sensitivity but high specificity (18% and 94%, respectively). Interestingly, not only ΔtESRseq but also ΔHZEI could distinguish variants that increased exon skipping from those that did not (t-test, p-values of 3.5 e-6 and 5.7 e-6, respectively). Again, ΔΨ could not discriminate these 2 groups of variants (t-test, p-value of 0.56). Moreover, we observed a statistically significant correlation between the level of BRCA2 exon 7 inclusion and the score differences produced by the ΔtESRseq and ΔHZEI approaches (Pearson correlation, p-values 1.1 e-6 and 0.9 e-3, respectively), whereas ΔΨ showed no correlation (Pearson correlation, p-value = 0.15) (S6 Fig and S3 Table). Finally, we completed our study by analyzing three additional datasets experimentally characterized by other laboratories, namely: (i) a set of 42 BRCA1 exon 6 variants [29], (ii) a set of 41 CFTR exon 12 variants [33,34] and (iii) a set of 24 NF1 exon 37 variants [35]. As shown in S4, S5 and S6 Tables, here again, we found that ΔtESRseq and ΔHZEI.had better sensitivity than ΔΨ for predicting which variants induce exon skipping (67–100% and 68–100% versus 0–33% sensitivity, respectively, depending on the dataset). Statistical analyses further highlighted the good performance of ΔtESRseq and especially ΔHZEI, but not that of ΔΨ for discriminating variants that lead to exon skipping and for predicting the severivity of the splicing defect within these 3 additional datasets (S7 Table).
We surmise that out of the three new in silico approaches expected to predict ESR-mutations [17–19], ΔtESRseq and ΔHZEI show the best performance at least for the five datasets analyzed in our study. Indeed, these two approaches displayed a better balance between sensitivity and specificity than ΔΨ, or the prior in silico method EX-SKIP, for predicting exon skipping-mutations (S8 Table). Importantly, we found that ΔtESRseq and ΔHZEI can be used to predict not only the direction but also the severity of the induced splicing defects, more negative score differences being indicative of higher exon skipping levels (S7 Table).
To evaluate the physiological pertinence of the MLH1 exon 10 minigene assays and the above described in silico predictions, we set to compare our results with data derived from the analysis of RNA obtained from carriers of MLH1 exon 10 variants, especially of those located outside splice sites. Patients’ RNA being rarely available, we had the opportunity to obtain patient RNA samples for only two SNVs of interest: (i) MLH1 c.793C>T (Patient P793CT.1) and (ii) MLH1 c.842C>T (Patient P842CT.1).
First, peripheral blood RNA of patient P793CT.1 (heterozygous for MLH1 c.793C>T) was analyzed by RT-PCR, by using primers targeting exons 8 and 12, and compared to those of three control individuals. Our results revealed a complex splicing pattern involving MLH1 exons 9, 10 and 11 in all individuals (Figs 5A and S4). These data are concordant with previous studies describing exon 10 as an alternative exon that is naturally partially skipped, either alone or in combination with exon 9 and/or exon 11 [25–27]. We also observed that, as compared to controls, the sample derived from patient P793CT.1 showed a lower amount of full-length transcripts (FL), and a higher amount of transcripts lacking exon 10 (Δ10), indicative of aberrant splicing. Sequencing of the FL products revealed the presence of WT (c.793C) exon 10 only (Fig 5A, right panel), which suggests that in this sample the exon 10-skipped products mostly derive from the mutant allele (c.793T), and that the variant-associated splicing defect is very severe. Importantly, the observation that c.793C>T is associated with a drastic splicing defect in endogenous MLH1 transcripts agrees with the data obtained in the minigene assays (Figs 1, 2 and S2), especially in the context of pSPL3m-M1e10 c.793C>T, which showed a very high level of exon skipping. To better evaluate the consequences of c.793C>T on MLH1 expression, and because Sanger sequencing is known to have low detection sensitivity, we then decided to measure the relative contribution of the WT and mutant alleles to the production of the full-length (FL) transcripts by using an allele-specific primer extension method. Our findings, shown on Fig 5B and 5C, indicate that the FL transcripts expressed from the mutant allele were in fact present in the blood cells of patient P793CT.1 but at very low level as compared to the WT allele (~10%). Given the results of our minigene assays, and RT-PCR analysis of patient’s RNA, we conclude that this allelic imbalance is mostly due to c.793C>T-induced exon 10 skipping. Moreover, if one assumes that, in the blood cells of Patient P793CT.1, both alleles are transcribed in equal amounts, then one can deduce that the pSPL3m-M1e10 assay closely reflects the effects detected in vivo in patient’s peripheral blood, at least for this variant.
We then analyzed LCL RNA from patient P842CT.1 (heterozygous for MLH1 c.842C>T) by comparing to those from 5 controls, including: 3 healthy individuals, and 2 Lynch syndrome patients (P791-5TG.1 and P882CT.1) carrying MLH1 c.791-5T>G and c.882C>T mutations directly altering the 3’ss or 5’ss of exon 10, respectively [36]. Results shown on S7 Fig indicate that patient P842CT.1 has a splicing pattern similar to that of patients P791-5TG.1 and P882CT.1, i.e. an apparent decrease in the amount of FL MLH1 transcripts, a mild increase in exon 9–10 skipping (Δ9–10) and a drastic increase in exon 10 skipping (Δ10), as compared to healthy controls. Δ10 transcripts were detected in LCLs treated with puromycin, strongly suggesting that the aberrant Δ10 out-of-frame transcripts (S4B Fig) are degraded by the NMD pathway in these cell lines. As expected, the level of Δ9–10 in-frame transcripts did not increase in the presence of the NMD-inhibitor puromycin. Sequencing of the FL RT-PCR products of patients P842CT.1 and P882CT.1 revealed the absence of c.842T and c.882T mutant FL transcripts (S7C Fig), indicating that c.842C>T and c.882C>T cause very severe splicing defects. These in vivo results clearly agree with the pSPL3m and pCAS2 minigene assays, which revealed predominant exon 10 skipping for both c.842C>T and c.882C>T (Figs 1, 2 and S2).
Importantly, these results highlight the physiological pertinence of the in silico predictions produced by the ΔtESRseq and ΔHZEI approaches. Indeed, these methods accurately predicted the variant-induced splicing aberrations observed in vivo in patients carrying exonic SNVs, as shown here for MLH1 variants c.793C>T and c.842C>T, and (ii) BRCA2 c.520C>T or c.617C>G [17,37]. As of note, the ΔtESRseq approach also accurately predicted the physiological effect of MLH1 c.794G>A [13]. We conclude that ΔtESRseq and ΔHZEI, but not ΔΨ, are promising tools for prioritizing exonic variants for splicing assays.
The present study was initiated to follow up on our observation that a large number of variants in the exon 7 of BRCA2 induce exon skipping [17]. Our hypothesis was that exonic splicing mutations were also underestimated in other exons and genes. We thus decided to analyze the impact on splicing of all SNVs identified in the exon 10 of MLH1 (n = 22), a gene implicated in Lynch syndrome. Before this study, only 5 SNVs in MLH1 exon 10 had been reported as causing aberrant splicing; more specifically, they were all shown to increase exon 10 skipping [13,24,36]. Our work not only confirmed those initial findings but, importantly, uncovered 12 new splicing mutations (8 exon skipping-, and 4 exon inclusion-mutations), bringing the number of MLH1 exon 10 splicing mutations to a total of 17. Hence, our results revealed a striking high proportion of splicing mutations in MLH1 exon 10 (77%), largely exceeding the fraction of splicing mutations detected in BRCA2 exon 7 (42%) [17]. Moreover, we found that the majority of MLH1 exon 10 splicing mutations (~60%) map outside the reference splice-site consensus sequences, indicating an important contribution of variant-induced ESR alterations in this exon.
Importantly, among the 12 new splicing mutations in MLH1 exon 10, we identified 4 SNVs causing increased exon inclusion. To our knowledge, this is the first report of SNVs having a positive impact on MLH1 splicing. Besides full-length transcripts, MLH1 is known to normally produce a fraction of transcripts lacking exon 10, such as Δ10, Δ9/10, Δ10/11 and Δ9/10/11 [25,26], with Δ9/10 being one of the most frequently reported alternative MLH1 isoforms [27]. It is possible that the 4 variants that increased exon 10 inclusion in our minigene assays also disturb MLH1 physiological alternative splicing leading to a higher production of FL transcripts and lower amount of Δ10, Δ9/10 and Δ9/10/11. Because the role of alternative MLH1 isoforms is still unknown [27], it is difficult to predict the biological and clinical consequences of these splicing alterations. Of note, a few cases of variant-induced exon inclusion have already been described in other genes. Particularly, previous studies have shown that mutations that increase exon inclusion can have a significant clinical impact. For instance, they can behave as protective factors, as is the case of SMN2 c.859G>C (p.Gly287Arg), a variant that attenuates the severity of spinal muscular atrophy (SMA) by increasing inclusion of SMN2 exon 7 [28,38]. Moreover, mutations inducing exon inclusion can also be harmful, as is the case for the majority of mutations identified in the exon 10 of the Microtubule Associated Protein Tau gene (MAPT) (reviewed in [39]). It has been shown that dysregulation of MAPT exon 10 splicing disrupts normal tau isoform ratio and leads to neurodegeneration and dementia: increased MAPT exon 10 skipping causes Pick disease, whereas increased inclusion typically causes FTDP-17 (frontotemporal dementia with parkinsonism linked to chromosome 17).
Given the challenging need in medical genetics for stratifying exonic variants for functional analyses, we decided to use the experimental data generated in this study to evaluate the predictive power of in silico tools at discerning splicing mutations. We found that the impact on splicing of the seven MLH1 exon 10 variants mapping within the splice-site consensus sequences (potential splice site-mutations) were correctly predicted by splice site-dedicated in silico tools (SSF, MES and HSF-ss). These findings confirm and extend previous studies that highlighted the good reliability of these algorithms for predicting exon skipping-mutations [12–15], further pinpointing their interest as filtering tools in variant stratification strategies. As for the variants located outside the reference splice sites, our minigene data revealed 10 ESR-mutations and 5 variants with no impact on splicing. We took advantage of these results and selected four additional experimental datasets previously described in other genes [17,29,33–35] to evaluate the discriminating power of 3 bioinformatics approaches recently described as suitable for predicting variant-induced ESR alterations [16–19]. Our findings revealed that the ΔtESRseq and ΔHZEI ESR-dedicated tools show the best performance in identifying ESR-mutations, outperforming the previous bioinformatics method EX-SKIP, whereas ΔΨ did not show compelling predictive power. Our results further indicate that both ΔtESRseq- and ΔHZEI- based approaches can predict the severity of the variant-induced splicing defects, underlining the quantitative character of these methods. It is possible that the ΔtESRseq- and ΔHZEI-based approaches produced somewhat similar results because they both rely on the appreciation of hexamer sequences as ESRs. Interestingly, a correlation between ESRseq and ZEI scores has already been reported [18] suggesting that most ESRs are indeed defined by 6-nucleotide stretches and that ESE sequences are more frequently represented in exons than in introns. Contrary to our initial expectations, ΔΨ displayed the weakest predictive power out of the 3 new in silico tools dedicated to identifying ESR alterations. This was unexpected as this method had been validated by using a large set of previously published splicing data, such as data on a plethora of MLH1 nucleotide variants [19]. Close inspection of the validation set used on that study revealed a large excess of intronic variants relative to exonic changes, and also a considerable under-representation of known ESR-mutations relative to the total number of variants included in the dataset. Most of the positive predictions reported for ΔΨ [19] thus correspond to intronic mutations directly affecting splice sites. This may explain why we detected an excess of true negative calls relative to true positive calls when using ΔΨ-values for predicting ESR-mutations in the five datasets analyzed in this study, and suggests that the ΔΨ-based method may be overall suitable for predicting mutations directly modifying splice sites but not entirely reliable for predicting those affecting ESRs. In sum, our findings suggest that both ΔtESRseq- and ΔHZEI-, based approaches can be used to stratify exonic variants for functional testing, and that this strategy may help identifying disease-causing variants. We cannot exclude that ΔΨ-values may be useful in particular conditions and, conversely, that the ΔtESRseq- and ΔHZEI-, based approaches may not be suitable for the analysis of certain exons or genes.
In the case of MLH1, it is clear that severe exon 10 skipping causes Lynch syndrome (reviewed in [27]). Skipping of MLH1 exon 10 leads to a shifted reading frame, resulting in a premature stop codon in exon 11 (MLH1 p.His264Leufs*2) and probable degradation of the aberrant transcripts by NMD. Variants inducing total exon 10 skipping cause a drastic loss in FL MLH1 protein and are therefore considered deleterious. In contrast, the clinical significance of variants inducing partial exon 10 skipping is still unknown. First, it is unclear if the amount of FL transcripts produced in the presence of such variants is enough to fulfill MLH1 function. Second, it is possible that remaining FL MLH1 transcripts carrying missense variants lead to production of nonfunctional proteins. Additional analyses are thus necessary to determine the biological and clinical significance of partial exon skipping-variants, including protein assays, assessment of patient clinical history and family data. Given that exon 10 codes for part of an important domain of the MLH1 protein (interaction with MUTSα) [40], we suspect that SNVs increasing exon 10 inclusion can either have a protective impact if co-occurring with variants showing the opposite effect, or a deleterious effect if introducing a missense change that severely impairs protein function. Thus, in the absence of further information, MLH1 missense SNVs inducing exon 10 inclusion, as well as those not affecting splicing, should be considered as variants of unknown significance (VUS). In contrast, the clinical classification of synonymous substitutions not affecting RNA splicing can eventually evolve from VUS to likely not pathogenic depending on expert panel assessment [22,23].
In conclusion, our results revealed an unexpected high number of splicing mutations in the exon 10 of MLH1, most of which affecting potential ESRs, and confirmed the predictive power of ΔtESRseq- and ΔHZEI-based approaches for pinpointing this type of mutations, at least in MLH1 exon 10, BRCA2 exon 7, BRCA1 exon 6, CFTR exon 12 and NF1 exon 37. In principle, the bioinformatics methods described in our study are amenable to automation and, as such, have the potential to be used as filtering tools for identifying disease-causing candidates among the large number of variants detected by high-throughput DNA sequencing.
Written informed consent was obtained from all individuals.
We collected all SNVs reported in the exon 10 of MLH1, until January 2013, by interrogating the following public databases: UMD-MLH1 (Universal Mutation Database-MLH1, http://www.umd.be/MLH1/) [22], LOVD (Leiden Open Variation Database, http://chromium.liacs.nl/LOVD2/colon_cancer/variants.php?select_db=MLH1), dbSNP (the Single Nucleotide Polymorphism database http://www.ncbi.nlm.nih.gov/SNP/), and UniProtKB/Swiss-Prot (the European protein sequence database, http://swissvar.expasy.org/cgi-bin/swissvar/result?global_textfield=MLH1) (Table 1 and Fig 1A).
Nucleotide numbering is based on the cDNA sequence of MLH1 (NCBI accession number NM_000249.3), c.1 denoting the first nucleotide of the translation initiation codon, as recommended by the Human Genome Variation Society.
In order to evaluate the impact on splicing of each MLH1 exon 10 variant, we performed functional assays based on the comparative analysis of the splicing pattern of wild-type and mutant MLH1 reporter minigenes. These minigenes were prepared by using two different vectors: pSPL3m and pCAS2. The pSPL3m plasmid, a modified version of the exon-trapping vector pSPL3 (Invitrogen) which in turn derives from pSPL1 [41], carries two chimeric exons (here named I and II, both containing rabbit β-globin and HIV Tat sequences) separated by an intron containing BamHI and MluI cloning sites (Fig 1B) [13]. Expression of the pSPL3m minigene is driven by the SV40 promoter. The pCAS2 vector carries two exons (here named A and B) with a sequence derived from the human SERPING1/C1NH gene, separated by an intron with BamHI and MluI cloning sites (S2A Fig). Expression of the pCAS2 minigene is under the control of a CMV promoter. The pCAS2 is a modified version of the previously described pCAS1 plasmid [13,42]. Two modifications were introduced into the exon A of pCAS2 relative to pCAS1: (i) the first 114 bp of exon A were deleted, and (ii) the SERPING1/CINH translation initiation codon was disrupted by replacing the sequence GATG (initiation codon underlined) by TCAC.
The wild-type genomic fragment MLH1 c.791-168_c.884+187 (MLH1 exon 10 and flanking intronic sequences) was inserted into the BamHI and MluI cloning sites of the reporter plasmids pSPL3m and pCAS2, yielding the three-exon hybrid minigenes pSPL3m-M1e10 and pCAS2-M1e10, respectively (Figs 1B and S2A). Minigenes carrying MLH1 exon 10 variants were prepared by site-directed mutagenesis by using the two-stage overlap extension PCR method [43] and the wild-type pSPL3m-M1e10 construct as template. Then, the mutant amplicons were digested with BamHI and MluI, and introduced into BamHI and MluI cloning sites of the pSPL3m-M1e10 minigene to replace the wild-type fragment. The inserts of all constructs were sequenced to ensure that no other mutations had been introduced during the cloning process. In some cases, as indicated (S2 Fig), mutant inserts were digested from the pSPL3m-M1e10 minigene with BamHI and MluI and then subcloned into pCAS2.
Next, wild-type and mutant minigenes (1μg/well) were transfected in parallel into HeLa cells grown at ~60% confluence in 6-well plates using the FuGENE 6 transfection reagent (Roche Applied Science). HeLa cells were cultivated in Dulbecco’s modified Eagle medium (Life Technologies) supplemented with 10% fetal calf serum in a 5% CO2 atmosphere at 37°C. Total RNA was extracted 24 hours after transfection using the NucleoSpin RNA II kit (Macherey Nagel) according to the manufacturer’s instructions. Then, the minigene transcripts were analyzed by semi-quantitative RT-PCR (30 cycles of amplification) in a 25 μl reaction volume by using the OneStep RT-PCR kit (Qiagen), 100 ng total RNA, and pSPL3m- or pCAS2-appropriate forward and reverse primers (SD6 and SA2 or pCAS-KO1-F and pCAS-2-R, respectively, as described in [13] and [37]). RT-PCR products were separated by electrophoresis on 2.5% agarose gels containing ethidium bromide and visualized by exposure to ultraviolet light under non-saturating conditions using the Gel Doc XR image acquisition system (Bio-RAD). Semi-quantitative analysis, gel extraction and sequencing of the RT-PCR products were carried out as previously described [42].
MLH1 exon 10 fragments (~30 bp-long) were analyzed for their splicing enhancer properties by performing a functional assay based on the splicing pattern of the pcDNA-Dup minigene [13]. This vector contains a β-globin-derived three-exon minigene with a middle exon particularly sensitive to the presence of exonic splicing regulatory signals. Expression of the minigene is under the control of the CMV promoter (Fig 2B). The exonic fragments of interest, wild-type or mutant, were obtained by annealing complementary 5’-phosphorylated oligonucleotides carrying 5’-EcoRI and 3’-BamHI compatible ends. Then, the exonic segments were inserted into the EcoRI and BamHI cloning sites of the middle exon of pcDNA-Dup to produce hybrid pcDNA-Dup-M1e10-R minigenes. All constructs were sequenced to ensure that no other mutations were introduced into the middle exon during the cloning process. Transfection, RNA extraction and RT-PCR analysis were performed as described for the splicing minigene reporter assay except that RT-PCR reactions were performed with 150 ng total RNA as template, T7-Pro (5’-TAATACGACTCACTATAGG-3’) and Dup-S4-Seq-3R (5’-CGTGCAGCTTGTCACAGTGC-3’) as forward and reverse primers respectively, and 28 cycles of amplification. RT-PCR products were analyzed by electrophoresis as described above for the splicing minigene reporter assay.
Three in silico tools were used to predict variant-induced alterations in 3’ and 5’ splice site strength, namely: SpliceSiteFinder-like (SSF, http://www.interactive-biosoftware.com), MaxEntScan (MES, http://genes.mit.edu/burgelab/maxent/Xmaxentscan_scoreseq.html; Maximum Entropy Model) and the splice site module of Human Splicing Finder (here named HSF-ss for splice site-dedicated HSF, http://www.umd.be/HSF/). These algorithms were interrogated simultaneously by using the integrated software tool Alamut (Interactive Biosoftware, France, http://www.interactive-biosoftware.com), as previously described [17].
Three newly developed in silico approaches were used to predict variant-induced alterations in exonic splicing regulatory elements (ESRs): (i) calculation of total ESRseq score changes (ΔtESRseq) by using the method previously described by our group [17] with a small modification (here, only exonic positions were taken into account), (ii) calculation of ΔHZEI values by using the HEXplorer method [18], and (iii) assignment of ΔΨ values based on the Splicing Regulatory Model (http://tools.genes.toronto.edu) recently described by Xiong and co-workers [19]. Moreover, as indicated, we also resorted to three previously established ESR-dedicated in silico tools: (i) EX-SKIP (http://ex-skip.img.cas.cz/) [29] in which we took into account the full nucleotide sequence of the exon of interest, (ii) ESEfinder (http://rulai.cshl.edu/cgi-bin/tools/ESE3/esefinder.cgi?process=home) [30,31], and (iii) the ESR module of Human Splicing Finder (here named HSF-SR for ESR-dedicated HSF, http://www.umd.be/HSF3/) [32].
Results are presented as the mean ± SD of three independent experiments. Data derived from comparisons of experimental and in silico analyses were compared by using either the one-way ANOVA test or the Student’s t-test, and the Pearson’s correlation coefficient, as indicated. More specifically, the ANOVA test was used for assessing the performance of the bioinformatics tools in discriminating 3 groups of variants (i.e. variants that increase exon skipping versus those with no effect on splicing versus those that increase exon inclusion), whereas Student’s t-test was used when only 2 groups of variants were taken into account (i.e. variants that increase exon skipping versus those that do not). Correlation between exon inclusion levels and in silico predictions was measured by calculating Pearson correlation coefficients (r). p-values and r are indicated in the figures. Results were considered significant when p-value <0.05. Statistical tests were performed by using BiostaTGV (http://marne.u707.jussieu.fr/biostatgv/).
The power to distinguish mutations that induce exon skipping from those that do not was further assessed, for each ESR-dedicated in silico method, by calculating sensitivity and specificity values (true positives x 100/(true positives + false negatives) and (true negatives x 100/(true negatives + false positives), respectively). Sensitivity and specificity were determined by taking into account the following thresholds: -0.5 for ΔtESRseq (arbitrary threshold), -20 for ΔHZEI (arbitrary threshold), and -0.05 for ΔΨ (threshold previously established by the authors, [19].
Peripheral blood samples were directly collected into PAXgene Blood RNA Tubes (Qiagen) from which total RNA was extracted by using the PAXgene Blood RNA kit, according to the manufacturer’s instructions. EBV-immortalized lymphoblastoid cell lines (LCLs) were cultivated in RPMI medium (Life Technologies) supplemented with 2 mM of L-glutamine and 10% fetal calf serum, at 37°C in a 5% CO2 atmosphere. Before RNA extraction, LCLs were transferred into 6-well plates, at 2.5x106 cells/well, and incubated for 5.5 hours with/without 200 μg/ml puromycin. Then, total RNA was extracted by using the NucleoSpin RNA II kit (Macherey Nagel). Written informed consent was obtained from all individuals.
The splicing pattern of MLH1 transcripts expressed in peripheral blood and in LCLs was analyzed by semi-quantitative RT-PCR using the OneStep RT-PCR kit (Qiagen) in 25 μl-final volume reactions containing 100 ng of total RNA, a forward primer located in MLH1 exon 8 (MLH1-RT-8Fbis, 5’-AAGGAGAGACAGTAGCTGATGTT-3’) and a reverse primer located in exon 12 (MLH1-12R, 5’-TGCTCAGAGGCTGCAGAAA-3’). To ensure that the assay was in the linear range, RT-PCR reactions were performed with 34 cycles of amplification (S4 Fig). Then, RT-PCR products were separated by electrophoresis on a 2% agarose gel, gel-purified and sequenced.
Allele specific expression (ASE) was measured by performing a SNaPshot assay (ABI Prism SNaPshot, Fig 5B). First, RT-PCR products spanning MLH1 exons 8 to 12 were obtained, as described above, from a peripheral blood RNA sample of a patient carrying the heterozygous c.793C>T substitution in MLH1 exon 10 (Patient P793CT.1). In parallel, a segment encompassing MLH1 exon 10 was amplified by PCR from the genomic DNA of the same patient by using the Multiplex PCR kit (Qiagen) according to the manufacturer’s instructions. Briefly, PCR reactions (35 cycles of amplification) were performed in a final volume of 25 μl with 100 ng of genomic DNA as template, a forward primer in MLH1 intron 9 (MLH1-10-Bam-F, 5’-GACCGGATCCTTGGAAAGTGGCGACAGG-3’) and a reverse primer in intron 10 (MLH1-10-Mlu-R, 5’-GACCACGCGTAATTAGTGAATAAATGAAGGAAAA-3’). Then, 5 μl aliquots of RT-PCR and PCR products were treated with one unit of Shrimp Alkaline Phosphatase (SAP, USB) and 8 units of Exonuclease I (Thermo Scientific) in the presence of SAP buffer in a final volume of 10 μl. After 1 hour at 37°C, the reactions were terminated by incubating at 75°C for 15 minutes. Next, 2 μl aliquots of treated RT-PCR and PCR products were subjected to SNaPshot reactions, in a final volume of 10 μl, by using the SNaPshot Multiplex Kit (Applied Biosystems) and a reverse primer targeting the sequence immediately downstream MLH1 c.793C>T (SNAP-M1.793-R, 5’-GGAAGTTGATTCTACCAGAC-3’). SNaPshot reactions were carried out by performing 25 cycles of primer extension (denaturation at 96°C for 10 sec, annealing at 50°C for 5 sec and elongation at 60°C for 30 sec). Next, the reactions were incubated with 1 unit of SAP at 37°C for 1 hour, and terminated at 75°C for 15 minutes. Finally, the extension products were separated by electrophoresis and analyzed quantitatively by using an ABI PRISM-3100 Genetic Analyzer (Applied Biosystems). SNaPshot results obtained from patient cDNA were normalized by those obtained from gDNA.
|
10.1371/journal.pgen.1003005 | Mining the Unknown: A Systems Approach to Metabolite Identification Combining Genetic and Metabolic Information | Recent genome-wide association studies (GWAS) with metabolomics data linked genetic variation in the human genome to differences in individual metabolite levels. A strong relevance of this metabolic individuality for biomedical and pharmaceutical research has been reported. However, a considerable amount of the molecules currently quantified by modern metabolomics techniques are chemically unidentified. The identification of these “unknown metabolites” is still a demanding and intricate task, limiting their usability as functional markers of metabolic processes. As a consequence, previous GWAS largely ignored unknown metabolites as metabolic traits for the analysis. Here we present a systems-level approach that combines genome-wide association analysis and Gaussian graphical modeling with metabolomics to predict the identity of the unknown metabolites. We apply our method to original data of 517 metabolic traits, of which 225 are unknowns, and genotyping information on 655,658 genetic variants, measured in 1,768 human blood samples. We report previously undescribed genotype–metabotype associations for six distinct gene loci (SLC22A2, COMT, CYP3A5, CYP2C18, GBA3, UGT3A1) and one locus not related to any known gene (rs12413935). Overlaying the inferred genetic associations, metabolic networks, and knowledge-based pathway information, we derive testable hypotheses on the biochemical identities of 106 unknown metabolites. As a proof of principle, we experimentally confirm nine concrete predictions. We demonstrate the benefit of our method for the functional interpretation of previous metabolomics biomarker studies on liver detoxification, hypertension, and insulin resistance. Our approach is generic in nature and can be directly transferred to metabolomics data from different experimental platforms.
| Genome-wide association studies on metabolomics data have demonstrated that genetic variation in metabolic enzymes and transporters leads to concentration changes in the respective metabolite levels. The conventional goal of these studies is the detection of novel interactions between the genome and the metabolic system, providing valuable insights for both basic research as well as clinical applications. In this study, we borrow the metabolomics GWAS concept for a novel, entirely different purpose. Metabolite measurements frequently produce signals where a certain substance can be reliably detected in the sample, but it has not yet been elucidated which specific metabolite this signal actually represents. The concept is comparable to a fingerprint: each one is uniquely identifiable, but as long as it is not registered in a database one cannot tell to whom this fingerprint belongs. Obviously, this issue tremendously reduces the usability of a metabolomics analyses. The genetic associations of such an “unknown,” however, give us concrete evidence of the metabolic pathway this substance is most probably involved in. Moreover, we complement the approach with a specific measure of correlation between metabolites, providing further evidence of the metabolic processes of the unknown. For a number of cases, this even allows for a concrete identity prediction, which we then experimentally validate in the lab.
| Recently, genome-wide association studies (GWAS) on metabolic quantitative traits have proven valuable tools to uncover the genetically determined metabolic individuality in the general population [1]–[5]. Interestingly, a great portion of the genetic loci that were found to significantly associate with levels of specific metabolites are within or in close proximity to metabolic enzymes or transporters with known disease or pharmaceutical relevance. Moreover, compared to GWAS with clinical endpoints the effect sizes of the genotypes are exceptionally high.
The number and type of the metabolic features that went into these GWAS was mainly defined by the metabolomics techniques used: Gieger et al. [1] and Illig et al. [2] used a targeted mass spectrometry (MS)-based approach giving access to the concentrations of 363 and 163 metabolites, respectively. Suhre et al. [3] and Nicholson et al. [4] applied untargeted nuclear magnetic resonance (NMR) based metabolomics techniques, yielding 59 metabolites that had been identified in the spectra prior to the GWAS and 579 manually selected peaks from the spectra, respectively. In Suhre et al. [5], 276 metabolites from an untargeted MS-based approach were analyzed.
While these previous GWAS focused on metabolic features with known identity, untargeted metabolomics approaches additionally provide quantifications of so-called “unknown metabolites”. An unknown metabolite is a small molecule that can reproducibly be detected and quantified in a metabolomics experiment, but whose chemical identity has not been elucidated yet. In an experiment using liquid chromatography (LC) coupled to MS, such an unknown would be defined by a specific retention time, one or multiple masses (e.g. from adducts), and a characteristic fragmentation pattern of the primary ion(s). An unknown observed by NMR spectroscopy would correspond to a pattern in the chemical shifts. Unknowns may constitute previously undocumented small molecules, such as rare xenobiotics or secondary products of metabolism, or they may represent molecules from established pathways which could not be assigned using current libraries of MS fragmentation patterns [6], [7] or NMR reference spectra [8].
The impact of unknown metabolites for biomedical research has been shown in recent metabolomics-based discovery studies of novel biomarkers for diseases and various disease-causing conditions. This includes studies investigating altered metabolite levels in blood for insulin resistance [9], type 2 diabetes [10], and heart disorders [11]. A considerable number of high-ranking hits reported in these biomarker studies represent unknown metabolites. As long as their chemical identities are not clarified the usability of unknown metabolites as functional biomarkers for further investigations and clinical applications is rather limited.
In mass-spectrometry-based metabolomics approaches, the assignment of chemical identity usually involves the interpretation and comparison of experiment-specific parameters, such as accurate masses, isotope distributions, fragmentation patterns, and chromatography retention times [12]–[14]. Various computer-based methods have been developed to automate this process. For example, Rasche and colleagues [15] elucidated structural information of unknown metabolites in a mass-spectrometry setup using a graph-theoretical approach. Their approach attempts to reconstruct the underlying fragmentation tree based on mass-spectra at varying collision energies. Other authors excluded false candidates for a given unknown by comparing observed and predicted chromatography retention times [16], [17], or by the automatic determination of sum formulas from isotope distributions [18]. Furthermore, Gipson et al. [19] and Weber et al. [20] integrated public metabolic pathway information with correlating peak pairs in order to facilitate metabolite identification. However, these methods might not be applicable for high-throughput metabolomics datasets that have been produced in a fee-for-service manner, since the mass spectra as such might not be readily available.
Approaching the problem from a conceptually different perspective, we here present a novel functional metabolomics method to predict the identities of unknown metabolites using a systems biological framework. By combining high-throughput genotyping data, metabolomics data, and literature-derived metabolic pathway information, we generate testable hypotheses on the metabolite identities based solely on the obtained metabolite quantifications (Figure 1). No further experiment-specific data such as retention times, isotope patterns and fragmentation patterns are required for this analysis.
The concept of our approach is based on the following observations from our previous work on genome-wide association studies and Gaussian graphical modeling (GGM) with metabolomics: We showed that GWAS with metabolic traits can reveal functional relationships between genetic loci encoding metabolic enzymes and metabolite concentration levels in the blood [1]–[3], [5]. A genetic variant can alter, for instance, the expression levels of mRNAs or affect the properties of the respective enzymes through changes of the protein sequence (e.g. enzyme activity, substrate specificity). Moreover, we found that GGMs, which are based on partial correlation coefficients, can identify biochemically related metabolites from high-throughput metabolomics data alone [21], [22]. These observations suggest that if an unknown compound displays a similar statistical association with a genetic locus in a GWAS or a known metabolite in a GGM, then this may provide specific information of where it is located in the metabolic network. Based on this information we can then derive testable hypotheses on the biochemical identity of the unknown metabolite. This annotation idea parallels classical concepts from functional genomics, where, for instance, co-expression between RNA transcripts is used to predict the function of poorly characterized genes [23], [24].
The manuscript is organized as follows: We first conduct a full genome-wide association study on 655,658 genotyped SNPs with concentrations of 225 unknown metabolites using fasting blood serum samples from a large German population cohort (n = 1768) [25]. We thereby extend our previous work on known metabolites [5] to a GWAS with hitherto unpublished unknown metabolic traits. We then compute a Gaussian graphical model including both known and unknown metabolites. In a third step, we integrate the results of the GWAS and GGM computations and combine them with metabolic pathway information from public databases to derive predictions for a total of 106 unknown metabolites. In order to validate the approach, we investigate six distinct cases, in which we derive specific identity predictions for a total of nine unknown metabolites, which we then confirm experimentally. Finally, we discuss the relevance of newly discovered genetic loci and unknown identity predictions in the context of existing disease biomarker discovery and pharmacogenomics studies.
All GWAS and GGM results, unknown metabolite classifications and pathway annotations are available as spreadsheets and in .graphml format in Dataset S1 or from our study website at http://cmb.helmholtz-muenchen.de/unknowns.
In the first step of our analysis, we conducted a GWAS with the concentrations of known and unknown metabolites, testing a total of 655,658 genotyped SNPs from the KORA cohort for association. Thus, in addition to the unknown metabolite data, we included the association data for known metabolites from our previous study [5] into the present analysis. Unknown metabolites are uniquely labeled in the format “X-12345”, which are identical throughout all published studies that use the Metabolon platform.
In total, we observe 34 distinct loci that display metabolite associations at a genome-wide significance level (Figure 2 and Dataset S1). Out of these 34 loci, 15 associate with at least one unknown compound. For 12 loci, an unknown compound constitutes the strongest association of all tested compounds. From the 213 unknown metabolites analyzed (see Methods for the determination of this metabolite subset), 28 show at least one genome-wide significant hit. These 28 associations at the 15 loci are presented in Table 1 along with all previously described GWAS hits to metabolic traits or other endpoints. Associating traits were determined from the GWAS catalog [26] for SNPs in LD (r2≥0.5) with the respective lead SNP. Seven of the 15 loci (SLC22A2, COMT, CYP3A5, CYP2C18, GBA3, UGT3A1, rs12413935) have not been described in GWAS with metabolic traits before and thus represent new genetic loci of metabolic individuality. Interestingly, genetic variants in strong LD with CYP2C18 have been reported to associate with warfarin maintenance dose [27].
In our previous GWAS with metabolic traits, we observed that metabolites associating with genetic variants in or near enzymes are likely to be functionally linked to these proteins. A SNP with detectable effects on the metabolome will, for instance, alter expression levels of mRNAs, or affect the properties of the respective enzymes (e.g. enzyme activity, substrate specificity) through modifications of the protein sequence. As an example of the latter case, the SNP rs4343 in the angiotensin converting enzyme (ACE) encoding gene was found to be associated with the activity of the enzyme [28] (See Table 1). To estimate the contribution of the first case, we compared our significant SNPs with expression quantitative trait loci (eQTLs) from published GWAS with expression levels. To this end, we queried the Genotype-Tissue Expression (GTEx) browser, an online eQTL database of the NIH GTEx roadmap project, which stores eQTL results for multiple human tissues (liver, lymphoblastoid, brain) [29]. For seven SNPs in three distinct loci (PYROXD2, CYP3A5, SPPL3), we found significant cis-eQTLs (p-value<2.7×10−9, see Methods) in GTEx. All identified eQTLs with p-values below 10−5 are listed in Dataset S1.
Based on the observation that SNPs in or in the vicinity of enzymes are mostly associating with functionally related metabolites in case of the knowns, we used the GWAS data to derive hypotheses on the potential identity of the respective unknowns. For instance, the SNP rs296391 in close proximity to the SULT2A1 gene (sulfotransferase family, cytosolic, 2A, dehydroepiandrosterone DHEA-preferring) strongly associates with the concentrations of the unknown metabolites X-11440 and X-11244 (p = 1.7×10−43 and p = 2.1×10−26, respectively). The enzyme encoded by SULT2A1, a bile salt sulfotransferase, converts steroids and bile acids into water-soluble sulfate conjugates for excretion [30]. Thus, we may speculate that X-11440 and X-11244 are biochemically related to steroids, bile acids, or water-soluble sulfate conjugates. Additional insights can be gained from genetic associations that involve both known and unknown metabolites. For instance, X-12510, X-11787, X-12093 and N-acetylornithine strongly associate with genetic variation at the NAT8 locus. NAT8 encodes the protein N-acetyltransferase 8. In this case, we may speculate that the unknowns represent similar substrates or products of the N-acetylation processes linked to this enzyme. Finally, we can link the results obtained here with results from other GWAS on metabolic traits. For example, the unknown metabolite X-13431 associates with a genetic variant in the ACADL (acyl-CoA dehydrogenase, long-chain) gene. This locus does not associate with any other metabolite in the present study, but was previously reported to associate with the medium-chain length carnitines C9 and C10:1 [1], [2]. Proteins from the ACAD family catalyze rate-limiting reactions in the β-oxidation pathway which generally associate with carnitines. This observation suggests that X-13431 may be a member of this medium-chain length carnitine family. These examples demonstrate that concrete information on the biochemical identity of unknown metabolites can be derived from our experimental dataset by using the GWAS approach.
In the second step of our analysis we focused solely on intrinsic relations between the measured metabolites and, in particular, on associations between known and unknown compounds. To this end, we applied Gaussian graphical models (GGMs), which we have previously shown to be able to reconstruct pathways involving directly related metabolites from cross-sectional blood serum metabolomics data [21], [22]. GGMs are based on partial correlation coefficients, that is, correlations between pairs of metabolites corrected for the effects of all remaining metabolites. Each known metabolite is annotated with a “super-pathway” corresponding to its general metabolic class, and a “sub-pathway” representing more specific metabolic pathways (see Dataset S1). In order to obtain a dataset that is independent of our genetic analysis, and to avoid circular arguments, co-variations in metabolite concentrations that are due to association with genetic variants (SNPs) were specifically removed from the data (see GGM methods for further details). A partial correlation was included in the model if it was significantly different from zero with α = 0.05 after Bonferroni correction, yielding a corrected significance level of = 7.9×10−7 and an absolute partial correlation cutoff of ζ = 0.178. The resulting GGM consists of a total of 399 out of 62,835 theoretically possible edges (0.64% connectivity, Figure 3A). In line with our previous observations [21], metabolites tend to be strongly connected within their respective metabolic class, while links between different classes are rare (see Text S1). Inspecting the GGM in detail, we observe that the unknowns are tightly integrated within the network and connected to known compounds of various metabolic classes. This is reflected both in the overall network (Figure 3A, Text S1) and in the top list of high-scoring GGM edges (Table 2), where 18 of the 30 strongest partial correlations comprise at least one unknown metabolite. The highest partial correlation in the dataset actually involves a known-unknown metabolite pair, namely 3-indoxylsulfate and the unknown metabolite X-12405 (ζ = 0.840). For pairs of known metabolites, we consistently observe associations of biochemically related metabolites from various metabolic pathways, such as the metabolites inosine and guanosine (ζ = 0.798), which are involved in nucleotide metabolism, or androsterone sulfate and epiandrosterone sulfate (ζ = 0.755), which represent related steroid hormone metabolites. Other pathways with related metabolite pairs include amino acid metabolism, lipid metabolism, bile acid metabolism, and xanthine metabolism. Following our line of reasoning, correlating pairs of a known and an unknown metabolite then directly point to specific pathways of cellular metabolism on which the unknown metabolite may lie. The investigation of the sub-network structure around the unknown compounds provides additional biochemical context for that compound.
We selected four high-scoring sub-networks in the GGM to show that this concept is indeed applicable to real data. The first two of these sub-networks consist of a series of intermediate compounds from purine metabolism, including guanosine, inosine, xanthine derivatives and urate (Figure 3B and 3C). In these cases, one can actually follow the addition and removal of chemical groups by following the edges in the GGM network: Most edges in these sub-networks correspond to the change of either a single methyl group at the purine double-ring structure or to the removal of a ribose residue in the reaction from nucleosides to xanthine variants. While the compounds in both sub-networks appear structurally similar, the distinction into two groups by the GGM is indeed biochemically sound. The metabolites in Figure 3B correspond to endogenous substances in the nucleoside pathway, whereas the molecules in Figure 3C relate to signals from xenobiotic metabolism of drugs and caffeine. Here, the unknown metabolites X-11422 and X-10810, as well as X-14473 and X-14374 are prominently placed in the networks, making them direct targets for closer inspection with respect to endogenous xanthines and xenobiotics, respectively.
The third sub-network comprises three androsterone sulfate variants, which belong to the class of steroid hormones (Figure 3D). We observe direct GGM links between the unknowns X-11450, X-11244 and X-11443 with both dehydroepiandrosterone sulfate (DHEAS) and epiandrosterone sulfate, suggesting androsterone derivatives as likely candidates for these three metabolites. The fourth sub-network involves different stereoisomers of bilirubin, which is the degradation product of the oxygen transporter hemoglobin [31] (Figure S1). In this sub-network, we observe high partial correlations between the bilirubin variants and a series of unknown metabolites (X-11441, X-11530, X-11442, X-11793, X-11809, X-14056, and X-14057). The seven unknown compounds in this GGM sub-network are thus likely to be involved in hemoglobin degradation processes. Taken together, the examples confirm that further information on the biochemical identity of unknown metabolites can be extracted from GGM networks.
The next step in our analysis was the integration of the GGM and GWAS approaches with general pathway information from external databases, in order to generate concrete predictions for the unknowns' metabolic pathway memberships. As a feasibility test, we first asked whether the local neighborhood of a metabolite in the GGM can be used to correctly predict its metabolic class. Using a majority-voting based classifier and subsequent permutation testing, we detected significant classification abilities (mean sensitivity 0.674, mean specificity 0.84, macro-averaged F1 score 0.72) far beyond random (p<10−8,). Detailed results can be found in Text S2. Note that we performed this approach only to demonstrate the systematic possibility to derive functional information from the GGM. The actual classification of the unknowns in the following will not be based on majority voting, but rather on the collection of all available functional information from GGM neighbors and GWAS hits.
We combined functional annotations for both GGM neighbors and GWAS hits for each unknown in order to derive specific pathway classifications. For unknowns that did not have a known metabolite neighbor in the GGM, we also investigated the 2- and 3-neighborhoods. Since these hits certainly represent weaker evidence than a direct GGM neighbor, we distinguish between ‘GGM hit’ and ‘direct GGM hit’ in the following. Functional annotations were obtained from three sources: (1) The sub-pathway assignment provided for each known metabolite in the GGM neighborhood, (2) the GO functional terms for the associated gene of all genome-wide significant GWAS hits, and (3) the KEGG pathways on which the associated genes lie. To the best of our knowledge, there is presently no consistent mapping between annotations from the different data sources available for both metabolites and genes, so we here had to perform the only non-automatic step in the analysis: By manual interpretation of different functional classes (Figure 4A), we derive a single consensus pathway annotation for a total of 106 of the unknown metabolites (Figure 4B). For 98 unknowns, we obtained annotations from the GGM network, with 74 of these hits representing direct GGM hits. From the 28 genetic hits introduced above, 27 were in a genetic region with gene annotation. Overlaying the direct edge GGM set and the GWAS set, we obtained 16 unknowns with both biochemical and genetic evidence (Figure 4C). A list of all functional evidence along with the respective predictions can be found in Table S1.
In the following, we selected several unknowns that were forwarded to detailed analysis and experimental validation. Five cases were obtained from the set of 16 high-confidence predictions in the previous section, since the combined evidence from GWAS and GGMs provides rich functional annotations that allow to derive possible compound candidates. Moreover, in order to demonstrate the power of GGMs in the absence of genetic associations, we selected one further case (HETE) where publically available pathway information was systematically exploited. Experimental validations were performed by running pure candidate compounds on the LC-MS/MS platform. For cases where no pure compound was available, we determined exact molecule masses and revisited the retention times and fragmentation spectra.
We investigated six metabolic scenarios in-depth and attempted experimental confirmation of the respective predictions (Table 3). In the following, we discuss three example cases, termed DIPEPTIDE, STEROID, and HETE (Figure 5). Three further examples, named CARNITINE, BILIRUBIN, and ASCORBATE, are presented as Text S3. In the discussion of these scenarios we now use all available evidence, the metabolite correlations, genetic associations, biochemical data, and in addition the molecular masses reported with the known and unknown compounds (which do not represent exact masses at this point). Note that the presented scenarios represent the only cases where a detailed investigation has been attempted. Moreover, the candidate compounds mentioned in the following paragraphs and the supplementary material are the only compounds that have been experimentally tested (there are no negative results not reported in this text).
We developed and validated a novel integrative approach for the biochemical characterization of “unknown metabolites” from high-throughput metabolomics and genotyping datasets. Our method allows for the functional annotation of previously unidentified metabolites and, as a consequence, enhances the interpretability of metabolomics data in genome-wide association studies and biomarker discovery. For the first time, we systematically evaluated genetic associations of unknown metabolites, thereby discovering seven new loci of metabolic individuality. By classifying a series of unknown metabolites, we gained new insights into the functional interplay between genetic variation and the metabolome both for previously reported and new loci. Furthermore, several of the unknown compounds that we identified as well as their newly associated loci were independently reported in disease-related studies. In the following, we discuss three genetic loci and their associated phenotypes.
The first example is a recent biomarker study, where Milburn et al. [34] reported an association of X-11593 with hepatic detoxification. In our GWAS, we find a strong association of X-11593 with the COMT locus, which encodes the catechol-O-methyltransferase enzyme. COMT is responsible for the inactivation of catecholamines such as L-dopa and various neuroactive drugs by O-methylation [35]. Following our identification approach, we experimentally confirmed the identity of X-11593 as O-methylascorbate. Notably, O-methylascorbate is a known product of ascorbate (vitamin C) O-methylation by COMT [36], [37]. Thus, our observations establish a link between O-methylascorbate blood levels, common genetic variation in the COMT locus and COMT-mediated liver detoxification processes.
The second example relates to the ACE gene locus, which is a known risk locus for cardiovascular disease, hypertension and kidney failure. The protein encoded by the ACE locus, angiotensin-converting enzyme, is an exopeptidase which cleaves dipeptides from vasoactive oligopeptides, and plays a central role in the blood pressure-controlling renin-angiotensin system [38]. Moreover, the ACE protein is a target for various pharmaceuticals (ACE inhibitors), especially in the treatment of hypertension [39]. In our study, we identified three unknowns as dipeptides (X-14205, X-14208 and X-14478), two of which also associated with the ACE locus. These dipeptides could thus represent novel, interesting biomarkers for the activity of ACE. Moreover, Steffens et al. [11] reported a connection between heart failure and X-11805, which is in close proximity to angiontensin-related peptides in the GGM. This connection might be revisited after a successful identification of X-11805 in a future study.
The third example is an explorative study to detect biomarkers for insulin sensitivity. Gall et al. [9] reported several known metabolites (most prominently α-hydroxybutyrate) as biomarkers for insulin resistance. They also reported a series of unknown metabolites among their top hits. In the present study, we investigated three of these unknowns: X-11793 associates with UGT1A (UDP glucuronosyltransferase 1) and represents a bilirubin-related substance. Moreover, we experimentally validated X-11421 and X-13431, which display a strong association with ACADM (acyl-Coenzyme A dehydrogenase, C-4 to C-12 straight chain), as acylcarnitines containing 10 and 9 carbon atoms, respectively. The identification of these latter two unknown metabolites as medium-chain length acylcarnitines is coherent with reports by Adams et al. [40]. The authors found elevated blood plasma acylcarnitine levels in women with type 2 diabetes. Functionally, they attributed this finding to incomplete β-oxidation. Thus, our identification of X-11421 and X-13431 now suggests incomplete β-oxidation as an explanation for the associations found by Gall et al. and implies that acylcarnitines containing 10 and 9 carbon atoms are potential biomarkers for insulin resistance.
In summary, we integrated high-throughput metabolomics and genotyping data from a large population cohort for elucidating the biochemical identities of unknown metabolites. To this end, we applied metabolomics genome-wide association studies and Gaussian graphical modeling in order to link these unknown metabolites with known metabolic classes and biological processes. For six specific scenarios, we went from systematic hypothesis generation over detailed investigation and identity prediction to direct experimental confirmation. Similar validations may now be undertaken for the remaining predictions that we report in Table S1. Finally, we demonstrated the benefit of our method by discussing several of these newly identified metabolites in the context of existing biomarker discovery studies on liver detoxification, hypertension and insulin resistance.
It is to be noted that our method does not specifically require genotyping data. Even metabolomics measurements alone, analyzed through the GGMs, may provide sufficient information for the classification and even precise identity prediction. The unknowns with GGM evidence but without GWAS hits in Figure 4 as well as the HETE scenario represent examples for this approach.
One limitation of our approach is the requirement for associations with functionally described loci or known metabolites. Certain metabolite groups might thus systematically not be identifiable. For instance, if the identity of a whole class of biochemically related molecules is unknown (which might be due to experimental reasons), then the GGM associations between those compounds will not aid in identity elucidation. The 118 unknown compounds for which we could not derive any classification might represent such cases. Thus, our functionally oriented method should be regarded as a complementary extension to the existing identity determination methods.
Accordingly, our approach can be extended in several directions. It can be combined with method-specific, automated techniques that further exclude sets of metabolites. Previously mentioned methods relying on mass-spectra [15] or chromatographic properties [17] are suitable candidates here. Moreover, the method can be directly transferred to other types of metabolomics datasets not specifically originating from MS experiments, such as NMR-based metabolomics.
Beyond the application to metabolite identification, our study demonstrates the general potential of functional metabolomics in the context of genome-wide association studies. The comprehensive metabolic picture provided by GGMs in combination with GWAS allows for the detailed analysis of metabolic functions, chemical classes, enzyme-metabolite relationships and metabolic pathways.
We used data from n = 1768 fasting serum samples used in a previously published genome-wide association study on a German population cohort. Details of the sample acquisition and experimental procedures can be found in [5]. Briefly, metabolic profiling was done using ultrahigh-performance liquid-phase chromatography and gas-chromatography separation, coupled with tandem mass spectrometry. The dataset contains a total of 292 known compounds and, in addition to the GWAS study in [5], 225 unknown compounds. Metabolite concentrations were log-transformed since a test of normality showed that in most cases the log-transformed concentrations were closer to a normal distribution than the untransformed values [5]. Genotyping was carried out using the Affymetrix GeneChip array 6.0. For our analyses, we only considered autosomal SNPs passing the following criteria: call rate >95%, Hardy-Weinberg-Equilibrium p-value p(HWE)>10−6, minor allele frequency MAF>1%. In total, 655,658 SNPs were left after filtering.
In order to avoid spurious false positive associations due to small sample sizes, only metabolic traits with at least 300 non-missing values were included and data-points of metabolic traits that lay more than 3 standard deviations off the mean were excluded by setting them to ‘missing’ in the analysis (leaving 273 known and 213 unknown metabolites). Genotypes are represented by 0, 1, and 2 for major allele homozygous, heterozygous, and minor allele homozygous, individuals respectively. We employed a linear model to test for associations between a SNP and a metabolite assuming an additive mode of inheritance. Statistical tests were carried out using the PLINK software (version 1.06) [41] with age and gender as covariates. Based on a conservative Bonferroni correction, associations with p-values<1.6×10−10 meet genome-wide significance, corresponding to a significance level of α = 0.05. SNP-to-gene assignments were derived via linkage disequilibrium (LD) from HAPMAP [42]. A SNP was associated with a gene whenever there was at least one other SNP lying in the transcribed region of this gene (that is from 5′UTR to 3′UTR) that displays an r2≥0.8 with the query SNP. A detailed description of the GWAS procedure can be found in [5].
Lookups of previously known associations between phenotypes and genetic variants were performed using the GWAS catalog [26]. We list a phenotype with one of our GWAS hits, if the phenotype was reported with at least one SNP that displays an LD r2≥0.5 with the respective “lead SNP”. Lookups of eQTLs were performed for all significant SNPs (474, see Dataset S1) using the GTEx database [29]. We applied a p-value cutoff of 2.7×10−9, corresponding to a significance level of 0.05 and correction for 474×40,000 tests (the number of SNPs times number of transcripts, conservative estimate). Detailed results up to a p-value of 10−5 can be found in Dataset S1.
For the GGM calculation, we require a full data matrix without missing values. From the original data matrix containing n = 1768 samples and 517 metabolites (thereof 292 knowns and 225 unknowns), we first excluded metabolites with more than 20% missing values (column direction), and then samples with more than 10% missing values (row direction). The filtered data matrix still contained n = 1764 samples with 355 metabolites (217 knowns and 138 unknowns). Remaining missing values were imputed with the ‘mice’ R package [43]. Note that the numbers of metabolites used in the GWAS and in the GGM analysis differ due to specific constraints for the treatment of missing values in the two methods.
Gaussian graphical models are induced by full-order partial correlation coefficients, i.e. pairwise correlations corrected against all remaining (n-2) variables. GGMs are based on linear regressions with multiple predictor variables. When regressing two random variables X and Y on the remaining variables in the data set, the partial correlation coefficient between X and Y is given by the Pearson correlation of the residuals from both regressions. Since our dataset contains more samples than variables, full-order partial correlations can be conveniently calculated by a matrix inversion operation. A significance cutoff of α = 0.05 with Bonferroni correction was applied. A detailed description of the GGM calculation procedure can be found in [21].
Age, gender and SNP effects were removed by adding the respective variables and SNPs states to the data matrix. For each pair of variables under investigation, Gaussian graphical models remove the effects of all remaining variables on this correlation (due to the above-mentioned linear regression approach). That is, adding a variable to the data matrix will automatically result in the removal of confounding effects of this variable on the correlations of all other variables. Note that age, gender and SNPs were not investigated as an actual node in the network but merely used for the correction procedure. For the later analysis steps, we then only considered metabolite-metabolite edges in the network. SNP states were coded as numerical values of 0, 1 and 2 (see previous section), such that the linear regressions that underlie the GGM correspond to an additive genetic model (cf. [5]). Gender represents a “dummy variable” [44] in the linear regression model which only takes values of 1 (male) and 0 (female).
Metabolic reactions were imported from three independent human metabolic reconstruction projects: (1) H. sapiens Recon 1 from the BiGG databases [45], (2) the Edinburgh Human Metabolic Network (EHMN) reconstruction [46] and (3) the KEGG PATHWAY database [47] as of January 2012. We attempted to create a highly accurate mapping between the different metabolite identifiers of the respective databases, in order to ensure the identity of each compound in our list. Entries referring to whole groups of metabolites, such as “phospholipid”, “fatty acid residue” or “proton acceptor” were excluded from our study. Furthermore, we did not consider metabolic cofactors such as “ATP”, “CO2”, and “SO4” etc. in our analysis, since such metabolites unspecifically participate in a plethora of metabolic reactions. For each enzyme catalyzing one or more reactions in our pathway model, we retrieved functional annotations from two independent sources: (i) GO-Terms from the Gene Ontology [48] and (ii) enzyme pathway annotations from the KEGG PATHWAY database [47].
All imported metabolic pathways along with metabolite database identifiers, excluded compounds and pathway annotations can be found in Dataset S1.
|
10.1371/journal.pntd.0002282 | Mapping the Genes for Susceptibility and Response to Leishmania tropica in Mouse | L. tropica can cause both cutaneous and visceral leishmaniasis in humans. Although the L. tropica-induced cutaneous disease has been long known, its potential to visceralize in humans was recognized only recently. As nothing is known about the genetics of host responses to this infection and their clinical impact, we developed an informative animal model. We described previously that the recombinant congenic strain CcS-16 carrying 12.5% genes from the resistant parental strain STS/A and 87.5% genes from the susceptible strain BALB/c is more susceptible to L. tropica than BALB/c. We used these strains to map and functionally characterize the gene-loci regulating the immune responses and pathology.
We analyzed genetics of response to L. tropica in infected F2 hybrids between BALB/c×CcS-16. CcS-16 strain carries STS-derived segments on nine chromosomes. We genotyped these segments in the F2 hybrid mice and tested their linkage with pathological changes and systemic immune responses.
We mapped 8 Ltr (Leishmania tropica response) loci. Four loci (Ltr2, Ltr3, Ltr6 and Ltr8) exhibit independent responses to L. tropica, while Ltr1, Ltr4, Ltr5 and Ltr7 were detected only in gene-gene interactions with other Ltr loci. Ltr3 exhibits the recently discovered phenomenon of transgenerational parental effect on parasite numbers in spleen. The most precise mapping (4.07 Mb) was achieved for Ltr1 (chr.2), which controls parasite numbers in lymph nodes. Five Ltr loci co-localize with loci controlling susceptibility to L. major, three are likely L. tropica specific. Individual Ltr loci affect different subsets of responses, exhibit organ specific effects and a separate control of parasite load and organ pathology.
We present the first identification of genetic loci controlling susceptibility to L. tropica. The different combinations of alleles controlling various symptoms of the disease likely co-determine different manifestations of disease induced by the same pathogen in individual mice.
| Leishmaniasis, a disease caused by Leishmania ssp. is among the most neglected infectious diseases. In humans, L. tropica causes cutaneous form of leishmaniasis, but can damage internal organs too. The reasons for this variability are not known, and its genetic basis was never investigated. Therefore, analysis of genes affecting host's responses to this infection can elucidate the characteristics of individual host-parasite interactions. Recombinant congenic strain CcS-16 carries 12.5% genes from the mouse strain STS/A on genetic background of the strain BALB/c, and it is more susceptible than BALB/c. In F2 hybrids between BALB/c and CcS-16 we detected and mapped eight gene-loci, Ltr1-8 (Leishmania tropica response 1-8) that control various manifestations of disease: skin lesions, splenomegaly, hepatomegaly, parasite numbers in spleen, liver, and inguinal lymph nodes, and serum level of CCL3, CCL5, and CCL7 after L. tropica infection. These loci are functionally heterogeneous - each influences a different set of responses to the pathogen. Five loci co-localize with the previously described loci that control susceptibility to L. major, three are species-specific. Ltr2 co-localizes not only with Lmr14 (Leishmania major response 14), but also with Ir2 influencing susceptibility to L. donovani and might therefore carry a common gene controlling susceptibility to leishmaniasis.
| Leishmaniasis is endemic in 98 countries on 5 continents, causing 20,000 to 40,000 deaths per year [1]. In the past decade the number of endemic regions have expanded, prevalence has increased and the number of unrecorded cases must have been substantial, because notification has been compulsory in only 32 of the 98 countries where 350 million people are at risk [1], [2]. Infection represents an important global health problem, as no safe and effective vaccine currently exists against any form of human leishmaniasis, and the treatment is hampered by serious side effects [3].
The disease is caused by obligate intracellular vector-borne parasites of the genus Leishmania. In the vertebrate host organism, Leishmania parasites infect so-called professional phagocytes (neutrophils, monocytes and macrophages) [4], as well as dendritic cells [5], immature myeloid precursor cells, sialoadhesin-positive stromal macrophages of the bone marrow, hepatocytes and fibroblasts [6]. Leishmaniasis includes asymptomatic infection and three main clinical syndromes. In the dermis, parasites cause the cutaneous form of the disease, which can be localized or diffuse; in the mucosa, they cause mucocutaneous leishmaniasis, and the metastatic spread of infection to the spleen and liver leads to visceral leishmaniasis (also known as kala-azar or black fever). Parasites can also enter other organs, such as lymph nodes, bone marrow and lungs, and in rare cases, can even reach the brain [4]. One of the major factors determining the type of pathology is the species of Leishmania [7]. However, the transmitting vector, as well as genotype, nutritional status of the host, and environmental and social factors also have a large impact on the outcome of the disease [4], [7]. That is why even patients infected by the same species of Leishmania develop different symptoms [7] and may differ in response to therapy [3]. The basis of this heterogeneity is not well understood [8], but part of this variation is likely genetic [4].
The search for loci and genes controlling leishmaniasis included candidate-gene approach, genome-wide linkage and association mapping. Genotyping of candidate genes, which have been chosen on the basis of previous immunological studies (hypothesis-driven approach) detected influence of polymorphism in HLA-Cw7, HLA-DQw3, HLA-DR, TNFA (tumor necrosis factor alpha), TNFB, IL4, IFNGR1 (interferon gamma receptor 1) [reviewed in [4]], TGFB1 (transforming growth factor, beta 1) [9], IL1 [10], IL6 [11], CCL2/MCP1 (chemokine (C-C motif) ligand 2) [12], CXCR1 (chemokine (C-X-C motif) receptor 1) [13], CXCR2 (chemokine (C-X-C motif) receptor 2) [14], FCN2 (ficolin-2) [15] and MBL2 (mannose-binding lectin (protein C) 2) [16] on response to different human leishmaniases.
Hypothesis-independent search for susceptibility genes included genome-wide linkage and association mapping. Bucheton and coworkers [17] performed a genome-wide linkage scan, identified a major susceptibility locus that controls the susceptibility to L. donovani on chromosome 22q12 [17] and found that polymorphism in IL2RB (interleukin 2 receptor, beta chain) in this chromosomal region is associated with susceptibility to visceral leishmaniasis [18]. Genome-wide search with the subsequent analysis of a putative susceptibility locus on chromosome 6q27 revealed that polymorphism in DLL1 (delta-like 1 (Drosophila)), the ligand for NOTCH3 (Neurogenic locus notch homolog protein 3) [19] is associated with susceptibility to visceral leishmaniasis caused by L. donovani and L. infantum chagasi. Delta1-Notch3 interactions bias the functional differentiation of activated CD4+ T cells [20]. GWAS (genome-wide association study) established that common variants in the HLA-DRB1-HLA-DQA1 HLA class II region contribute to susceptibility to L. donovani and L. infantum chagasi [21].
Genome-wide linkage in mouse revealed susceptibility genes Nramp1 (Natural resistance-associated macrophage protein 1)/Slc11a1 (solute carrier family 11 (proton-coupled divalent metal ion transporters), member 1) [22] and Fli1 (Friend leukaemia virus integration 1) [23] and the role of these genes has been also established in humans [13], [24], [25]. NRAMP1, which controls susceptibility to L. donovani and L. infantum functions as a divalent metal pH-dependent efflux pump at the phagosomal membrane of macrophages and neutrophils [26]. It is also expressed in dendritic cells and influences major histocompatibility complex class II expression and antigen-presenting cell function [27]. Susceptible mouse allele carries a “null” mutation that abolishes gene function (it is a natural knockout) [28], whereas polymorphisms in the promoter, exon3 and the intron of human SLC11A1 [24], are expected to have a smaller impact on gene function. The Friend leukaemia virus integration gene, linked with wound healing, influences cutaneous leishmaniasis caused by L. major in mouse [23] and by L. braziliensis in human [25]. It remains to be tested, whether natural polymorphisms detected in mouse genes bg (beige)/Lyst (lysosomal trafficking regulator) [29] and cationic amino acid transporter Slc7a2 (solute carrier family 7 (cationic amino acid transporter, y+ system), member 2) [30] influencing response to L. donovani [31] and L. major [30], respectively, plays role also in humans. However, nothing is known about genes controlling L. tropica-induced disease in humans.
L. tropica causes cutaneous leishmaniasis in humans, but it can also visceralize. Although cutaneous disease due to L. tropica is known for a long time, its potential to visceralize in humans has been recognized only relatively recently [32]. Visceralized L. tropica was also identified as the cause of an initially not understood systemic illness in veterans returning from endemic areas in the Middle East [33]. This finding stimulated interest in less typical symptoms induced by this parasite. It was found that L. tropica caused visceral disease in Kenya [34], as well as classical visceral leishmaniasis (kala-azar) in India [35], [36] and in Iran [37], and disseminated cutaneous leishmaniasis accompanied with visceral leishmaniasis in Iran [38]. L. tropica was also implicated in development of mucosal leishmaniasis in Iran [39]. The reasons of this variability are not known.
A suitable animal model for study of this parasite would therefore contribute to genetic dissection of the functional and clinical manifestations of infection. Golden hamsters (Mesocricetus auratus) have been considered to be the best model host for L. tropica infection, but this host is not inbred and therefore not suitable for genetic dissection. Fortunately, several L. tropica strains from Afghanistan, India [40], and Turkey [41] have been reported to cause cutaneous disease in inbred BALB/c mice. Extension of analysis to the strains C57BL/6J, C57BL/10SgSnAi and gene-deficient mice on their backgrounds indicated role of IL-10 and TGFβ in regulation of parasite numbers in ears of infected mice [42].
We studied susceptibility to L. tropica using BALB/c-c-STS/A (CcS/Dem) recombinant congenic (RC) strains [43], which differ greatly in susceptibility to L. major [44], [45]. Parental strains BALB/c, STS and RC strains CcS-3, CcS-5, CcS-11, CcS-12, CcS-16, CcS-18, and CcS-20 were infected with L. tropica and skin lesions, cytokine and chemokine levels in serum, splenomegaly, hepatomegaly, and parasite numbers in organs were measured [46]. These experiments revealed that manifestations of the disease after infection with L. tropica are strongly influenced by genotype of the host. We have found that females of the RC strain CcS-16 that contains 12.5% genes of the resistant donor strain STS and 87.5% genes of the susceptible strain BALB/c [43], [47] developed the largest skin lesions and exhibited a unique systemic chemokine reaction, characterized by additional transient early peaks of CCL3 and CCL5, which were present neither in CcS-16 males nor in any other tested RC strain [46]. In order to establish the genetic basis of these differences, we prepared F2 hybrids between BALB/c and CcS-16, infected them with L. tropica and measured their skin lesions, splenomegaly, hepatomegaly, parasite numbers in spleen, liver and inguinal lymph nodes, and serum level of CCL3, CCL5 and CCL7 during the transient early peak. The strain CcS-16 carries STS-derived segments on nine chromosomes. They were genotyped in the F2 hybrid mice and their linkage with pathological symptoms and systemic immune responses was determined, which revealed eight controlling genes.
Females of strains BALB/c (16 infected, 16 uninfected) and CcS-16 (15 infected, 11 uninfected) were 8 to 19 weeks old (mean age 12 weeks, median age 12 weeks) at the time of infection. When used for these experiments, strain CcS-16 was in more than 90 generations of inbreeding. The parts of its genome inherited from the BALB/c or STS parents were defined [48]. 247 female F2 hybrids between CcS-16 and BALB/c (age 9 to 16 weeks at the time of infection, mean age 13 weeks, median 13 weeks) were produced at the Institute of Molecular Genetics AS CR, v.v.i.. Mice were kept in individually ventilated cages (Ehret, Emmendingen, Germany) and tested in two experimental groups. Both groups of F2 hybrids were derived from the same F1 parents; second experiment started seven weeks after the first. 2 mice died shortly after inoculation and were excluded from experiments. Among analyzed F2 hybrids, first experiment consisted of 111 mice, of which 51 mice originated from a cross (BALB/c×CcS-16)F2 (mean age 11.9 weeks, median 12 weeks; 3 mice died before the end of an experiment), 60 mice originated from a cross (CcS-16×BALB/c)F2 (mean age 12.6 weeks, median age 13 weeks; 1 mouse died before the end of an experiment). According to the nomenclature rules, the first strain listed in the cross symbol is the female parent, the second the male. The second experiment contained 134 mice, of which 64 mice originated from a cross (BALB/c×CcS-16)F2 (mean age 12.6 weeks, median 16 weeks; 2 mice died before the end of an experiment), 70 mice originated from a cross (CcS-16×BALB/c)F2 (mean age 13.4 weeks, median age 13 weeks; 6 mice died before the end of an experiment). The numbers of mice analyzed for individual phenotypes are given in Supplementary Table S1.
All experimental protocols utilized in this study comply with the Czech Government Requirements under the Policy of Animal Protection Law (No. 246/1992) and with the regulations of the Ministry of Agriculture of the Czech Republic (No. 207/2004), which are in agreement with all relevant European Union guidelines for work with animals and were approved by the Institutional Animal Care Committee of the Institute of Molecular Genetics AS CR and by Departmental Expert Committee for the Approval of Projects of Experiments on Animals of the Academy of Sciences of the Czech Republic (permission Nr. 37/2007).
Leishmania tropica from Urfa, Turkey (MHOM/1999/TR/SU23) was used for infecting mice. Amastigotes were transformed to promastigotes using SNB-9 [49], and 1×107 stationary phase promastigotes from subculture 2 were inoculated in 50 µl of sterile Phosphate Buffer Saline (PBS) s.c. into the tail base, with promastigote secretory gel (PSG) collected from the midgut of L. tropica-infected Phlebotomus sergenti females (laboratory colony originating from L. tropica focus in Urfa). PSG was collected as described [50]. The amount corresponding to one sand fly female was used per mouse.
The size of the skin lesions was measured every second week using the Profi LCD Electronic Digital Caliper Messschieber Schieblehre Messer (Shenzhen Xtension Technology Co., Ltd. Guangdong, China), which has accuracy 0.02 mm. Blood was collected every 2 weeks in volume from 60 to 180 µl, and serum was frozen at −30°C for further analysis. The mice were killed 43 weeks after inoculation. Blood, spleen, liver and inguinal lymph nodes were collected for later analysis.
Parasite load was measured in frozen lymph nodes, spleen, and liver samples using PCR-ELISA according to the previously published protocol [51]. Briefly, total DNA was isolated using a TRI reagent (Molecular Research Center, Cincinnati, USA) standard procedure (http://www.mrcgene.com/tri.htm). For PCR, two primers (digoxigenin-labeled F 5′-ATT TTA CAC CAA CCC CCA GTT-3′ and biotin-labeled R 5′-GTG GGG GAG GGG CGT TCT-3′ (VBC Genomics Biosciences Research, Austria) were used for amplification of the 120-bp conservative region of the kinetoplast minicircle of Leishmania parasite, and 50 ng of extracted DNA was used per each PCR reaction. For a positive control, 20 ng of L. tropica DNA per reaction was amplified as a highest concentration of standard. A 30-cycle PCR reaction was used for quantification of parasites in lymph nodes; 33 cycles for spleen, and 40 cycles for liver. Parasite load was determined by analysis of the PCR product by the modified ELISA protocol (Pharmingen, San Diego, USA). Concentration of Leishmania DNA was determined using the ELISA Reader Tecan and the curve fitter program KIM-E (Schoeller Pharma, Prague, Czech Republic) with least squares-based linear regression analysis.
Levels of GM-CSF (granulocyte-macrophage colony-stimulating factor), CCL2 (chemokine ligand 2)/MCP-1 (monocyte chemotactic protein-1), CCL3/MIP-1α (macrophage inflammatory protein-1α), CCL4/MIP-1β (macrophage inflammatory protein-1β), CCL5/RANTES (regulated upon activation, normal T-cell expressed, and secreted) and CCL7/MCP-3 (monocyte chemotactic protein-3) in serum were determined using Mouse chemokine 6-plex kit (eBioscience, Vienna, Austria). The kit contains two sets of beads of different size internally dyed with different intensities of fluorescent dye. The set of small beads was used for GM-CSF, CCL5/RANTES and CCL4/MIP-1β and the set of large beads for CCL3/MIP-1α, CCL2/MCP-1 and CCL7/MCP-3. The beads are coated with antibodies specifically reacting with each of the analytes (chemokines) to be detected in the multiplex system. A biotin secondary antibody mixture binds to the analytes captured by the first antibody. Streptavidin-phycoerythrin binds to the biotin conjugate and emits a fluorescent signal. The test procedure was performed in the 96 well filter plates (Millipore, USA) according to the protocol of manufacturer. Beads were analyzed on flow cytometer LSR II (BD Biosciences, San Jose, USA). Lyophilized GM-CSF and chemokines (CCL2/MCP-1, CCL3/MIP1α, CCL4/MIP1β, CCL5/RANTES, CCL7/MCP-3) supplied in the kit were used as standards. Concentration was evaluated by Flow Cytomix Pro 2.4 software (eBioscience, Vienna, Austria). The limit of detection of each analyte was determined to be for GM-CSF 12.2 pg/ml, CCL2/MCP-1 42 pg/ml, CCL7/MCP-3 1.4 pg/ml, CCL3/MIP-1α 1.8 pg/ml, CCL4/MIP-1β 14.9 pg/ml, and for CCL5/RANTES 6.1 pg/ml.
DNA was isolated from tails using a proteinase procedure [52] with modifications described in [51]. The strain CcS-16 differs from BALB/c at STS-derived regions on nine chromosomes [48 and unpublished results]. These differential regions were typed in the F2 hybrid mice between CcS-16 and BALB/c using 23 microsatellite markers (Generi Biotech, Hradec Králové, Czech Republic): D2Mit156, D2Mit389, D2Nds3, D2Mit257, D2Mit283, D2Mit52, D3Mit25, D3Mit11, D4Mit153, D6Mit48, D6Mit320, D10Mit67, D10Mit103, D11Mit139, D11Mit242, D11Nds18, D11Mit37, D16Mit126, D17Mit38, D17Mit130, D18Mit35, D18Mit40 and D18Mit49 (Supplementary Table S2). The maximum distance between any two markers in the chromosomal segments derived from the strain STS or from the nearest BALB/c derived markers was 14.16 cM. The DNA genotyping by PCR was performed as described elsewhere [53]. The genotyping for microsatellite markers with fragment length difference of less than 10 bp was performed by using ORIGINS Elchrom Scientific electrophoresis (Elchrom Scientific AG, Cham, Switzerland) according to manufacturer's instruction. Briefly, DNA was amplified as described in [53]. Each PCR product was mixed with 5 µl of loading buffer and electrophoresed using Spreadex EL300 gel and Spreadex EL400 gel (Elchrom Scientific AG, Cham, Switzerland) for product with size of less than 150 bp or more than 150 bp, respectively. The best gel and proper running time was selected using ElQuantTM Software (Elchrom Scientific AG, Cham, Switzerland). 30 mM TAE buffer was used as a running buffer. Running temperature was set to 20°C and to 50°C, when voltage was set to 120 V and 100 V, respectively. After finishing the electrophoresis gel was stained by ethidium bromide and the results were read by GENE bio-imaging system (Syngene, Cambridge, UK).
The role of genetic factors in control of the tested pathological and immunological symptoms was examined with ANOVA using the program Statistica for Windows 8.0 (StatSoft, Inc., Tulsa, Oklahoma, USA). Marker, grandparent-of-origin effect and age were fixed factors and the experiment was considered as a random factor. In order to obtain normal distribution of the analyzed parameters, the obtained values were transformed, each as required by its distribution, as shown in the legends of the Tables. Markers and interactions with P<0.05 were combined in a single comparison.
To obtain whole genome significance values (corrected P-values) the observed P-values (αT) were adjusted according to Lander and Schork [54] using the formula:where G = 1.75 Morgan (the length of the segregating part of the genome: 12.5% of 14 M); C = 9 (number of chromosomes segregating in cross between CcS-16 and BALB/c, respectively); ρ = 1.5 for F2 hybrids; h(T) = the observed statistic (F ratio).
The percent of total phenotypic variance accounted for by a QTL and its interaction terms was computed by subtracting the sums of squares of the model without this variable from the sum of squares of the full model and this difference divided by the total regression sums of squares:
Differences in skin lesions development between strains BALB/c and CcS-16 are controlled by two loci, which are not dependent on or influenced by interaction with other genes (main effects) (Table 1, Figure 1). Ltr2 (Leishmania tropica response 2) linked to D2Nds3 (Figure 1A) and D2Mit389 influences lesion size at week 19 (corrected P = 0.004, Bonferroni corr. P = 0.049), 21 (corrected P = 0.0020, Bonferroni corr. P = 0.024) and 31 (corrected P = 0.0152, Bonferroni corr. P = 0.18) after infection, Ltr3 that controls lesion size at week 21 after infection is linked to D3Mit11 (corrected P = 0.042, Bonferroni corr. P = 0.5) (Figure 1B). STS allele of both Ltr2 and Ltr3 determines larger lesions. STS allele of Ltr4 marked by D4Mit153 (which also controls parasite numbers in liver and in lymph nodes) has an opposite effect on the studied trait; its STS allele is associated with smaller lesions at week 27 after infection. Figure 1C and Figure 1D show the strong additive effects of Ltr2 and Ltr3, and Ltr2, Ltr3 and Ltr4, respectively. However, Ltr3 and Ltr4 effects on skin lesions (nominal P value = 0.00048 and 0.00096, respectively, corr. P value = 0.024 and 0.045, respectively) were not significant after Bonferroni correction for number of tested weeks of infection and for whole genome significance. Although lesions were larger in the second experiment, no significant interaction between experimental group and markers was observed.
Genetic analysis of F2 hybrids has revealed identical genetic control of serum levels of CCL3 and CCL5 at week 7 after infection (Table 5, 6). Ltr3 linked to D3Mit11 determines levels of both CCL3 (corrected P = 0.0046) and CCL5 (corrected P = 0.010), its BALB/c allele is associated with higher chemokine levels (Table 5). Ltr3 has not only individual (main) effect on chemokines levels, but also influences levels of CCL3 (corrected P = 0.014) and CCL5 (corrected P = 0.0012) in interaction with Ltr7 linked to D17Mit130. The largest effect is seen by Ltr3 when Ltr7 is SS. In that genotypic situation the Ltr3 CC alleles cause more than 300×higher levels of CCL3 and 28×higher levels of CCL5 than the Ltr3 SS alleles (Table 6). It is likely that this very large size of this effect in Ltr7 SS mice makes the Ltr3 effects visible as a main effect, although smaller, in F2 hybrids irrespective of their Ltr7 genotype.
CCL7 level is controlled with two loci with an opposite effect on the studied trait. The homozygosity for the STS allele of Ltr2 (SS) determines higher CCL7 level (corrected P = 0.002), whereas homozygosity for the BALB/c allele of Ltr8 (CC) is associated with higher level of this chemokine (corrected P = 0.013) (Table 5). No significant interaction between experimental group and marker was observed. Older mice had higher levels of CCL7 in serum than the younger ones, but we did not observe any interactions between marker and age (nominal P (Ltr2) = 0.80, nominal P (Ltr8) = 0.64). Levels of CCL7 in serum of infected mice are also influenced by interaction of Ltr2 linked to D2Mit257 and Ltr6 linked to D11Mit37 (corrected P = 0.016), the highest CCL7 levels are observed in STS allele (SS) homozygotes in Ltr6 in combination with heterozygotes (CS) or STS allele (SS) homozygotes in Ltr2 (Table 6).
Although chemokine levels were higher in the first experiment, no significant interaction between experimental group and markers was observed.
No linkage was found for GM-CSF, CCL2/MCP-1 and CCL4/MIP-1β.
The present study provides the first insight into the genetic architecture of susceptibility to L. tropica. We have described eight loci on seven chromosomes (Figure 2 [10], [12], [55]–[83]) and shown that the presence of individual symptoms of disease is controlled by different subsets of host's genes. The identification of host's genes responsible for the specific symptoms of the disease induced by different Leishmania species will contribute to the understanding of mechanisms of pathogenesis of leishmaniasis, similarly as comparative parasite genomics led to identification of differentially distributed genes in Leishmania species inducing different pathology [84], [85], and analysis of specific virulence factors revealed how different Leishmania species subvert or circumvent host's defenses [7]. Such analysis will provide description of individual predisposition to specific symptoms of disease and its probable course. Moreover, the possibility to compare genetics of response to several Leishmania species will further help to understand the genetic basis of general and species-specific responses of the host. This will synergize with the future information about genome sequence of L. tropica and about interaction of its specific virulence factors with the immune system.
Our data show that interaction of mice with L. tropica parasites is complex and involves numerous genes and responses (Table 7). We have detected eight loci that in the strain CcS-16 control host-parasite interaction (Table 7, Figure 2). All eight Ltr loci are involved in gene-gene interactions (Figure 3), four loci (Ltr2, Ltr3, Ltr6, Ltr8) have also individual effect, while effects of Ltr1, Ltr4, Ltr5 and Ltr7 are seen only in interaction with other Ltr loci. This is not surprising, as the average proportion of genetic variation explained by epistatic QTLs in mice in different systems was estimated to be 49% [86] and gene-gene interactions were observed also in response to other pathogens such as L. major [87]–[89], Trypanosoma brucei brucei [53], Salmonella enteritidis [90], Plasmodium falciparum [91] and Mycobacterium leprae [92].
The loci described here have heterogeneous effects (Table 7). Ltr1 on chromosome 2 controls in interaction with Ltr4 only parasite numbers in lymph nodes, whereas the more distal Ltr2 on the same chromosome influences development of skin lesions, splenomegaly (in interaction with Ltr3), hepatomegaly, parasite load in liver and level of CCL7 in serum. Multiple functions are also exerted by Ltr3 on chromosome 3, which controls splenomegaly (in interaction with Ltr2), parasite numbers in spleen, and levels of CCL3 and CCL5 in serum. We have analyzed genetic control of early levels of chemokines, as there is a unique early peak in the CcS-16 females [46]. However, comparison of genetic control of CCL3 and CCL5 levels with genetic control of development of skin lesions indicates that there is no simple correlation between the chemokines levels and manifestations of disease. Ltr4 on chromosome 4 controls in interaction with Ltr1 and Ltr8 parasite numbers in lymph nodes and in liver, respectively. Ltr5 on chromosome 10 influences in interaction with Ltr7 or Ltr8 splenomegaly. Ltr6 influences parasite numbers in spleens and level of CCL7 in serum (in interaction with Ltr2). Ltr7 controls splenomegaly (in interaction with Ltr5) and in interaction with Ltr3 level of both CCL3 and CCL5 in serum. Ltr8 controls splenomegaly (as a main effect gene and in interaction with Ltr5), parasite numbers in liver (in interaction with Ltr4) and level of CCL7 in serum. Ltr1 and Ltr5 control only one parameter, whereas other loci have multiple effects. Some multiple effects could reflect causal relationship – e.g. CCL7 influences recruitment of monocytes to spleen [93], which could contribute to splenomegaly. The observed multiple effects of some Ltr loci might also suggest that some such loci might represent complexes of two or more closely linked Ltr genes. This issue will be resolved by future recombinational analysis.
We have detected also loci that control symptoms, such as splenomegaly, in which the strains BALB/c and CcS-16 do not differ [46]. This is because in an inbred strain the final outcome of response is exerted by multiple genes, which often have opposite effects, masking each other. In the F2 hybrids these genes segregate and can be therefore detected.
Reliability and validity of the described loci is supported by the fact that they have been detected by analysis of different phenotypes and their statistical significance was corrected for whole genome testing and where appropriate also by conservative Bonferroni correction. The relatively high proportion of variance explained by the mapped loci (Table 1–6) might be partly due to a limited variability of the tested manifestations of the disease.
Most inbred mouse strains that were produced without intentionally selectively bred for a specific quantitative phenotype (like susceptibility to specific infections) inherited from their non-inbred ancestors randomly susceptible alleles at some loci and resistant alleles at others, so that their overall susceptibility phenotype depends on the relative number of both. STS is resistant to L. tropica and does not develop skin lesions [24], however some STS-derived segments carried by CcS-16 on chromosome 2 (Ltr2) and possibly also on chromosome 3 (Ltr3) are associated with larger lesions. Similarly, STS-derived alleles of Ltr2 and Ltr6 are associated with higher parasite load in liver and spleen, respectively. This finding is not unique as susceptibility alleles originating from resistant strains were found in studies of colon cancer [94] and L. major [95] susceptibility; a low-responder allele was identified in a strain exhibiting high response to IL-2 [96] or producing a high level of IFNγ [97], whereas a high responder allele was found in a strain producing low level of IL-4 [98].
Loci Ltr3 and Ltr6 influencing parasite numbers in spleen (Table 2) were significant only in the cross (BALB/c×CcS-16)F2, but not in the cross (CcS-16×BALB/c)F2, hence the outcome in these crosses that are theoretically genetically identical depends on the strain of the female or male used originally to produce the F1 hybrids, which were then crossed with each other to produce the F2 hybrids for the tests. Thus, this is a special type of a transgenerational parental effect as the mothers and fathers of the F2 hybrids were genetically identical. Recently, examples of transgenerational parental effects have been described in several species [reviewed in [99]] and several possible mechanisms have been proposed. Our observation may reflect a parental effect due to modification of the developing immune system of fetuses or youngs by maternal environment, maternal nutritional effects, or epigenetic effects, and it offers a possibility to characterize the transgenerationally regulated functional pathways.
Control of parasite elimination differs among organs: the loci Ltr1 and Ltr4 interact to control parasite numbers in inguinal lymph nodes, while Ltr4 in interaction with Ltr8 influences parasite load in liver (Table 4). Parasite load in liver is also controlled by Ltr2 (Table 2), whereas parasite burden in spleen is influenced by Ltr3 and Ltr6 (Table 2). These data show that parasite elimination in lymph nodes, liver and spleen are controlled differently, suggesting a predominantly organ specific control of parasite load. Mechanistic studies analyzing response to L. tropica in different organs are not yet available, but generally organ specific responses described here are compatible with the mechanistic studies of other parasites. The enzymes inducible nitric oxide synthase and phagocyte NADPH oxidase, which are required for the control of L. major, display organ- and stage-specific anti-Leishmania effects [76], [100]. Inducible nitric oxide synthase has been shown to control resistance to parasites in skin and draining lymph nodes, but not in spleen of the resistant strain C57BL/6 [100]. On the other hand, activity of phagocyte NADPH oxidase is essential for the clearance of L. major in the spleen, but it is dispensable for the resolution of the acute skin lesions and it exerted only a limited effect on the containment of the parasites in the draining lymph node [76]. Similarly, bg/Lyst (lysosomal trafficking regulator) is involved in control of parasite numbers of L. donovani in spleen, but not in liver [31]. On the other hand VCAM-1 (vascular cell adhesion molecule-1) and VLA-4 (very late antigen-4) interactions influenced early L. donovani burden in liver, but not in spleen [82].
Comparison of genetic control of parasite numbers in spleen and splenomegaly, or parasite numbers in liver and hepatomegaly shows that control of parasites elimination and organ pathology overlap only partially. For example Ltr3 controls both parasite numbers in spleen and splenomegaly, but Ltr6 is involved in control of parasite numbers in spleen, but not in splenomegaly, whereas Ltr2, Ltr8, Ltr5, and Ltr7 are involved only in control of splenomegaly (Table 2, 3, 7). Similarly, Ltr2 influences both parasite load in liver and hepatomegaly, but parasite load in liver is controlled also by interaction of Ltr4 with Ltr8. The differences in genetic control of parasite numbers and organ pathology induced by the parasites are probably due to the fact that during a chronic disease the organ damage is a combined result of speed of elimination of parasite on one hand and changes caused by reaction to parasite (such as influx of immune cells, inflammatory responses) and healing processes on the other hand. It is therefore likely that these processes are regulated by different sets of genes.
It is important to understand that as in any QTL study, failure to find a linkage between a phenotype and a marker does not rule out that such linkage may exist, although its phenotypic effect are likely smaller than in the detected linkages. So for a QTL, which affects several but not all parameters of a complex disease, this indicates that it has predominant effects on some parameters, although it might modify to a lesser extent other parameters as well.
Usually, a standard inbred-strain mapping experiment using F2 hybrids will map a QTL into a 20- to 40-cM interval [105]. In the RC strains 54% of their donor strain genome reside in segments of medium length (5–25 cM) [106]. However, RC strains can carry on some chromosomes very short segments of the donor strain origin. This feature of the RCS system allowed us previously to narrow the location of Lmr9 (Leishmania major response 9) on chromosome 4 to a segment of 1.9 cM (6.79 Mb) without any additional crosses [101]. The short length of this segment, which controls levels of serum IgE in L. major infected mice, enabled us to detect a human homolog of this locus on human chromosome 8q12 and show that it controls susceptibility to atopy [107]. In another study, we were able to precisely map Tbbr2 (Trypanosoma brucei brucei response 2) to 2.15 Mb [53].
In the present F2 mapping experiment the shortest locus Ltr1 is 4.07 Mb long (Figure 2). Although most Ltr loci contain several possible candidate genes, here we list (Figure 2)[10], [12], [55]–[83] only those that have been shown previously to influence infection with Leishmania ssp.. However, the effects of many of Ltr loci might be caused by genes that are at the present not considered as candidates. Currently we are producing mice with recombinant haplotypes that carry individual Ltr loci in a very short segment on chromosome. The testing of these strains will restrict the present number of the candidate genes to the most likely ones.
We present the first description of genetic architecture of response to L. tropica in any species. We observed organ specific control of infection and distinct control of parasite load and organ pathology, the typical characteristics of immune response to many pathogens observed in all infections where multiple disease parameters were studied (L. major [4], L. donovani [4], Borrelia burgdorferi [102], Toxoplasma gondii [108], Trypanosoma congolense [109], and Chlamydia psittaci [110]). In addition, the genetic control of response to L. tropica exhibits heterogeneity of gene effects, gene-gene interactions, and trans-generational parental effects. These complexities of genetic control have been invoked [111] to explain the very large fraction of heritability that has not been detectable in genome-wide association studies (GWAS) [112], a power deficiency that likely cannot be ameliorated by further increases of the number of tested SNPs or by whole genome sequencing. Identification of these complexities in the present study will open way to elucidation of their functional basis and detection of homologous processes in humans.
|
10.1371/journal.ppat.1004610 | Protective Efficacy of Centralized and Polyvalent Envelope Immunogens in an Attenuated Equine Lentivirus Vaccine | Lentiviral Envelope (Env) antigenic variation and related immune evasion present major hurdles to effective vaccine development. Centralized Env immunogens that minimize the genetic distance between vaccine proteins and circulating viral isolates are an area of increasing study in HIV vaccinology. To date, the efficacy of centralized immunogens has not been evaluated in the context of an animal model that could provide both immunogenicity and protective efficacy data. We previously reported on a live-attenuated (attenuated) equine infectious anemia (EIAV) virus vaccine, which provides 100% protection from disease after virulent, homologous, virus challenge. Further, protective efficacy demonstrated a significant, inverse, linear relationship between EIAV Env divergence and protection from disease when vaccinates were challenged with viral strains of increasing Env divergence from the vaccine strain Env. Here, we sought to comprehensively examine the protective efficacy of centralized immunogens in our attenuated vaccine platform. We developed, constructed, and extensively tested a consensus Env, which in a virulent proviral backbone generated a fully replication-competent pathogenic virus, and compared this consensus Env to an ancestral Env in our attenuated proviral backbone. A polyvalent attenuated vaccine was established for comparison to the centralized vaccines. Additionally, an engineered quasispecies challenge model was created for rigorous assessment of protective efficacy. Twenty-four EIAV-naïve animals were vaccinated and challenged along with six-control animals six months post-second inoculation. Pre-challenge data indicated the consensus Env was more broadly immunogenic than the Env of the other attenuated vaccines. However, challenge data demonstrated a significant increase in protective efficacy of the polyvalent vaccine. These findings reveal, for the first time, a consensus Env immunogen that generated a fully-functional, replication-competent lentivirus, which when experimentally evaluated, demonstrated broader immunogenicity that does not equate to higher protective efficacy.
| Our best effort for containment of the global HIV epidemic is the development of a broadly protective vaccine. Current research has focused on vaccines that can generate a protective immune response against numerous strains of the virus. For this reason, vaccines with centralized HIV genes as immunogens, which merge HIV genetic information and potentially protect against multiple viral strains in a single inoculation, are an increasing area of interest to the field. Existing published studies have not evaluated centralized immunogens in the context of attenuated vaccines, which to date, have demonstrated the highest level of vaccine protection in lentiviral studies. Furthermore, centralized immunogen studies have also not included protective efficacy findings accomplished through challenge with highly pathogenic virus strains. In this study we not only examine the immunogenicity of these immunogens in an animal model, but we also for the first time evaluate the ability of centralized immunogens to induce protection against virulent virus challenge.
| The scientific community has aggressively sought after the development of a universal HIV vaccine that can prevail over the extraordinary levels of antigenic diversity in the fight against HIV and AIDS. The considerable extent of genomic variation found between isolates and within clades, and to a larger extent within the circulating recombinant forms, make for an effectual blockade to vaccine protection. Different strategies of vaccine composition and delivery have been proposed that are actively and widely being examined. A majority of these vaccines target the Env protein, as lentiviral antigenic variation is most pronounced in the viral Env proteins that serve as initial primary targets for host immune responses [1]–[5].
Centralized Env immunogens are one of the more promising contemporary approaches to overcoming HIV antigenic diversity [1], [6]. Centralized sequences attempt to minimize the genetic distance between vaccine proteins and the circulating isolates that pose a threat to public health. The centralized genes are generated through the computational determination of consensus genes (the most common amino acid at each position), ancestral genes (modelling ancestral states through phylogenetics), or center of the tree sequences (phylogenetic determination of a central isolates) [1], [4], [7], [8]. Centralized genes have been investigated as effective vaccine approaches in the HIV field both as DNA and/or protein immunogens [6], [9]–[19]. To date, however, the efficacy of centralized immunogens has not been fully explored in the context of an attenuated lentiviral vaccine model that could provide both immunogenicity data as well as protective efficacy data via virulent challenge in an animal model.
Equine infectious anemia virus (EIAV), a macrophage-tropic lentivirus, causes a persistent infection and chronic disease in equids [20]. Infection, transmitted via blood-feeding horse flies, occurs in three stages: acute, chronic and inapparent. Acute and chronic stages are defined by episodes of clinical disease that are triggered by waves of viremia, and distinguished by fever, anemia, thrombocytopenia, edema, and various wasting signs. By one year post-infection animals typically progress to life-long inapparent carriers, but continue to harbor steady-state levels of viral replication in monocyte-rich tissue reservoirs [20]–[22]. Stress or immune suppression of inapparent carriers can induce increases in viral replication and potentially a recrudescence of disease [20], [23]. Among virulent lentiviruses, EIAV is unique in that, despite aggressive virus replication and associated rapid antigenic variation, greater than 90% of infected animals progress from chronic disease to an inapparent carrier stage, by a strict immunologic control over virus replication [20]. The EIAV system hence serves as a unique animal model for the natural immunologic control of lentiviral replication and disease. Further, EIAV inapparent carriers have proven to be resistant to subsequent virus exposure to diverse viral strains, indicating the development of a high level of prophylactic immunity. Thus, the EIAV system provides a valuable model for identifying critical immune correlates of protection and ascertaining the potential for developing effective prophylactic lentivirus vaccines [24].
While the disease processes for EIAV and HIV have distinguishing dynamics, key similarities between the two virus systems make EIAV an extremely valuable tool and model for AIDS vaccine development [24]–[26]. EIAV and HIV are transmitted parenterally and share a macrophage/monocyte tropism [26], [27]. EIAV quasispecies also possess high levels of antigenic heterogeneity and their Env proteins share architectural characteristics such as extensive glycosylation and immune decoys [24], [25], [28], [29]. These features, all of which are critical elements associated with initial virus exposure, coupled to a very similar immune maturation process of the EIAV-infected equine to HIV-infected humans, are fundamental factors relevant to vaccine efficacy [24], [30].
We previously reported serial studies evaluating the efficacy of an attenuated EIAV proviral vaccine containing a mutation in the viral S2 accessory gene (EIAVD9) [31]–[33]. The results of these studies indicate that horses inoculated with the EIAVD9 viral vaccine were 100% protected from disease by virulent, albeit homologous, EIAV challenge. Thus, the EIAV system mirrors other animal lentivirus vaccine models, which consistently identify attenuated vaccines as producing the highest level of vaccine protection. [31]–[38]. Our latest EIAVD9 data demonstrated the effects of challenge virus Env sequence variation on vaccine protection [39]–[41]. We identified a significant, inverse, linear correlation between vaccine efficacy and increasing divergence of the challenge virus Env surface protein, gp90, compared to the vaccine virus gp90 protein. The study demonstrated that the 100% protection of immunized horses from disease after challenge by virus with a homologous gp90 (EV0), dropped to approximately 60% protection when a challenge virus gp90 was 6% divergent (EV6), and nose-dived to less than 50% protection against challenge with a gp90 that was 13% (EV13) divergent from the vaccine strain. Most recently, we demonstrated that the attenuated vaccine strain progressively evolved during the seven-month pre-challenge period and that the observed protection from disease was significantly associated with divergence from the original vaccine strain, not the overall diversity of the vaccine Env quasispecies present on the day of challenge (DOC) [39].
Despite numerous studies on the immunogenicity of centralized Env proteins, use of these noteworthy immunogens in an attenuated vaccine model, accompanied by virulent virus challenge, has yet to be reported. In the current study, we sought to directly build upon our current model and the series of described EIAVD9 vaccine studies. Our attenuated vaccine model, coupled with the well-characterized genomic and phylogenetic ancestry of the Env gene of our EIAV strain, enabled a thorough, unparalleled evaluation of centralized sequence vaccine efficacy not as readily modelled in other lentiviral systems. The presented studies evaluated multiple derivatives of centralized Env immunogens, both consensus and ancestral, in our proviral attenuated vaccine backbone. The studies were designed to first, develop and test a consensus immunogen for functionality through examination of replication and pathogenic potential in proviral backbones; second, compare the protective efficacy of the consensus immunogen in an attenuated backbone to attenuated strains containing our ancestral Env [34] as well as a polyvalent Env attenuated strain mixture; and third, to develop and utilize a more stringent challenge model in the form of an engineered quasispecies.
Consensus gene development of the EIAV Env protein focused on the gp90 region of the gene as genomic evolution and antigenic variation in the transmembrane (gp45) protein has been shown to be minimal among characterized longitudinal EIAV isolates [42], [43]. To engineer a consensus Env, the gp90 genes of approximately 90 naturally occurring isolates from an experimental infection [42], [44] were aligned. Virus isolates included the inoculum strain Env as well as isolates from all three stages of disease (acute, chronic, and inapparent). Isolates therefore included an ancestral strain (EV0) and its descendant strains that evolved innately between day zero and 1200 days post-infection (DPI). Consensus sequences were derived primarily from codon-aligned nucleotide sequences and secondarily from amino acid alignments. The consensus sequence from the nucleotide alignment was translated, compared to the consensus sequence from the amino acid alignment for congruence and the resolution of ambiguities, and termed ConEnv. To evaluate the veracity of the derived ConEnv sequence and discriminate it against other potential consensus Env proteins, additional consensus sequences were designed. Consensus Envs representing the individual febrile episodes (six) and inapparent stage isolates were generated from the isolate amino acid sequences and thereafter a consensus from those engineered sequences was constructed as well. This “consensus from all consensus” method is similar to creating consensus Envs from each HIV clade and subsequently creating a consensus of those clades. The ConEnv consensus sequence was then examined by phylogenetic comparison with these control consensus sequences, the Env sequences involved in ConEnv construction, and sequences targeted as partner Envs for vaccine and challenge strains of this study (Fig. 1). The “consensus from all consensus” sequence was phylogenetically closely related to the ancestral Env (EV0) emerging on the main ancestral root amongst the acute disease Env genes. ConEnv shared the same ancestral root as the EV6 strain, manifesting ancestrally between EV0 and EV6. Genetic distance calculations demonstrated the ConEnv gp90 sequence was 4%, 6%, and 11% divergent from the EV0, EV6, and EV13 strains, respectively. The majority of the disparate residues occurred within the designated variable regions, specifically V3 through V7 (Fig. 2). Hence, the ConEnv sequence was a strong consensus sequence representative of the 90 isolates, capturing epitopes broadly characteristic of the family of EIAV isolates.
To fully assess the competency of the ConEnv protein to function indistinguishably from a naturally occurring Env protein, ConEnv was evaluated in the context of both attenuated and pathogenic EIAV proviruses. Commercially synthesized ConEnv was cloned into the attenuated EIAVD9 backbone [31], [32] and the pathogenic EIAVUK3 backbone [45], with the resultant proviral strains termed ConD9, and EVCon, respectively. Attenuated and pathogenic proviral clones were sequenced to verify the ConEnv gene, and then transfected into equine dermal (ED) cells for production of infectious virus stocks. Virus stocks were titered and characterized for in vitro replication kinetics [46]–[48]. Both proviral strains, EVCon (pathogenic) and ConD9 (attenuated) demonstrated typical in vitro kinetics, emulating their parental and variant strain counterparts, and peaked in virus production at approximately ten DPI.
In vivo analysis of the proviral pathogenic and attenuated ConEnv strains, by experimental infections of equids, confirmed characteristic EIAV clinical and virological profiles of both pathogenic and attenuated infections (Fig. 3). Typical attenuated and avirulent replication properties were observed for the ConD9 strain. Low level, viral replication kinetics (averaging between 102–103 copies RNA/ml) which failed to progress to clinical disease over a 100 day observation period were observed. Conversely, pathogenesis and virulence, characterized by standard viremic replication kinetics (averaging between103–104 copies RNA/ml and peaking at 106 copies RNA/ml), including the induction of acute, and progression to chronic disease were observed with the EVCon strain [20], [43], [49]. Hence, for the first time, a synthetic consensus lentiviral Env was demonstrated to not only be fully functional in the context of replication competence in vitro and in vivo, but also capable of inducing traditional and virulent lentiviral disease.
The final pre-vaccine trial evaluation was an assessment of ConEnv's immunogenicity. Assays of antibody responses elicited by EVCon, EV0, EV6, and EV13 experimental infections, with variant and consensus viruses, indicated primarily distinct neutralization phenotypes for the individual variant Envs; each variant virus was neutralized by immune serum from homologous virus infections, but not from heterologous virus infections, except for marginal neutralization of the EVCon strain by the EV0 heterologous serum (Fig. 4). However, sera produced by the EVCon virus infections were capable of not only neutralizing its homologous strain, but also neutralized the EV0 and EV6 heterologous strains. Thus, these data demonstrate that the ConEnv was similar to EV0, EV6, and EV13 in replication and virulence properties [34], yet distinct in immune properties as a result of the defined Env sequence variation. Much like the HIV consensus Env recombinant proteins that have been reported [6], [15]–[18], this consensus Env, in the context of a fully functional virus, demonstrated immunogenicity induction of neutralizing antibodies with broader recognition of epitopes than that of the naturally occurring isolates from which the ConEnv was derived.
To directly evaluate the consensus Env as well as the general premise of centralized immunogens, we compared the consensus Env attenuated strain (ConD9) with an ancestral Env attenuated vaccine strain (EIAVD9 or D9). Proficiency of the centralized immunogens was further scrutinized by inclusion of a third attenuated vaccine regimen. The third arm of the study, a polyvalent attenuated strain mixture, was chosen as the most rigorous match to the centralized immunogens. The polyvalent attenuated quasispecies was constructed utilizing the D9 backbone. A trivalent attenuated mixture was assembled with the D9 as one of three strains. The EV6 and EV13 Envs were cloned into the D9 backbone to create 6D9 and 13D9, respectively, the final two strains of the polyvalent mix. The polyvalent attenuated mixture, a 1∶1∶1 mix of D9, 6D9, and 13D9, otherwise termed TriD9, was tested in vivo for TCID50 dosage verification in a group of eight ponies.
Twenty-four ponies of mixed age and gender were divided randomly into groups of eight animals and inoculated intravenously with two, 3×103 TCID50 doses of the ancestral D9 vaccine, the consensus ConD9 or polyvalent TriD9 vaccines, at four-week intervals. The inoculated ponies were monitored daily for clinical signs of EIA, and blood samples were taken at regular intervals for standard measurements of disease, virus replication, and host immune responses, as described previously [31]–[33], [50]. Figs. 5–7 display the clinical profiles of vaccinated animals. One of the eight animals in the polyvalent TriD9 group developed clinical complications pre-challenge that compromised continued use of the animal in the study and was thus removed from the trial. All twenty-three vaccinates (3 trial groups) exhibited no clinical signs of EIA disease from the attenuated vaccine strains during the seven month observation period, a time frame that allows complete maturation of vaccine immunity prior to virus challenge [22], [30], [51] (Figs. 5A–7A).
An engineered quasispecies challenge model, EVMX, was developed as a rigorous assessment of immune protection. Based on previous studies [28], [34], equivalent (1∶1∶1) TCID50 dosages of the virulent EV0, EV6, and EV13 strains combined to create the virulent, well-defined, swarm. Six months following the second vaccination, the immunized ponies were challenged intravenously every other day with three, 3×103 TCID50, EVMX inoculations. A control group consisting of six EIAV-naïve ponies was also challenged with EVMX (Figs. 5B–7B). Analyses of vaccinate day of challenge (DOC) viral loads demonstrated all three attenuated viral regimens replicated to similar levels, averaging, over the seven month pre-challenge period, between 2×103 and 3×103 copies RNA/ml plasma (Figs. 5–7). Despite these similarities in viral vaccine replication, however, trial groups displayed markedly different levels of EVMX-induced disease. Four ancestral D9, six consensus ConD9, and three polyvalent TriD9 animals displayed clinical signs of EIA disease during the observation period post challenge (Figs. 5A–7A). Chronic disease was observed in the majority of vaccinates that experienced initial acute disease. All six control animals of each variant virus challenge group developed clinical EIA disease, indicating 100% virulence of the quasispecies challenge under the current experimental infection conditions. DOC vaccine immune responses in all groups were also indicative of mature immune responses (Table 1, Table 2, S1 Fig.). The polyvalent TriD9 group demonstrated the highest Env-specific serum antibody titer and avidity, although it was only significantly different from the ancestral D9 regimen (P = 0.017, P<0.0001, respectively), and not from the consensus ConD9 regimen (S1 Fig.). While neutralizing antibodies were detectable in all three groups, they could not be associated with protection. Similarly, DOC cellular immunity levels were similar but not correlated with protection levels (Table 2). For example, measured in vivo cytokine responses were the lowest in the polyvalent TriD9 vaccinates that showed the highest level of protection.
The percentage of animals within each trial group protected from clinical EIA was plotted as a function of days post-challenge for survival analysis (Fig. 8). Both centralized Env vaccine groups had subjects that succumbed to EVMX disease within a typical time frame of 2–3 weeks post-challenge. The polyvalent, TriD9 group, however, demonstrated a delay in the onset of disease with the first animal not breaking until 81 days post-challenge (Fig. 8). Overall, protection curves of all three vaccine groups were significantly different from one another (ANOVA, P<0.0001). Polyvalent TriD9 vaccinates demonstrated the highest levels of protection that was significantly different from the unvaccinated controls and the consensus ConD9 curves (P<0.0001, P = 0.0002). While the consensus ConD9 group had the lowest level of protection, it was significant as compared to unvaccinated subjects (P = 0.001). Analysis of the trend revealed a significant relationship between the complexity of immunogen and protective efficacy (P = 0.02). Ultimately, the consensus ConD9 strain, while pre-trial appearing to be more broadly immunogenic, demonstrated the lowest level of protection. The polyvalent TriD9 regimen demonstrated the highest level of protection against a quasispecies challenge. The current study not only highlighted important information towards HIV vaccine development and highlighted the importance of rigorous challenge strain engineering.
The collective success of attenuated vaccine regimens makes it the ideal modality to rigorously test novel vaccine concepts. Previous work by our group and others have demonstrated attenuated vaccines to be an extremely useful tool for vaccine development, regardless of the potential for marketable advancement of the attenuated platform due to the safety concerns associated with HIV vaccines. Consequently, we resolved to examine the efficacy of centralized Env immunogens in our well-established EIAV attenuated vaccine model. Results presented here reveal, for the first time, a consensus Env in a fully replication-competent attenuated virus backbone that can confer protective efficacy against virulent virus challenge; however, it does not induce the highest level of protection as compared to ancestral or polyvalent vaccines.
A concept not articulated in most reports on centralized Env sequences are the specifics involved in the determination of the consensus sequence. Creation of an ideal consensus sequence requires careful consideration of various parameters. Our original attempts to construct a consensus gp90 gene focused on a more basic approach than what we utilized for the current study. Initial alignments were performed utilizing the three sequences that comprise the EVMX quasispecies challenge strain, EV0, EV6, and EV13 [34]. Phylogenetic analysis of this consensus gp90 demonstrated that the consensus sequence was located high on the ancestral root and would not have represented a good consensus sequence as it was highly related to the ancestral, or EV0 gp90. Likewise, the “consensus of all consensus” method utilized in both the HIV and influenza fields [6], [9], [15], [18], [19], [52]–[56] resulted in a gp90 sequence that phylogenetically was more related to early disease isolates on the ancestral root (Fig. 1) than to the consensus gene generated from all 90 isolate sequences (nucleotide and amino acid). In vitro analysis of the virus constructed from the second-generation consensus gp90, assembled from the 90-isolate alignment and cloned into a proviral backbone, was found to infect equine cells, but was not fully replication competent. Therefore, if we were to analyze this Env in a single-round infection assay it would falsely appear to be fully functional. Hence, the 90-isolate sequence alignment was re-evaluated and a higher level of hand-editing performed, especially in the highly variable V3 region, prior to consensus generation. This thorough computational analysis and consideration of the consensus Env limited the potential sampling bias that obfuscates computational engineering of protein immunogens [1], [57]. Likewise, our use of an actual ancestoral Env also reduced the potential sampling bias that is problematic to computationally constructed most recent common ancestors [1], [57]. Additionally, in the absence of replication analysis and study in an attenuated model, the incapability of the consensus Env to functional naturally would not have been observed and a gp90 less representative of a fully functional Env would have been examined. Ultimately, this is the first successful construction of a consensus envelope lentivirus construct with full replication and virulence properties.
The highest protective efficacy against disease was observed in the polyvalent TriD9 vaccinates. Survival analysis revealed the polyvalent TriD9 disease curve was significantly different from the naïve controls and importantly, the consensus ConD9 curve. Considering the EVMX challenge quasispecies virus strain composition, the polyvalent TriD9 regimen displaying the highest levels of protective efficacy might be anticipated; however, the ancestral D9 vaccinates also protected to a higher degree than the consensus ConD9 vaccinates. Pre-challenge analysis of clinical and virological factors would not have predicted these results. All three attenuated vaccine regimens displayed similar levels of pre-challenge viral strain replication. Pre-trial evaluation of the capability to induce neutralizing antibodies (Fig. 4) indicated that the ConEnv produced a broader immune response than the ancestral gp90 (EV0/D9) and also the EV6 and EV13 gp90 proteins. Immune analysis of DOC responses did not reveal a direct correlation of protection. Neutralizing antibodies and cellular immune responses were not associated with protective efficacy. Although a significantly higher level of Env-specific antibodies were found in the polyvalent TriD9 vaccinates, the consensus ConD9 vaccinates, similarly, had a higher antibody response as compared to the ancestral D9. In the case of the consensus ConD9 vaccinates, immune response data is suggestive of potential issues related to the artificial nature of the consensus Env and its ability to induce broadly protective responses. Protection in all three vaccine groups could be due to anamnestic responses that could be related to conserved regions or conformational epitopes that allow for protective response in lieu of sequence identity. Although total antibody binding does not reveal directly correlative data (S1 Fig.), mapping of the reactive epitopes will be key to determining if a region of the gp90 potentially conferred more protective responses. These studies are currently underway. A notable observation of the polyvalent TriD9 vaccinates was the delayed onset of disease (Fig. 8). The quasispecies challenge virus eventually broke through causing its first case of disease at the late time point of 81 days post-challenge. The nature of this break through is an interesting study of evolution: did the EVMX Env evolve at a faster rate than the polyvalent TriD9 and escape the protective immune responses or did recombination allow the escape. Current studies are being performed to enable future characterization of break-through febrile isolates.
As part of the current report we generated an engineered quasispecies challenge virus mix containing different degrees of Env variation. The majority of current lentiviral vaccine studies employ a single strain challenge model. Heterologous challenge models, while more rigorous than homologous challenge models, are commonly single viral strains. Data presented here indicate that more comprehensive challenge models, that include variable Env proteins, should be developed for the study of lentiviral vaccines to better test the vaccine modalities being investigated. The polyvalent TriD9 and EVMX vaccine and challenge strains were at their basic phenotype, a homologous pairing. However, the complexity of a pathogenic, diverse, challenge strain resulted in a notable difference and reduction in the protective efficacy as compared to previous studies where the vaccine and challenge strains were matched in their Env sequences. Improved challenge model development for animal lentiviral studies that comprehensively test protective efficacy are critical tools required for broadly protective vaccine development.
The studies presented here demonstrate definitively that polyvalent attenuated vaccine regimens have significantly higher levels of protection as compared to centralized immunogens. Although it is not possible with absolute confidence to extrapolate the results of vaccine studies in any single animal lentivirus system to other animal lentiviruses or to HIV, the data presented here certainly highlight the priority of ascertaining centralized immunogens on HIV vaccine efficacy in the context of higher animal models that include challenge studies which can inform on the true protective nature of the proposed immunogens.
Consensus gp90 Env protein sequence was determined by alignment of nucleotide (codon alignment) and amino acid sequences from naturally arisen EIAV isolates originated from an experimental infection (pony #567) [34], [42]–[44] in the Geneious Pro package of software (Biomatters, Ltd.). Alignments were hand edited where necessary, especially in the highly variable loop region, and ambiguities resolved through partner aligning of nucleotides and amino acids. Phylogenetic characterization of the consensus Env was were constructed by the neighbor-joining method of Jukes Cantor corrected distances with the optimality criterion set to distance as measured in PAUP [58] and implemented in the Geneious Pro 5.0.4 (Biomatters Ltd., NZ) package of software. Statistical significance of branchings and clustering was assessed by bootstrap resampling of 1000 pseudoreplicates on the complete data set; trees were rooted to the original infectious ancestral Env, EV0. The trees were edited for publication using FigTree Version 1.1.2. Resultant gp90 sequence was synthesized (GeneArt, Regensberg, Germany). The consensus gp90 was cloned into pathogenic and attenuated EIAV backbones using methods and restriction sites previously described [34], [47], [48]. Viral stocks were prepared as previously described [34], [48]. Viral stock titers were determined utilizing our infectious center assay (cell-based ELISA) in fetal equine kidney cells, described previously [46].
All equine procedures were conducted in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health at the Gluck Equine Research Center of the University of Kentucky according to protocols approved by the University of Kentucky IACUC (#01058A2006). The animals were monitored daily and maintained as described previously [31], [32], [34], [43], [49]. Eight outbred, mixed-breed ponies were separated into two groups of four and experimentally inoculated intravenously with 103 TCID50 of either chimeric strain ConD9 or EVCon. Rectal temperatures and clinical status were recorded daily. Platelet numbers were determined using the IDEXX VetAutoread Hematology Analyzer (IDEXX Laboratories Inc., Westbrook ME). Clinical EIA (fever) episodes were determined on the basis of rectal temperature and platelet count (rectal temperature >39°C; platelet number <100,000/µl of whole blood) in combination with the viremic presence of infectious plasma virus (≥105) [20], [26], [43], [49], [59]. Samples of whole blood, serum, and plasma were collected weekly as well as daily during fever episodes. Plasma samples were stored at −80°C until used to determine plasma viral RNA level.
The challenge strain EVMX quasispecies was produced by combining equivalent infectious titers (TCID50) of the variant challenge strains EV0, EV6, and EV13 [34]. Viral stocks were prepared as previously described [34], [48]. Viral stock titers were determined utilizing our infectious center assay (cell-based ELISA) in fetal equine kidney cells, described previously [46].
Equine procedures were conducted at the Gluck Equine Research Center of the University of Kentucky according to protocols approved by the University of Kentucky IACUC. Thirty-six mixed age and gender outbred ponies, serognegative for EIAV, were utilized. Daily rectal temperatures and clinical status were recorded. CBC analysis of whole blood was performed using an IDEXX QBC Vet Autoreader. Hematocrit and platelet numbers were monitored weekly. The EIAVD9 stock was produced and vaccinations performed as described [31], [34]. Twenty-three vaccinated and six naïve ponies were challenged with 3×103 TCID50 of EVMX. The ponies were monitored daily for clinical symptoms of EIA, and blood was drawn at regular intervals (weekly, daily if febrile) for assays of platelets, viral replication, and virus-specific immune responses. During the course of these experiments ponies that demonstrated severe disease-associated symptoms resulting in distress as outlined by the University of Kentucky IACUC were euthanized.
Plasma samples from all subjects were analyzed for the levels of viral RNA per milliliter of plasma using a previously described quantitative real-time multiplex RT-PCR assay based on gag-specific amplification primers [60]. The standard RNA curve was linear in the range of 101 molecules as a lower limit and 108 molecules as an upper limit.
Our in vivo method for assessing immune reactivity to specific peptides, described previously [61], was used to explore cellular immune responses. Forty-four 20-mer peptides, overlapping by 10 residues, spanning the EIAVD9 gp90 were generated (GenScript USA Inc., Piscataway, NJ). An additional 15 peptides, specific for the variable regions ConD9 gp90, were included in the pool for analysis of those vaccinates. Vaccinates were screened for gp90 specific cellular immune responses one week prior to day of challenge. A 2 mm skin biopsy was collected and homogenized and RNA extracted. IFNγ gene expression was determined by real-time PCR, as previously described [61]. Amplification efficiencies were determined using Linreg [62] and samples with amplification efficiencies above 99% were included for further analyses. Beta-glucuronidase (β-GUS) was used as housekeeping gene and the ΔΔCT method [63] was used to determine relative gene expression with saline injection site for each vaccinate used as the calibrator. Relative quantity (RQ) was calculated as 2−ΔΔCT.
Serum IgG antibody reactivity to EIAV envelope glycoproteins was assayed quantitatively (end point titer) and qualitatively (avidity index, conformation ratio) using our standard concanavalin A (ConA) ELISA procedures as described previously [51]. Virus neutralizing activity to the historical reference strain EIAVPV, and vaccine-specific virus stocks EVCon and EVMX, mediated by immune sera, was assessed in an indirect cell-ELISA based infectious center assay using a constant amount of infectious EIAV and sequential 2-fold dilutions of serum [46], [51].
Significance of protection from disease was performed by survival curve analysis as implemented in GraphPad Prism version 6.0d (San Diego, CA). Significance of survival curves were determined utilizing One-way ANOVA with Bonferroni's multiple comparison's test as well as survival analysis of Kaplan Meier plots with Logrank test for trend.
All equine procedures were conducted in accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health at the Gluck Equine Research Center of the University of Kentucky according to protocols approved by the University of Kentucky IACUC (#01058A2006).
|
10.1371/journal.pcbi.1005800 | Bayesian refinement of protein structures and ensembles against SAXS data using molecular dynamics | Small-angle X-ray scattering is an increasingly popular technique used to detect protein structures and ensembles in solution. However, the refinement of structures and ensembles against SAXS data is often ambiguous due to the low information content of SAXS data, unknown systematic errors, and unknown scattering contributions from the solvent. We offer a solution to such problems by combining Bayesian inference with all-atom molecular dynamics simulations and explicit-solvent SAXS calculations. The Bayesian formulation correctly weights the SAXS data versus prior physical knowledge, it quantifies the precision or ambiguity of fitted structures and ensembles, and it accounts for unknown systematic errors due to poor buffer matching. The method further provides a probabilistic criterion for identifying the number of states required to explain the SAXS data. The method is validated by refining ensembles of a periplasmic binding protein against calculated SAXS curves. Subsequently, we derive the solution ensembles of the eukaryotic chaperone heat shock protein 90 (Hsp90) against experimental SAXS data. We find that the SAXS data of the apo state of Hsp90 is compatible with a single wide-open conformation, whereas the SAXS data of Hsp90 bound to ATP or to an ATP-analogue strongly suggest heterogenous ensembles of a closed and a wide-open state.
| In solution, many proteins adopt ensembles of multiple distinct states. The relative concentrations of the states are tightly controlled by factors such as pH, phosphorylation, or ligand binding, and a misbalance between the states underlies diseases such as cancer or neurodegeneration. However, detecting protein ensembles in experimental data has remained challenging. We present a statistically founded procedure for refining protein structures and ensembles against X-ray solution scattering data by combining atomistic simulations with Bayesian inference.
| Proteins are dynamic nanomachines that often populate heterogeneous ensembles of multiple distinct structural states. Controlling the relative population of such states is pivotal for the correct functioning of biological cells, and any misbalance between states may lead to severe conditions such as cancer or neurodegeneration. Detecting, understanding, and manipulating heterogeneous protein ensembles has therefore remained a central goal of molecular biophysics [1].
Deriving solution ensembles of proteins from structural experimental data has remained challenging, mainly because the information content of the data is typically insufficient to define all degrees of freedom of the ensemble [2, 3]. Consequently, upon fitting of structures or ensembles against experimental data, the data must be complemented by a physical model that restrains the protein into physically reasonable conformations, thereby reducing the risk of overfitting the model. Bayesian inference provides a route founded on probability theory for combining experimental data with physical models [4]. Applied to structure determination, Bayesian inference may become computationally expensive and technically challenging since it requires explicit sampling of the conformational space of the protein. However, it also holds a number of key advances over more simple optimization algorithms, as it provides statistically founded procedures (i) to weight the experimental data versus prior physical knowledge, and (ii) to quantify the uncertainty (or ambiguity) of the fitted structural model [5]. Due to its probabilistic rigor, Bayesian inference has been gaining increased popularity in various fields of biophysics, and it has hence been successfully applied for the refinement of structures against restraints from NMR, EPR, cryo-EM, and single-particle X-ray diffraction [5–11]. Following the pioneering work by Rieping et al., we refer to structural modeling based on Bayesian statistics as ‘inferential structure determination’ (ISD) [6].
Small-angle X-ray scattering (SAXS) is an increasing popular method that is in principle capable of detecting biomolecular structures and ensembles in solution [12, 13]. However, due to the low information content of SAXS data, refining structures or ensembles without overfitting poses a major challenge. For the refinement of individual structures against SAXS data, two routes have been suggested to reduce the risk of overfitting: first, during refinement, nearly all degrees of freedom of the biomolecule are constrained, leading to methods such as rigid-body modelling or normal mode fitting [14–17]. Second, physical information may be added to the low-information SAXS data, for instance by coupling a force field-based molecular dynamics (MD) simulation to the data with an energetic restraint [18–20]. Here, we follow the second route, building upon our method of SAXS-driven MD simulations [18]. SAXS-driven MD simulations drive biomolecular structures into conformations that are compatible with the data, using a differentiable harmonic restraint to the data. Critically, the method employs explicit-solvent calculations for predicting SAXS curves from the simulations frames, which were shown to provide accurate prediction for small and wide angles without the need of adjusting fitting parameters for the hydration layer or excluded solvent (see Fig 1) [21, 22]. In other words, the method uses a highly accurate and predictive ‘forward model’. However, as formulated previously, the method was not Bayesian and, consequently, did not yet benefit from advantages of ISD-related approaches (see above).
Many methods for the refinement of heterogenous ensembles against experimental data follow a “sample-and-select” strategy [23]. Accordingly, first, an ensemble is proposed by sampling from a computationally efficient physical model, such as a coarse-grained force field. Second, a limited number of structures or clusters are picked from the proposed ensemble. Third, the weights of the structures or clusters are modified in a statistically meaningful manner until the data back-calculated from the refined ensemble agrees with the given experimental data. Examples of this strategy in the context SAXS are the EROS, BSS-SAXS, and EOM methods [24–26], yet a number of related approaches have been suggested ([23] and references therein). Interesting recent developments proposed the Bayesian derivation of continuous ensembles from experimental data, including a development for SAXS data [27, 28].
In this work, we take an alternative approach for the refinement of ensembles against SAXS data. First, following the ISD approach, we embed SAXS-driven MD simulations into a Bayesian inference framework. Hence, we derive posterior distributions of protein structures in the light of the SAXS data and the applied force field. Because we simulate with a physically accurate all-atom force field, we employ an accurate and informative prior of protein structures. Additional unknown parameters, termed nuisance parameters in context of ISD [6], are not chosen ad hoc but instead estimated simultaneously with the protein structures. Specifically, two remaining fitting parameters as well as a systematic uncertainty due to the buffer subtraction are taken as nuisance parameters. Second, we extend the concept of ISD towards an ensemble of a small number of structural states, allowing us to estimate the structural weights simultaneously with the structures and the nuisance parameters. In combination, due to the commitment to Bayesian inference, the new method provides statistically founded confidence intervals for both the structures and the structural weights. In addition, we show that the posterior distribution for the structural weights can be used as a criterion for detecting the number of states in the ensemble that is required to explain the data.
We consider proteins that adopt an ensemble of a small number of distinct states. Typical examples would be proteins that exist in a mixture of active and inactive states, apo and holo states, or in a mixture of a few states along a more complex conformational cycle. We aim to derive the coordinates R = (R1, …, RN) as well as the relative weights (or concentrations) w = (w1, …, wN) of the states from given experimental SAXS data, where N is the number of states. Hence, the term ‘ensemble’ does not refer to the thermodynamic ensemble, but instead to a specific set (R, w). Notably, since the ensemble reduces to a single structure by setting N = 1, Bayesian structure refinement (instead of ensemble refinement) is contained in the method presented here as a special case.
Since the number of independent data points in a SAXS curve is much smaller than the number of degrees of freedom of any protein, it is unlikely that only a single ensemble (R, w) fits the SAXS data, but instead a wide range of ensembles are typically compatible with the data. A statistically founded procedure to infer the ensemble from the data can be formulated as the conditional probability p(R, w, θ|D, K) that quantifies the plausibility of the ensemble (R, w) in the light of the SAXS data D and prior physical knowledge K [4]. The symbol θ summarizes nuisance parameters, which are of limited interest but required for evaluating whether the ensemble (R, w) is compatible with the data D (see below). The posterior distribution is most conveniently evaluated using Bayes’ theorem:
p ( R , w , θ | D , K ) ∝ L ( D | R , w , θ , K ) π ( R | K ) π ( w | K ) π ( θ | K ) . (1)
Here, L(D|R, w, θ, K) denotes the likelihood that the data D were measured given the ensemble (R, w) and nuisance parameters θ. The functions π(R|K), π(w|K), and π(θ|K) denote the prior distributions of possible protein conformations, weights, and nuisance parameters. Due to the low information content of SAXS data, L(D|R, w, θ, K) provides only limited information, i.e., L is a wide function of R. Hence, in order to draw structural conclusions from the data, that is, to arrive at a reasonably narrow posterior distribution, it is critical to impose an informative tight prior π(R|K) of protein conformations, which is here achieved by applying an accurate physical model. In the method presented here, the prior π(R|K) is naturally given through an unbiased MD simulation, where K represents the physical laws and the force field underlying the MD simulation.
Formulating a likelihood function L for SAXS refinement is not straightforward because experimental SAXS data report purely statistical errors, whereas systematic errors, for instance due to poor buffer matching, are typically unknown. For data recorded at modern single-photon counting detectors, systematic error may dominate the overall uncertainty, suggesting that systematic errors strongly contribute to the likelihood L. In addition, for comparing experimental with calculated SAXS curves, free fitting parameters must be adjusted [21, 29–31]. Since both the systematic errors and fitting parameters are a priori unknown, a full Bayesian treatment requires that those parameter are simultaneously estimated with the structures and weights. Hence, systematic errors as well as fitting parameters are treated as nuisance parameters θ in the present method.
In practice, one is mainly interested in the ensemble (R, w), but not in the nuisance parameters θ. The statistically correct way of reducing the general posterior p in Eq 1 to the posterior of the structural ensemble is to marginalize out the nuisance parameters,
p ( R , w | D , K ) = ∫ d θ p ( R , w , θ | D , K ) . (2)
In our method, the fitting parameters are marginalized out analytically at the level of the likelihood, whereas systematic errors are explicitly sampled and marginalized out numerically (Methods and materials). To visualize the high-dimensional p(R, w|D, K), the posterior may be further projected onto intuitively important coordinates, such as the distance between two protein domains, the radius of gyration of the protein, or the weight of an interesting state.
By taking the negative logarithm of the posterior, Eq 1 takes the form of a hybrid energy that is commonly applied for structure refinement [5, 9], yet corrected with the contributions from the prior distributions:
E hybrid = V ff ( R , K ) + E exp ( R , w , θ , D , K ) - β - 1 ln [ π ( w | K ) π ( θ | K ) ] . (3)
Here, the posterior was identified with a hybrid energy Ehybrid = −β−1 ln p(R, w, θ|D, K), where β denotes the inverse temperature. The prior for the protein structures is taken from the MD force field energy as Vff(R, K) = −β−1 ln π(R|K), after marginalizing out the solvent coordinates (see Methods and materials). The experiment-derived energy is given via the likelihood, Eexp = −β−1 ln L, adding an energetic penalty if the SAXS curve calculated from the ensemble (R, w) is incompatible with the data D.
Having translated the probabilities into energies, all parameters can be sampled using established methods. Accordingly, sampling of protein structures R is conducted using Newtonian dynamics. Here, the force on atom ℓ is given via gradients of the hybrid energy with respect to the atomic positions, Fℓ = −∇ℓ Ehybrid, evaluated at fixed w and fixed nuisance parameters. The fitting parameters, as shown below, are marginalized analytically at the level of the likelihood. The remaining nuisance parameter, namely the systematic error σbuf, as well as the weights w are sampled using Gibbs sampling, that is, Monte-Carlo moves at fixed protein coordinates R. Calculations of the SAXS intensity and intensity gradients from (R, w), as required for sampling the posterior, were conducted with the explicit-solvent algorithms established previously [18, 21], taking accurate atomistic models for both the hydration layer and the excluded solvent (Fig 1). Details on the likelihood function, assumed priors, force calculations, and sampling algorithms are provided in the Methods and Materials.
The weights w are normalized and have non-negative elements, i.e., the relevant weight space is given by the (N − 1)-simplex. Sampling of the weight space was accelerated using umbrella sampling along the weights [32]. This is computationally convenient because it allows calculation of the posterior from a set of short independent simulations. More critically, this allows us to compute the posterior for the complete weight space, including the “edge” of the simplex, where at least one of the weights wj is zero (1 ≤ j ≤ N). However, note that weight vectors w with elements equal to zero specify smaller ensembles with a reduced number of states. Consequently, the posterior of an ensemble of N states includes all smaller ensembles as a special case, thereby proving a probabilistic criterion for choosing the number of states required to explain the experimental data: if the posterior peaks at the edge of the simplex, a smaller ensemble provides a plausible model; in turn, if the posterior near the edge is small compared to the posterior’s maximum, a smaller ensemble is implausible.
In the following, Bayesian ensemble refinement is demonstrated for two test proteins: leucine binding protein (LBP) using calculated SAXS data and heat shock protein 90 (Hsp90) using experimental SAXS data. We assumed that both proteins adopt a two-state ensemble of an open and a closed structure (N = 2). We further assume that the closed structure is known, whereas (i) the coordinates of the open structure as well as (ii) the relative open/closed weights are simultaneously refined against SAXS data. Such scenarios are quite common, as a compact holo or ground state structure might be accessible to X-ray crystallography, whereas more flexible apo or excited state structures often do not crystallize. Applying the method proposed here to larger ensembles of N > 2 is conceptually possible but beyond the scope of this article. With increasing number of states N, due to increasing number of required umbrella windows, the computational cost would scale exponentially with N − 1.
LBP is a typical representative of the superfamily of periplasmic binding proteins involved in chemotaxis and solute uptake over membranes [33]. LBP is a well-characterized two-domain protein, exhibiting a transition from an open (apo) to a closed (holo) state triggered by ligand binding (Fig 2A/2B) [34, 35]. Free simulations of the closed and open state suggested center-of-mass distances between the N- and C-terminal domains of ∼3 nm and 3.25 nm, respectively, which is compatible with experimental SAXS data of the homologous LIVBP [18]. We theoretically computed SAXS curves of the open and closed states (Fig 2C, solid lines), as well as linear combinations, thereby modeling SAXS data from heterogeneous ensembles of known open/closed weights of 0:100, 25:75, 50:50, 75:25, and 100:0 (Fig 2C, dashed lines).
The posteriors of the ensembles p(R, w|D, K) refined against these five SAXS curves are presented in Fig 2D. To visualize the high-dimensional posterior, we projected the posterior onto two characteristic coordinates: (i) the weight of the open state wopen, implying the weight (1 − wopen) for the closed state, and (ii) the interdomain distance dNC of the open state (illustrated in Fig 2B). Evidently, all derived posterior distributions peak at the correct wopen. In addition, the posteriors refined against SAXS curves of non-zero open-state content (Fig 2D, four right panels) peak at the physically correct interdomain distance of ∼3.25 nm (Fig 2, see also the marginalized posteriors in S1A Fig). In addition, the RMSD to the mean open structure taken from umbrella simulations, restrained to weights at the maxima of the respective posterior, reveals that the refinement simulations rapidly approach the correct open state (S5 Fig). These findings demonstrate that the MD simulations were capable of translating the information in the SAXS curve into the underlying heterogeneous open/closed ensemble.
The width of the posteriors rigorously quantify the degree of structural knowledge that can (and cannot) be inferred from the SAXS curve, i.e., the posteriors quantify the ambiguity of the refined ensemble. For the LBP ensemble refinement, the marginalized posteriors suggest 65% confidence intervals (CI) for wopen and dNC in the order of ±15% and ±0.07 nm, respectively (Fig 3A and S1A Fig, S1 and S2 Tables). In addition, the posteriors in Fig 2D suggest some correlation between wopen and dNC, as apparent from the posterior’s diagonal elongated shapes, suggesting that the SAXS curves are compatible with an increased wopen given that the open state is modeled more compact.
Hsp90 is a chaperone that interacts with more than 200 proteins in eucaryotic cells [39–42]. It constitutes up to 2% of the cellular protein mass [43]. Since many client proteins of Hsp90 are oncogenic, Hsp90 has been suggested as a promising target for anti-cancer therapies [44, 45]. Structurally, Hsp90 is a homodimer, where each protomer contains three domains: an N-terminal domain with the ATP binding site, a middle domain forming the interaction sites for client proteins, and a C-terminal domain responsible for Hsp90 dimerization (Fig 4A). Crystallographic, cryo-EM, and SAXS studies established that Hsp90 carries out large-scale conformational transitions between a V-shaped open state and a compact closed state, controlled by binding of ATP, ATP analogues, and client proteins [38, 46]. However, ligands do not dictate a single well-defined state, but instead merely shift the equilibria of heterogeneous ensembles between open and closed conformations [47, 48]. Only recently it was found that sufficient time spent in the open state is crucial for correct Hsp90 functioning, highlighting the importance of controlling the open/closed equilibria of the chaperone [49].
Based on experimental SAXS data of yeast Hsp90 in the apo state, Hsp90 bound to the slowly hydrolyzing ATP-analogue AMPPNP, and Hsp90 bound to ATP (Fig 4C, colored curves [37]), we derived heterogeneous solution ensembles of the Hsp90 dimer using Bayesian ensemble refinement. Hsp90 ensembles were modeled as two-state ensemble of (i) the closed state, taken from the yeast crystal structure (Fig 4A), and (ii) an initially unknown open state. Starting the simulations from a nearly closed conformation, both the structure and the relative weight wopen of the open state were simultaneously refined against the SAXS data. The SAXS curves of the refined two-state ensembles exhibited reasonable agreement with the experimental curves (Fig 4C).
The residuals between calculated and experimental SAXS curves are analyzed in Fig 5. Here, panel (A) shows the residuals normalized purely by the statistical experimental errors, ΔI(q)/σexp. The large residuals at low q (Fig 5A, red and yellow) reflect that the MD force field did not allow conformations that accurately fullfil the data within statistical experimental errors, possibly because accurately reproducing the data would require an energetically unfavorable conformational transition (such as partial unfolding). In other words, the Bayesian analysis revealed that, in the light of the force field, substantial systematic errors at low q are more plausible than an ensemble that accurately matches the experimental data. Indeed, as shown in Fig 5B the residuals normalized with respect to the total errors including statistical and systematic errors, ΔI(q)/σtot, reveal reasonably low values over the entire q-range. As outlined in the Methods, we modelled systematic errors as a consequence of poor buffer matching, but the analysis can not exclude other sources of remaining discrepancies such as a small fraction of aggregated Hsp90. Further, in this work, we can not fully exclude the possibility that a more continuous ensemble, as supported by recent Förster resonance energy transfer (FRET) study [50], might provide a more accurate description of the experimental conditions.
Fig 4D presents the posterior distributions p(R, w|D, K) of the Hsp90 ensembles, projected onto two intuitive degrees of freedom: (i) the weight wopen of the open state, implying the weight (1 − wopen) for the closed state, and (ii) the radius of gyration R g open of the refined open state, which naturally quantifies the degree of opening of the open state. The marginal posteriors p ( R g open | D , K ) for the three ensembles, obtained by marginalizing the posteriors in Fig 4D with respect to wopen, are presented in Fig 6 as colored lines. Evidently, the refined structures of the open state were similar in all three ensembles, exhibiting large R g open values of ∼6.3nm. These R g open values are ∼1.3 nm and ∼1.7 nm larger than the radii of gyration of the open form of the bacterial HtpG homologue in the crystal and in solution environment, respectively [38, 48], but they are compatible with previously reported open conformations of eukaryotic apo Hsp90s [51]. Hence, the open structures of the three refined open/closed heterogeneous ensembles are characterizing by a wide open conformation, as visualized in Fig 4B.
Fig 4E presents the marginal posteriors of the weight of the open state, p(wopen|D, K), obtained by marginalizing the posteriors in Fig 4D with respect to R g open. Evidently, wopen strongly differs between the three ensembles. The posteriors suggest closed:open populations of 68:32 and 52:48 for the AMPPNP- and ATP-bound states, respectively, with 65% confidence intervals of ±18% (S4 Table). Hence, for the AMPPNP- and ATP-bound states, a model of a single state is very implausible in the light of the MD force field and the SAXS data. These findings resemble results from rigid-body SAXS modeling of a bacterial HtpG homologue that suggested heterogeneous closed/open ensembles in the AMPPNP-bound state, yet without providing confidence intervals [48]. For the Hsp90 apo state, the posterior suggests that wopen is with 65% confidence within the interval [78%,100%], suggesting that a single open state as well as a heterogeneous ensemble with a large wopen are both compatible with the SAXS data and the MD force field.
We have presented a method for the refinement of a single protein structure or of an ensemble of structures against SAXS data, applicable to ensembles of a small number of distinct states. By combining Bayesian inference with atomistic MD simulations, the method is capable of inferring the structures and structural weights that gave rise to the SAXS data. The method does not merely derive a single “best fit” against the experimental data, but instead provides the joint posterior distribution of structures and weights, thus quantifying the plausibility of all possible structures and ensembles in the light of data D and physical knowledge K. The width of the posteriors yield confidence intervals founded on probability theory for both the structures and the weights, that is, the method quantifies the precision of the refined ensemble. Such reliable confidence intervals are required for deciding whether a structural model is convincingly supported by available SAXS data, or whether additional data are needed to unambiguously prove a model. We stress that the confidence intervals derived here should not be confused with the spread of “best fits” obtained by multiple repetitions of an optimization algorithm, as common, for instance, when fitting low-resolution bead models against SAXS data [52]. Repeated best fits test the convergence of the optimization algorithm but do not provide a statistically founded confidence interval.
Since we enforced exhaustive sampling of the weight space using umbrella sampling, the posterior includes smaller ensembles with a reduced number of states as a special case, as given by weight vectors w with one or multiple zero elements. We showed that this feature provides a rigorous criterion for deciding on the number of states required to explain the experimental data. For the apo state of Hsp90, we found that the SAXS data is compatible with a single open state, as well as with a heterogenous open-closed ensemble with a large weight of the open state. In contrast, for the AMPPNP- and ATP-bound states of Hsp90, we found that a single state is unlikely in the light of the SAXS data and the MD force field, whereas a model of two states provides a much more plausible model. Critically, Bayesian inference further allows us to assign a confidence to these qualitative statements. Namely, the odds that a single state underlies the SAXS data is 80% for the apo state, 20% AMPPNP-bound, and 6% for the ATP-bound state. As such, the researcher may decide whether such confidence is sufficient to decide on the number of states, or whether additional data, e.g. from FRET, should be included to further increase the confidence on the number of states [53].
A property of Bayesian methods is that the computed posterior depends on the chosen priors. For Bayesian SAXS refinement, the posterior p(R, w) most critically depends on the prior for protein conformations π(R, K), which is given through the applied force field. In this work, we applied a physically accurate all-atom force field, which provides a more accurate description of the energy landscape as compared to rigid-body or coarse-grained force fields. However, despite major force field improvements in recent years [54], it can not be excluded that certain force fields bias the refinement simulations towards unphysical states, in particular for proteins with large disordered regions [55]. Hence, we recommend to use the most recent and best validated force fields.
Depending on the size of the system and the inherent autocorrelation times, exhaustive sampling of the posterior may become challenging. Due to the use of umbrella sampling along wopen, we here observed rapid convergence of the marginalized posterior p(wopen|D, K), both for LBP and Hsp90 (S6 Fig). The 2D posterior p(wopen, dNC|D, K) for LBP seemed converged at moderate computational effort of 50 ns per umbrella window (Fig 2D), whereas the 2D posterior p ( w open , R g open | D , K ) for Hsp90 converged more slowly, as apparently from the somewhat scattered posteriors (Fig 4D). Hence, in future refinement simulations of very large systems such as the system of Hsp90, the sampling may benefit from additional enhanced sampling methods.
The computational cost of the simulations presented here are increased by only ∼15% as compared to standard MD simulations, suggesting that the calculations are well feasible on modern hardware. However, MD simulations are obviously more expensive than simplified methods such as rigid-body modelling.
The sampling of structural weights has some similarity with previous sample-and-select methods that reweighed a set of structures against SAXS data using, for instance, Bayesian or maximum-entropy criteria [16, 24–26]. However, at variance with previous methods, we refined the weights simultaneously with the structures, fitting parameters, and systematic errors. This difference is not a technical subtlety but is instead critical to estimate the correct uncertainty of the weights: In our method, by commitment to Bayesian inference, the uncertainty (or ignorance) about the structure, fitting parameters, and systematic errors are propagated into the uncertainty of the weights. In other words, when estimating the weights, and in contrast to previous methods, we do not assume any precise knowledge about the structure, fitting parameters, and systematic errors that, in truth, we do not have. This difference rationalizes why the uncertainties of fitted weights reported previously are much smaller than the uncertainties derived here via the full Bayesian treatment [24].
The refinement simulations presented here differ from previous methods for structure and ensemble refinement against SAXS data by a number of additional elements. First, since our refinement simulations are steered by the experimental SAXS data, the simulations are capable of sampling large-scale conformational transitions, which would not be sampled in an equilibrium simulation due to limited simulation time. An example is the open/close transition of Hsp90 that occurs on the time scale of many seconds at experimental conditions [49]. As such, our method does not strictly require the application of coarse-grained simulations [24, 25] or other simplified physical models [16, 26] to visit the relevant conformational states. Second, because we apply purely the MD force field and the SAXS data but no additional constraints, the refinement is not limited to rigid-body motions or normal modes, which were previously used to refine structures against SAXS data [14–17]. Hence, prior to the refinement simulations, our method does not require the ad-hoc definition of rigid bodies and flexible linkers, which may not be obvious. Third, in contrast to previous refinement methods, SAXS curves were computed using explicit-solvent algorithms, avoiding any solvent-related fitting parameters [21, 56].
In this study, we built upon the concept of “inferential structure determination” (ISD), which was originally formulated to model NMR data with a single structural state [5, 6]. In short, we presented an ISD method for SAXS data using all-atom MD simulations. In addition, we extended the ISD concept towards the refinement of a small number of states (typically two states), but the method is not intended for the refinement of continuous and highly heterogeneous ensembles. Hence, our approach complements methods for the reweighing of continuous ensembles against experimental data, as required for modeling of intrinsically disordered proteins [27, 57, 58], and it further complements maximum-entropy-based methods for biasing ensembles with experimental data [59–61].
We developed the method with a focus on SAXS data, but the calculations may be readily complemented by other sources of structural information. For instance, the refinement may be additionally guided by multiple sets of small-angle neutron scattering (SANS) data, optionally measured at various D2O contrasts and differently deuterated solutes. Similar to the SAXS-guided refinement, such SANS-guided refinement simulations will benefit from the fitting-free explicit-solvent scattering calculations applied here. Alternatively, the refinement simulations may be complemented by additional distance restraints from double electronelectron resonance (DEER) or FRET. Such future developments, complementing the method proposed here, may provide a route to MD-based Bayesian integrative modeling.
A common source of systematic errors in SAXS experiments is poor buffer matching. We therefore modeled the systematic errors σbuf as a consequence of a buffer density mismatch δρbuf between the protein solution and the pure buffer. Following previous work [18], δρbuf can be translated into an uncertainty σbuf of the calculated intensity Ic(q), contributing to the likelihood function (see below).
We recently found excellent agreement of SAXS curves predicted from explicit-solvent MD simulations with experimental curves, if the experimental curves Iexp(q) were adjusted by only two fitting parameters following Ifit(q) = fIexp(q) + c, where f denotes the fitted absolute scale and c denotes a fitted constant offset [21], and q is the momentum transfer. Hence, we take for the likelihood function
L ( D | R , w , θ , K ) ∝ exp [ - N indep 2 N q ∑ i = 1 N q [ I c ( q i , R , w ) - ( f I exp ( q i ) + c ) ] 2 f 2 σ exp 2 ( q i ) + σ calc 2 ( q i ) + σ buf 2 ( q i ; δ ρ buf ) ] , (4)
where θ = {f, c, δρbuf}. As shown below, the fitting parameters f and c can be marginalized out analytically. The symbols σexp and σcalc denote statistical errors of the experimental and calculated intensities, respectively. The calculated intensity Ic is a weighted average over the intensities of the N states, I c ( q i , R , w ) = ∑ j = 1 N w j I ( q i , R j ). The symbols Nq and Nindep denote the total and the independent number of data points in the SAXS curve, respectively. Nindep was estimated by the number of Shannon-Nyqvist channels Nindep = qmax Dp/π, where Dp is the maximum diameter of the protein and qmax is the maximum momentum transfer of the SAXS curve [13]. Hence, the factor Nindep/Nq is an empirical correction that accounts for the fact that the number of independent data points Nindep in a SAXS curve is typically much smaller than the number of q-points Nq reported in experimental SAXS curves. Without the factor Nindep/Nq, the information content in the data would be overrated in comparison with the information in the priors. Critically, this correction assumes that the data Iexp corresponds to a “smoothed” SAXS curve, and that the experimental errors σexp(qi) denote the true uncertainty of point qi in the light of correlations of Iexp along q.
A flat prior was applied for the fitting parameters, π(f) = π(c) = 1. Notably, since the likelihood function is nonzero only for a very narrow f-range, applying the scale-invariant Jeffreys’ prior would change the posterior only marginally. The prior for the protein structure Rj of state j was taken from an unbiased MD simulation. Hence, π(Rj|K) is given by a Boltzmann factor of the force field energy Vff, marginalized with respect to all solvent coordinates rsol (water and ions), π(Ri|K) ∝ ∫drsol exp[− βVff(Ri, rsol, K)]. Assuming no prior information on the weights, π(w|K) was taken as a flat Dirichlet distribution. For the buffer density mismatch δρbuf, a Gaussian prior was taken as π ( δ ρ buf ) ∝ exp [ - δ ρ buf 2 / ( 2 ϵ buf 2 ) ]. Here, ϵbuf is a free parameter that quantifies the uncertainty of the density of an experimental buffer. Typical values for ϵbuf would be 0.1 to 0.5% of the density of water, yet we found that the choice for ϵbuf had only a small effect on p(R, w).
The buffer-subtracted SAXS curves were derived by explicit-solvent calculations, as described previously [18, 21]. Because the explicit solvent provides an accurate model for the hydration layer and excluded solvent, these calculations did not require any solvent-related fitting parameters, in contrast to implicit-solvent SAXS calculations.
In short, a spatial envelope was constructed around the protein at a distance of at least 8Å from all protein atoms (Fig 1). All protein and solvent atoms within the envelope were taken into account for the calculation of the SAXS curve, as visualized in Fig 1. Likewise, scattering contributions from the excluded solvent were computed from solvent atoms within the envelope taken from a pure-water MD simulation. A memory time constant of τ = 500 ps was applied during both LBP and Hsp90 simulations. The orientational average (or spherical quadrature) was conducted numerically using 1200 q-vectors per absolute value of q, distributed by the spiral method. During SAXS refinement simulations, the SAXS curves were updated on-the-fly every 0.5 ps. The statistical uncertainty σc of the calculated intensity was computed by applying standard Gaussian error propagation to the SAXS intensity calculations we described previously [21]. After averaging over a few hundred MD frames, σc is typically small compared to the other uncertainties that contribute to the likelihood function (σbuf and σexp).
The SAXS curves of the purely open and purely closed states of LBP (Fig 2, solid lines) were computed from 100-nanosecond free simulations of the open and closed state.
The two fitting parameters f, corresponding to the absolute scale of the SAXS curve, and the offset c, can be marginalized out analytically at the level of the likelihood. Assuming Gaussian errors, we take for the likelihood
L(D|R,w,f,c,δρbuf,K)∝exp[ −12ζχ2 ] (5)
with
χ 2 = ∑ i = 1 N q τ i , f [ I c ( q i , R , w ) - ( f I exp ( q i ) + c ) ] 2 , (6)
where we introduced the symbol ζ = Nindep/Nq, as well as the precision of the ith q-point as follows:
τi,f=1σ2(qi)=1f2σexp2(qi)+σc2(qi)+σbuf2(qi;δρbuf). (7)
Here, we used that the uncertainties from the experiment σexp, from the calculation σc, and from the buffer subtraction σbuf are independent, suggesting that the respective variances add up to the total variance σ2(qi). The precision τi,f depends on the fitted scale f because the experimental errors σexp must be scaled simultaneously with the experimental intensities Iexp. To allow us to marginalize out the scale f analytically, we use that the errors in the small-angle regime are much smaller than the intensities, suggesting that purely values of f close to it’s maximum-likelihood estimate fml contribute to the marginalized likelihood. As a consequence, replacing f by fml in eq 7 has only a small effect on the marginalized likelihood. We therefore use for the precision in the following
τi=[ fml2σexp2(qi)+σcalc2(qi)+σbuf2(qi;δρbuf) ]−1. (8)
In the first calculation step, while fml is still unknown, it may be simply estimated from the non-weighted averages of the calculated and experimental intensities, following fml ≈ ∑i Ic(qi)/∑i Iexp(qi).
To keep the nomenclature clear, let us introduce additional symbols. Let T : = ∑ i = 1 N q τ i denote the sum over all precisions. The τ-weighted average over q-points is
⟨ X ⟩ = T - 1 ∑ i = 1 N q τ i X ( q i ) . (9)
With the last definition, the τ-weighted variances of the calculated and experimental SAXS intensities are
s c 2 = ⟨ I c 2 ⟩ - ⟨ I c ⟩ 2 (10) s exp 2 = ⟨ I exp 2 ⟩ - ⟨ I exp ⟩ 2 , (11)
respectively, and the τ-weighted Pearson correlation coefficient between the calculated and experimental data points is
P = ⟨ I c I exp ⟩ - ⟨ I c ⟩ ⟨ I exp ⟩ s c s exp . (12)
The maximum likelihood estimates for the fitting parameters f and c are
f ml = P s c s exp (13) c ml = ⟨ I c ⟩ - f ml ⟨ I exp ⟩ , (14)
respectively. The residual between Ic and Iexp that cannot be fitted by the parameters f and c is
χ ^ 2 = T [ s c 2 - ( f ml s exp ) 2 ] (15) = T ⟨ [ I c - ( f ml I exp + c ml ) ] 2 ⟩ . (16)
The last equality is derived using eqs 10 to 14. The likelihood Lmarg marginalized with respect to fitting parameters f and c is obtained by integrating over f and c. Since no prior information on f and c is available, we assumed flat prior distributions, π(f) = π(c) = 1. A straightforward calculation yields:
L marg ( D | R , w , σ buf , K ) ∝ ∫ d f ∫ d c L ( D | R , w , f , c , σ buf , K ) π ( f ) π ( c ) ∝ 1 T s exp exp ( - 1 2 ζ χ ^ 2 ) (17)
Here, we dropped the normalization factors and other constants of the likelihood because these only lead to an irrelevant constant offset in the experiment-derived energies.
In order to sample the posterior distribution using Newtonian dynamics, Lmarg is reformulated as its energy analogue
E exp = - β - 1 ln L marg . (18)
Using eqs 10 through 18, the experiment-derived force on atom ℓ of state j to is
F j , ℓ = - ∂ ∂ r j , ℓ E exp (19) = - β - 1 ζ ∑ i = 1 N q τ i [ I c ( q i ) - ( f ml I exp ( q i ) + c ml ) ] ∂ I c ( q i , R , w ) ∂ r j , ℓ , (20)
where rj,ℓ denotes the Cartesian coordinates of atom ℓ in state j (j = 1, …, N). In general, the calculated SAXS intensity Ic is a weighted average over the intensities of the N states:
I c ( q i , R , w ) = ∑ j = 1 N w j I ( q i , R j ) , (21)
where wj and I(qi, Rj) denote the normalized weight (∑j wj = 1) and the SAXS intensity of state j, respectively. Following eq 21, the derivative of Ic with respect to rj,ℓ depends purely on the SAXS intensity of state j:
∂ I c ( q i , R , w ) ∂ r j , ℓ = w j ∂ I ( q i , R j ) ∂ r j , ℓ . (22)
Note that, for the simulations conducted in this study, one closed state (j = 1) was assumed to adopt a fixed know structure, whereas a second open state (j = 2) was refined against SAXS data. Hence, the forces Fj,ℓ were purely evaluated for the second flexible state. However, the equations presented above are suitable for simultaneously refining multiple states against SAXS data. The derivative ∂I(qi, Rj)/∂rj,ℓ was computed as described previously [18, 21].
For the simulations of this study, we applied the likelihood function defined in eqs 5 and 6, using both the absolute scale f and the constant offset c as unknown fitting parameters. However, there may be applications for which the fitting of a constant offset c is undesirable. Hence, for the sake of completeness, we report the expressions for marginalizing out purely the absolute scale f.
Then, the likelihood takes the form of eqs 5 and 6 with the parameter c set to zero. The maximum-likelihood estimate for the scale f evaluates to f ml ′ = 〈 I c I exp 〉 / 〈 I exp 2 〉, and the residual between Ic and Iexp changes to
χ ′ ^ 2 = T ( ⟨ I c 2 ⟩ - ⟨ I c I exp ⟩ 2 / ⟨ I exp 2 ⟩ ) = T ⟨ [ I c - f ml ′ I exp ] 2 ⟩ . (23)
The marginalized likelihood is
L marg ′ ∝ 1 [ T ⟨ I exp ⟩ ] 1 / 2 exp ( - 1 2 ζ χ ′ ^ 2 ) , (24)
and the force on atom ℓ of state j F j , ℓ = - β - 1 ζ ∑ i = 1 N q τ i [ I c ( q i ) - f ml I exp ( q i ) ] ∂ I c ( q i ) ∂ r j , ℓ . (25)
The weights of the N states (N = 2 in this study), as well as the uncertainty of the buffer density δρbuf were sampled using Gibbs sampling, that is, using Monte-Carlo (MC) moves with all other parameters fixed. At each time step at which the SAXS intensities were updated (0.5 ps in this study), 20 rounds of MC moves of δρbuf and wopen were conducted. In each round, 20 MC moves of δρbuf were conducted (at fixed wopen), followed by 20 MC moves of wopen (at fixed δρbuf). Typical posteriors of the parameter δρbuf are shown in S3 Fig.
Proposed MC moves of δρbuf were taken from a uniform distribution in the interval [0, 6ϵbuf). Proposed MC moves for the weight vector w = (w1, …, wN) were taken from a flat Dirichlet distribution. Hence, proposed w satisfied ∑ i = 1 N w i = 1 and were uniformly distributed over the standard (N − 1)-simplex, that is, the prior π(w) was a constant. Such w were drawn from the flat Dirichlet distribution by randomly partitioning the interval [0, 1], as follows:
We noticed that restricting the sampling of wi to the interval [0, 1] may lead to artifacts at “edge” of the (N − 1)-simplex, presumably as a consequence of the weighted running averages used for computing SAXS curves on-the-fly during MD simulations [18]. To avoid a boundary in the physically relevant weights space, the sampled weight space was extended to unphysical but mathematically well-defined weights slightly outside the (N − 1)-simplex (outside the interval [0, 1] in case of N = 2). This was achieved by scaling the proposed weight vector w, followed by a shift along the vector with all elements equal to unity, j = (1, …, 1), as follows:
w ′ = ( 1 + ξ N ) w - ξ j . (26)
The parameter ξ was set to 0.1 in this study. This transformation keeps the prior of w′ uniform on the (N − 1)-simplex, and it keeps the weight vector normalized (∑ i = 1 N w i ′ = 1). However, it allows one to draw samples of w i ′ from the interval [−ξ, 1 + Nξ − ξ]. For N = 2, for instance, samples of w i ′ are drawn from the interval [−ξ, 1 + ξ].
The proposed MC move was accepted with probability Paccept according to the Metropolis algorithm,
P accept = min { 1 , p marg ′ / p marg } , (27)
where the prime indicates the posterior after the MC move. Further, the symbol pmarg denotes the posterior distribution after marginalizing out the fitting parameters, which is given by
pmarg(R,w,δρbuf|D,K)∝Lmarg(D|R,w,δρbuf,K)π(R|K)π(w|K)π(δρbuf|K). (28)
For each MC move pmarg was evaluated using eq 17 as well as the priors for w (a constant in this study) and π(σbuf) (a Gaussian in this study, see section on prior distributions).
Obtaining a (reasonably) converged posterior distribution as a function of weights and protein coordinates would require very long simulations. To ensure exhaustive sampling of the weights space and, hence, to accelerate the convergence of the posterior, we used umbrella sampling along the weights [32]. Further, umbrella sampling is technically convenient because it allows the calculation of the posterior from a set of independent simulations.
For the two-state refinement used here, one-dimensional umbrella sampling was sufficient. Accordingly, the weight of the open state wopen was decomposed into Nwin = 11 umbrella windows w open ( k ) = { 0 , 0 . 1 , … , 1 . 0 } (k = 1, …, Nwin). During MC moves of the weights, a harmonic umbrella potential was applied V k ( b ) = f w ( w open - w open ( k ) ) 2 / 2 or, equivalently, the MC moves were accepted or rejected based on the biased posterior
p marg , k ( b ) = p marg e - β V k ( b ) . (29)
An umbrella force constant of fw = 1000 kJ/mol was applied. A typical set of umbrella histograms is shown in S4 Fig, demonstrating sufficient overlap between neighboring histograms. After the simulations had finished, the umbrella windows were combined to the unbiased posterior using the weighted histogram analysis method (WHAM), as implemented in the g_wham software [62, 63].
Fig 7 visualizes the algorithm used to compute the posteriors. Accordingly, the simulation system is set up from the initial coordinates R, and initial values for the weights w and the buffer density mismatch δρbuf are defined. The system is freely simulated for τ (the memory time constant for on-the-fly SAXS calculations [18]), using purely the MD force field Vff. The free simulation is required is required because, using the explicit-solvent SAXS predictions, the SAXS curve cannot be computed from a single frame but instead requires averaging over solvent fluctuations. Within the free simulation, an initial estimate for the calculated SAXS intensity Ic(qi, R, w) is obtained. A typical value for τ is 300 ps, suggesting that the computed SAXS curves account for fluctuations on the several hundred picosecond time scale. Subsequently, the experiment-derived energy Eexp is gradually turned on within the following τ.
The following steps are repeated until the requested simulation time is reached for each umbrella window along the weights: (i) MD simulation using forces derived from the hybrid energy, i.e., using the potential Vff + Eexp; (ii) update of Ic(qi, R, w) based on the current MD frame and using a cumulative weighted average [18], as previously suggested for NMR refinement [64]; (iii) a few hundred MC moves of weights wi and δρbuf (see above); (iv) update Ic(qi, R, w) with the final weights, and update the systematic error σbuf with the final δρbuf, as described previously [18]. After the simulations from all umbrella windows have finished, the biased posteriors from all windows are combined into the unbiased posterior using WHAM [62].
The crystal structures of the apo and holo states of LBP were taken from the protein data bank (PDB codes 1USG and 1USI [35]). For the simulation of Hsp90 the structure of ATP-bound yeast Hsp90 was used (PDB code 2CG9) [36]. The co-chaperone proteins SBA1 and ATP ligands were removed from the structure of HspP90 and leucine ligand was removed from the LBP structure. Flexible linkers missing in the Hsp90 crystal structure were added using Modeller [65]. A swap of the N-terminal β-strand (residues 1-9), which prevented the opening of the protein, was removed using the Coot software [66].
The structures were placed in simulation boxes of a rhombic dodecahedron with distance between the protein and the box surface of 1.5 and 4 nm for LBP and Hsp90, respectively. The systems were solvated by explicit water. Sodium and chloride ions were added to obtain a salt concentration of 100 mM. Here, the number of sodium and chloride ions was adjusted to neutralize the system. The energies of the systems were minimized using the steepest descent algorithm for 2000 steps. Subsequently, the systems were equilibrated with position restraints on the backbone atoms for 10 and 20 ns for LBP and Hsp90, respectively.
To obtain an initial open structure of Hsp90, we carried out pulling simulations along the distance of the two N-terminal domains. Accordingly, the center-of-mass distance between the two N-terminal domains was increased from 4 nm to 8 nm within 40 ns, using a pulling speed of 0.1 nm/ns. The obtained open structure was resolvated in a larger simulation box with the distance between the protein and the box surface of 3 nm, and the structure was equilibrated for another 20 ns with position restraints on the backbone atoms. The final structure was used as a starting structure for SAXS refinement. The Hsp90 system contained approximately 1.5 × 106 atoms.
Standard MD simulations were performed using the GROMACS simulation software (version 4.6) [67]. All SAXS calculations were done with an in-house modification of GROMACS 4.6, which is available from the authors upon request. Protein interactions of LBP and Hsp90 were taken from the CHARMM27 and CHARMM22* forcefields, respectively [68, 69], and water was modeled by the TIP3P potential [70]. Hydrogen atoms of the proteins were modeled as virtual interaction sites allowing an integration timestep of 4 fs. Electrostatic interactions were treated with the particle-mesh Ewald scheme [71, 72]. The cutoff of 1.2 nm was applied to the direct-space Coulomb and Lennard-Jones interactions. The bond lengths and angles of water molecules were constrained with the SETTLE algorithm [73], and all other bonds were constrained with LINCS [74]. The pressure was set to 1 bar using the Berendsen barostat (τ = 1 ps) [75]. During equilibration runs, the temperature was controlled at 300 K with the Berendsen thermostat (τ = 0.5 ps) [75]. During SAXS-driven simulations, in contrast, a tight stochastic dynamics integration scheme was applied, motivated from the fact that SAXS-driven MD is not strictly energy conservative [76].
For LBP simulations, the target curves for the refinement were modeled from calculated SAXS curves of the closed state Iclosed(q) (Fig 2C, solid dark blue curve) and open state Iopen(q) (Fig 2C, solid light blue curve), as follows:
I exp , w ( q ) = w open exp I open ( q ) + ( 1 - w open exp ) I closed ( q ) (30)
In this study, we tested ensemble refinement against SAXS data computed with the following w open exp: 0, 25%, 50%, 75%, or 100% (Fig 2C, solid and dashed curves). Hence, since Iexp,w(q) was computed theoretically, the true weight of the open state was known, allowing us
All simulations of LBP were started from the closed state (Fig 2A). The simulations were coupled to the target SAXS curve at Nq = 25 q-points, which were evenly distributed between 0 and 8 nm−1. The two-state ensemble refinement was conducted using umbrella sampling along the weight wopen of the open state (see above). Each umbrella window was simulated for 40 ns, where the first 2 ns were removed for equilibration. The posterior distributions of wopen and of the interdomain distance derived from these simulations are presented in Figs 2D, 3A and S1A Fig.
For comparison, a single state (instead of the ensemble of two states) was refined against each of the five curves Iexp,w(q), using five simulations of 10 ns each and removing the first 2 ns for equilibration. S1B Fig presents the posteriors of the interdomain distance dNC resulting from refining a single structure against SAXS curves that, in truth, represent heterogenous open/closed ensemble. Notably, the single-state refinements try to explain those SAXS curves with intermediate (partially open) structures.
For the refinement simulations of Hsp90, the simulations were coupled to the target SAXS curve at Nq = 30 q-points, which were evenly distributed between 0.1 and 3 nm−1. The q-range below 0.1 nm−1 was omitted because the experimental data exhibited some deviation from the ideal Guinier behaviour. For some umbrella windows, Hsp90 was required to carry out large-scale conformational transitions. To accelerate those transitions, each window was first simulated for 8 ns with a ten-fold increased experiment-derived energy Eexp. Subsequently, the simulation of each umbrella window was continued for another 20 ns using the statistically founded Eexp that leads to the correct posterior (eq 18). From those simulations, the first 2 ns were removed for further equilibration, and the remaining simulation time was used to compute the posterior. An example of the umbrella histograms along the weight coordinate is shown in S4 Fig. To further improve the sampling close to the maxima of the posteriors, the simulations of the umbrella window at the peak of p(wopen|D, K) plus two neighboring windows were prolonged for another 15 ns.
|
10.1371/journal.pbio.2005853 | Mice learn to avoid regret | Regret can be defined as the subjective experience of recognizing that one has made a mistake and that a better alternative could have been selected. The experience of regret is thought to carry negative utility. This typically takes two distinct forms: augmenting immediate postregret valuations to make up for losses, and augmenting long-term changes in decision-making strategies to avoid future instances of regret altogether. While the short-term changes in valuation have been studied in human psychology, economics, neuroscience, and even recently in nonhuman-primate and rodent neurophysiology, the latter long-term process has received far less attention, with no reports of regret avoidance in nonhuman decision-making paradigms. We trained 31 mice in a novel variant of the Restaurant Row economic decision-making task, in which mice make decisions of whether to spend time from a limited budget to achieve food rewards of varying costs (delays). Importantly, we tested mice longitudinally for 70 consecutive days, during which the task provided their only source of food. Thus, decision strategies were interdependent across both trials and days. We separated principal commitment decisions from secondary reevaluation decisions across space and time and found evidence for regret-like behaviors following change-of-mind decisions that corrected prior economically disadvantageous choices. Immediately following change-of-mind events, subsequent decisions appeared to make up for lost effort by altering willingness to wait, decision speed, and pellet consumption speed, consistent with past reports of regret in rodents. As mice were exposed to an increasingly reward-scarce environment, we found they adapted and refined distinct economic decision-making strategies over the course of weeks to maximize reinforcement rate. However, we also found that even without changes in reinforcement rate, mice transitioned from an early strategy rooted in foraging to a strategy rooted in deliberation and planning that prevented future regret-inducing change-of-mind episodes from occurring. These data suggest that mice are learning to avoid future regret, independent of and separate from reinforcement rate maximization.
| Regret describes a unique postdecision phenomenon in which losses are realized as a fault of one’s own actions. Regret is often hypothesized to have an inherent negative utility, and humans will often incur costs so as to avoid the risk of future regret. However, current models of nonhuman decision-making are based on reward maximization hypotheses. We recently found that rats express regret behaviorally and neurophysiologically on neuroeconomic foraging tasks; however, it remains unknown whether nonhuman animals will change strategies so as to avoid regret, even in the absence of changes in the achieved rate of reinforcement. Here, we provide the first evidence that mice change strategies to avoid future regret, independent of and separate from reinforcement rate maximization. Our data suggest mice accomplish this by shifting from a foraging decision-making strategy that produces change-of-mind decisions after investment mistakes to one rooted in deliberation that learns to plan ahead.
| Regretful experiences comprise those in which an individual recognizes a better decision could have been made in the past. Humans assert a strong desire to avoid feeling regret [1]. Regret can have an immediate impact on influencing subsequent valuations, but it can also motivate individuals to learn to avoid future regret-provoking scenarios altogether [2]. Recently, the experience of regret has been demonstrated in nonhuman animals, sharing principal neurophysiological and behavioral correlates of regret with humans [3–4]. However, it remains unclear if nonhuman animals are capable of learning from regret in order to avoid recurring episodes in the future.
Counterfactual reasoning, or considering what might have been, is a critical tenet of experiencing regret [5–6]. This entails reflecting on potentially better alternatives that could have been selected in place of a recent decision. Thus, owning a sense of choice responsibility and acknowledging error of one’s own agency is central to regret. Following the experience of regret, humans often report a change in mood and augment subsequent decisions in an attempt at self-justification or in efforts to make up for their losses [7–8]. These immediate effects of regret on behavior describe a phenomenon distinct from the notion that individuals will also learn to take longitudinal measures to avoid future scenarios that may induce regret.
Neuroeconomic decision-making tasks offer a controlled laboratory approach to operationalize and characterize decision-making processes comparable across species [9–12]. Recently, a study by Steiner and Redish reported the first evidence of regret in rodents tested on a spatial decision-making task (Restaurant Row) [4]. In this task, food-restricted rats were trained to spend a limited time budget earning food rewards of varying costs (delays) and demonstrated stable subjective valuation policies of willingness to wait contingent upon cued offer costs. In rare instances in which rats disadvantageously violated their decision policies and skipped low-cost offers only to discover worse offers on subsequent trials (e.g., made “economic mistakes”), they looked back at the previous reward site and displayed corrective decisions that made up for lost time. These behaviors coincided with neural representations of retrospective missed opportunities in the orbitofrontal cortex, consistent with human and nonhuman-primate reports of counterfactual “might-have-been” representations [2–4,8,13–15]. While these data demonstrate that rats are responsive to the immediate effects of regret, the regret instances were too sparse to determine whether rats also showed long-term consequences of these regret phenomena. Thus, it remains unknown if nonhuman animals are capable of learning from such regret-related experiences, leaving open the question of whether nonhuman animals adopt longitudinal changes in economic decision-making strategies that prevent future instances of regret from occurring in the first place.
In the present study (Fig 1), we trained food-restricted mice to traverse a square maze with 4 feeding sites (restaurants), each with unique spatial cues and providing a different flavor (Fig 1B). On entry into each restaurant, mice were informed of the delay that they would be required to wait to get the food from that restaurant. In this novel variant of the Restaurant Row task, each restaurant contained 2 distinct zones: an offer zone and a wait zone. Mice were informed of the delay on entry into the offer zone, but delay countdowns did not begin until mice moved into the wait zone. Thus, in the offer zone, mice could either enter the wait zone (to wait out the delay) or skip (to proceed on to the next restaurant). After making an initial enter decision, mice had the opportunity to make a secondary reevaluative decision to abandon the wait zone (quit) during delay countdowns (S1 Video). Just like rats, mice revealed preferences for different flavors that varied between animals but were stable across days, indicating subjective valuations for each flavor were used to guide motivated behaviors. Varying flavors, as opposed to varying pellet number, allowed us to manipulate reward value without introducing differences in feeding times between restaurants (as time is a limited commodity on this task). Costs were measured as different delays mice would have to wait to earn a food reward on that trial, detracting from their session’s limited 1 h time budget. Delays were randomly selected between a range of offers for each trial. Tones sounded upon restaurant entry whose pitch indicated offer cost and descended in pitch stepwise during countdowns once in the wait zone.
Taken together, in this task, mice must make serial judgements in a self-paced manner, weighing subjective valuations for different flavors against offer costs and balancing the economic utility of sustaining overall food intake against earning more rewards of a desirable flavor. In doing so, cognitive flexibility and self-control become critical components of decision-making valuation processes in this task, assessed in 2 separate stages of decision conflict (in the offer and wait zones). Importantly, because mice had 1 h to work for their sole source of food for the day, trials on this task were interdependent both within and across days. Therefore, this was an economic task in which time must be budgeted in order to become self-sufficient across days. Here, we tested mice for 70 consecutive d. Thus, the key to strategy development on this task is the learning that takes place across days, for instance, when performance on a given day produces poor yield. Monitoring longitudinal changes in decision-making strategy can provide novel insight into regret-related learning experiences.
How mice were trained on the Restaurant Row task allowed us to characterize the development of and changes in economic decision-making strategies. Mice progressed from a reward-rich to a reward-scarce environment in blocks of stages of training across days (Fig 1A). Each block was defined by the range of possible costs that could be encountered when offers were randomly selected on the start of each trial upon entry into each restaurant’s offer zone. The first block (green epoch) spanned 7 d in which all offers were always 1 s (Fig 1A). During this time, mice quickly learned the structure of the task (Fig 2), becoming self-sufficient and stabilizing the number of pellets earned (Fig 2A), reinforcement rate (Fig 2B), and number of laps run (Fig 2C). During this block, mice rapidly developed stable flavor preferences and learned to skip offers for less-preferred flavors and enter offers for more-preferred flavors, entering versus skipping at roughly equal rates overall while rarely quitting (Fig 2D and 2E, S1A–S1E Fig). The second block (yellow epoch) spanned 5 d in which offers could range between 1–5 s. The third block (orange epoch) spanned 5 d in which offers could range between 1–15 s. Lastly, the fourth and final block (red epoch, beginning on day 18) lasted until the end of this experiment (day 70), in which offers could range between 1–30 s. Note that because the mice had a limited 1 h time budget to get all of their food for the day, these changes in offer distributions produced increasingly reward-scarce environments that required more complex strategies to maximize rate of reward.
Upon transitioning to the 1−30 s offer block, mice suffered a large drop in total number of pellets earned (Fig 2A, repeated measures ANOVA, F = 9.46, p < 0.01) and reinforcement rate (increase in time between earnings, Fig 2B, F = 253.93, p < 0.0001). With this came a number of changes in decision-making behaviors that took place immediately, on an intermediate timescale, and on a delayed long-term timescale. Decreases in food intake and reinforcement rate were driven by an immediate significant increase in proportion of total offers entered (Fig 2D, F = 56.10, p < 0.0001) coupled with a significant increase in proportion of entered offers quit (Fig 2E, F = 472.88, p < 0.0001) as mice experienced long delays in the wait zone for the first time. This suggests that mice were apt to accept expensive offers in the offer zone even though they did not actually earn those offers in the wait zone (S2C, S2G, S2I and S2J Fig). This also suggests that choosing to enter versus skip in the offer zone and choosing to opt out of waiting in the wait zone may access separate valuation algorithms. We quantified this disparity in economic valuations by calculating separate “thresholds” of willingness to enter in the offer zone and willingness to wait in the wait zone as a function of offer cost. Following the 1−30 s transition, offer zone thresholds significantly increased (maxed out at approximately 30 s) and became significantly higher than wait zone thresholds (Fig 2F, offer zone change: F = 151.65, p < 0.0001; offer zone versus wait zone: F = 59.85, p < 0.0001). Furthermore, we found that these immediate behavioral changes were more robust in more-preferred restaurants, suggesting asymmetries in suboptimal decision-making strategies upon transition from a reward-rich to a reward-scarce environment were dependent on differences in subjective valuation algorithms (S1A Fig, see S1 Text).
Because performance on this task served as the only source of food for these mice, decision-making policies that might have been sufficient in reward-rich environments must change when they are no longer sufficient in reward-scarce environments. We found that mice demonstrated behavioral adaptations over the 2 wk following the transition to the 1−30 s offer range so that by approximately day 32, they had effectively restored overall food intake (Fig 2A, change across 2 wk: F = 355.21, p < 0.0001; post-2 wk compared to baseline: F = 0.80, p = 0.37) and reinforcement rates (Fig 2B, change across 2 wk: F = 183.68, p < 0.0001; post-2 wk compared to baseline: F = 0.24, p = 0.63) to baseline levels similar to what was observed in a reward-rich environment (Fig 2A and 2B). Note that the restored reinforcement rates renormalization, indicated by the pink epoch in Fig 2, was not imposed by the experimenters but was due to changes in the behavior of the mice under unchanged experimental rules (1−30 s offers). Mice accomplished this by running more laps to compensate for food loss (Fig 2C, F = 221.61, p < 0.0001) without altering economic decision-making policies. That is, we observed no changes in wait zone thresholds during this 2-wk period (Fig 2F, F = 2.57, p = 0.11). By entering the majority of offers indiscriminately with respect to cost (Fig 2D, proportion trials entered > 0.5: t = 31.22, p < 0.0001, S2C Fig), mice found themselves sampling more offers in the wait zone they were also unwilling to wait for, leading to an increase in quitting (Fig 2E, F = 55.37, p < 0.0001, S2G Fig).
Investing a greater portion of a limited time budget waiting for rewards that are ultimately abandoned appears, at face value, to be a wasteful decision-making strategy. Yet mice were able to restore food intake and reinforcement rates using this strategy. We characterized how mice allocated their limited time budget and quantified time spent among various separable behaviors that made up the total 1-h session (Fig 3). We first calculated the percent of total budget engaged in making offer zone decisions to skip versus enter, wait zone decisions to quit versus earn, postearn consumption behaviors, and travel time between restaurants (Fig 3A). We also calculated the average time spent engaged in a single bout of each decision process (Fig 3B–3F). The percent of total session time allocated to quit events (Fig 3A, F = 306.72, p < 0.0001), as well as average time spent waiting before quitting (Fig 3C, F = 44.21, p < 0.0001), significantly increased immediately following the transition to 1−30 s offers. Thus, time spent waiting in the wait zone before engaging in change-of-mind behaviors drove the immediate decrease in reinforcement rates and overall loss of food intake. Note that this waiting and then quitting behavior entails investing time that provided no reward. Over the subsequent 2 wk, time spent waiting before quitting significantly decreased as mice restored food intake and reinforcement rates (Fig 3C, F = 781.55, p < 0.0001). This suggests that mice learned to quit more efficiently in the wait zone. We calculated economic efficiency of wait zone quits (Fig 4B) by measuring how much time was remaining in the countdown at the moment of quitting relative to an individual’s wait zone threshold. Over these 2 wk, mice learned to quit in a more economically advantageous manner before excess time was invested. That is, mice learned to quit while the time remaining in the countdown was still above wait zone thresholds (Fig 4B, F = 64.00, p < 0.0001, S1P Fig, S3 Fig, see S1 Text), avoiding quitting at a timepoint when it would have been advantageous to otherwise finish waiting. This suggests that wait zone–quit reevaluations were corrective actions that opposed erroneous principal valuations in the offer zone. Interestingly, mice struggled to learn to quit efficiently in more preferred restaurants, reflecting a reluctance to apply adaptive opt-out foraging strategies in situations with high subjective valuation biases (S1K and S1P Fig see S1 Text). Despite increasing change-of-mind efficiency, because the frequency of quit events increased along this 2 wk time course, the fraction of the session budget allocated to quit events remained significantly elevated compared to baseline (Fig 3A, F = 105.90, p < 0.0001).
After mice successfully restored food intake and reinforcement rates by refining a foraging strategy, we found a distinct, delayed phase of additional learning that took place with prolonged training in the absence of any further changes in food intake (pink epoch, Fig 2A, F = 1.82, p = 0.18), reinforcement rates (pink epoch, Fig 2B, F = 0.01, p = 0.95), or laps run (pink epoch, Fig 2C, F = 1.54, p = 0.21). The proportion of enter-then-quit decisions decreased over the remainder of the experiment (Fig 2E, F = 159.30, p < 0.0001) as mice learned to reject offers in the offer zone that they were unwilling to remain committed to once in the wait zone (S2D–S2H Fig). This is reflected in a decrease in offer zone thresholds until they were in register with wait zone thresholds by the end of the experiment (pink epoch, Fig 2F, offer zone change: F = 812.40, p < 0.0001; offer zone versus wait zone at day 70: F = 0.17, p = 0.68). As a result, mice learned to skip more often in the offer zone (pink epoch, Fig 2D, F = 116.85, p < 0.0001). We calculated the economic efficiency of offer zone decisions by measuring the likelihood of skipping offers above wait zone thresholds relative to the likelihood of entering offers above wait zone threshold and found that offer zone decisions became more efficient during the pink epoch (Fig 4A, F = 474.94, p < 0.0001). As a result, the proportion of session budget allocated to quit events declined back to baseline levels (pink epoch, Fig 3A, budget quitting change: F = 1639.61, p < 0.0001, day 70 compared to baseline: F = 0.17, p = 0.68). The only change observed in average time spent per decision across decision processes during this phase of learning was in offer zone time, which increased over extended training as skip frequency increased (pink epoch, Fig 3B, offer zone time: F = 490.14, p < 0.0001; wait zone quit time: F = 0.10, p = 0.75; earn time: F = 0.11, p = 0.74; linger time: F = 0.73, p = 0.39; travel time: F = 0.01, p = 0.94).
Upon closer examination of offer zone behaviors (Fig 5), we found marked changes following the 1–30 s transition in skip decisions but not in enter decisions. We calculated the reaction time from offer onset until either a skip or enter decision was made. We also tracked each animal’s x and y location path trajectory as they passed through the offer zone. From this, we could capture the degree to which animals interrupted smooth offer zone passes with “pause and look” reorientation behaviors known as vicarious trial and error (VTE). VTE is a well-studied behavioral phenomenon that reveals ongoing deliberation and planning during moments of embodied indecision, supported by numerous electrophysiological experiments reporting concurrent neural representations of possible future outcomes compared serially [16–25]. The physical “hemming and hawing” characteristic of VTE is best measured by calculating changes in velocity vectors of discrete body x and y positions over time as dx and dy. From this, we can calculate the momentary change in angle, Phi, as dPhi. When this metric is integrated over the duration of the pass through the offer zone, VTE is measured as the absolute integrated angular velocity, or IdPhi, until either a skip or enter decision was made (Fig 5A and 5B, day 70 examples path traces). In a reward-rich environment, offer zone reaction time became more rapid (green-yellow-orange epochs, Fig 5C, F = 157.78, p < 0.0001), and paths measured by IdPhi became more stereotyped (green-yellow-orange epochs, Fig 5D, F = 150.19, p < 0.0001) as mice learned the structure of the task and made ballistic decisions. However, in a reward-scarce environment, skip reaction time (Fig 5C, F = 92.00, p < 0.0001) and skip VTE (Fig 5D, F = 117.80, p < 0.0001) began to increase following the transition to 1–30 s offers. These behaviors stabilized after food intake, and reinforcement rates were restored for the remainder of the experiment (pink epoch, Fig 5C, skip time: F = 2.21, p = 0.14; Fig 5D, skip VTE: F = 0.45, p = 0.50) as offer zone thresholds declined (Fig 2F) and skip frequency increased (Fig 2D). This suggests that mice enacted deliberative strategies in the offer zone after prolonged training. Mice learned to plan to skip expensive offers that previously would have been rapidly entered and then ultimately quit. Furthermore, following the transition to 1–30 s offers, enter decisions remained fast (Fig 5C, F = 1.73, p = 0.19) with low VTE (Fig 5D, F = 0.97, p = 0.32), suggesting enter decisions that ultimately led to quits were economically disadvantageous snap judgements in the offer zone that were subsequently reevaluated and corrected in the wait zone. Skip reaction time and VTE were higher in more preferred restaurants (S1G–S1J Fig), suggesting decisions to skip expensive offers for desired flavors were more difficult. Furthermore, refining the economic efficiency of this deliberative strategy was more difficult to learn in more-preferred restaurants (S1O Fig, S4 Fig, S5 Fig, see S1 Text).
This opens an intriguing question: if the changes that took place with prolonged training did not change the efficiency of food receipt, and if the only change after the development of deliberative strategies was a reversal of the increase in quit frequency, what does a reduction in change-of-mind decisions serve these animals? Given that there was no gain in food intake or reinforcement rate nor decrease in energy expenditure, what might be the driving force behind this delayed learning process?
A strength of the Restaurant Row task is its capability of measuring how economic decisions in one trial influence economic decisions in the following trial. This between-trial sequence feature of Restaurant Row captures post-decision-making phenomena, like regret [4]. A key factor in experiencing regret is the realization that a user-driven mistake has been made and that an alternative response could have led to a more ideal outcome. A change-of-mind quit decision in this novel variant of the Restaurant Row task thus presents an economic scenario in which mice take action to opt out of and abandon ongoing investments in the wait zone following an economically disadvantageous enter decision. As shown above, quits are economically advantageous reevaluations of prior snap judgements made in the offer zone. Thus, quit events reveal a potential economic scenario in which an agent’s decision has led to an economically disadvantageous option, whereby a counterfactual opportunity (“should have skipped it in the first place”) could provoke a regret-like experience.
Economic theories of human decision-making have hypothesized that regret adds a negative component to a utility function [1,7,26–28]. These theories suggest that an important driving force for human decision-making is the avoidance of future regret [2,8,29–31]. In order to test if decisions following enter-then-quit sequences carry added negative utility akin to regret previously demonstrated in Restaurant Row, we examined decision outcomes in the subsequent restaurant encounter following change-of-mind decisions compared to those following skip decisions (Fig 6). We compared enter-then-quit events to skip events (Fig 6A) that were matched for total time spent in the first restaurant before ultimately turning down the offer and advancing to the subsequent restaurant (Fig 6B). For example, we compared a skip decision that used up 2 s of offer zone time to an enter-then-quit sequence that used up a total of 2 s of combined offer zone and wait zone time. Consistent with previous reports in rats who attempted to make up for lost effort following regret, we found that, following quits, mice were more likely to accept offers in the next trial (Fig 6C, F = 39.26, p < 0.0001), did so quickly (Fig 6D, F = 163.28, p < 0.0001), and upon earning subsequent rewards, rapidly consumed food and exited the reward site (Fig 6E F = 191.89, p < 0.0001), compared to trials following skips. Quit-induced effects on subsequent trials existed across the entire experiment (Fig 6F–6H) and remained, even after controlling for flavor preferences (S6 Fig, see S1 Text). This suggests that enter-then-quit sequences were capable of augmenting subsequent valuations, even when change-of-mind reevaluations were matched to skip decisions for resource depletion and even during early stages of training amidst simpler foraging strategies before deliberative strategies developed.
Taken together, on a multiple-week timescale, mice transitioned from a foraging strategy that learned to become efficient (Fig 4B) to a distinct deliberative strategy that separately learned to become efficient later (Fig 4A). This change in strategy effectively traded enter-then-quit reevaluative decisions in the wait zone for skip decisions during principal valuations in the offer zone, with no overt benefit other than reducing the frequency of change-of-mind events. Quit events and skip events came from the same distribution of offer lengths (S7 Fig). Based on these data, it seems that not only can a change-of-mind experience have an immediate impact on subsequent valuations but it can also impact longer-term learning in mice capable of augmenting decision-making strategies. The resulting decision-making strategy appears to be one rooted in deliberation and planning as a means of avoiding future change-of-mind scenarios altogether.
Numerous studies have demonstrated that human individuals develop long-term strategies to avoid future instances of regret [2,7–8,14]. This phenomenon is distinct from the ability of regret to drive compensatory augmentations in valuation processes of immediately subsequent opportunities. While the immediate effects of regret have been demonstrated in rodents [4], long-term regret-avoidance learning, however, has not been previously observed. Here, we provide support not only for growing evidence that rodents (mice as well as rats) are capable of experiencing regret-like episodes but also that such experiences, separate from and independent of reinforcement maximization, can drive long-term changes in decision-making strategies.
Much of the animal learning literature has focused primarily on reinforcement maximization as the sole motivator of reward-related learning in decision-making paradigms [32–35]. That is, the goal of increasing reward reinforcement rate is thought to underlie animal behavior. Temporal difference error algorithms demonstrate a well-characterized mechanism of reward maximization–driven motivation in reinforcement learning theory [33–36]. Such learning algorithms, supported by neural representations of escalating response vigor and reward expectancies in mesolimbic dopamine systems, update behavioral policies or learn novel contingencies in order to optimize a given cost function and produce maximum reward yield [37–42]. Behavioral and neurophysiological data in both humans and nonhuman animals support a reward maximization theory of learning algorithms.
In the present study, we found evidence of reward-maximization learning algorithms as mice progressed from reward-rich to reward-scarce environments and made increasingly efficient wait zone decisions in a self-paced manner on a time-sensitive economic decision-making task during which they earned their only source of food. We also found distinct learning processes separated across space and time in the offer zone that took place on a much longer timescale. We found that mice reduced the frequency of wait zone change-of-mind decisions by learning to plan ahead in the offer zone, without any additional gain in reinforcement rates or reduction in energy expenditure. Other hypothesized drivers of human learning besides reinforcement maximization and energy expenditure minimization include managing affective states, particularly ameliorating or minimizing negative affect [43–44]. Avoiding pain, stress, threat, or anxiety is a well-studied motivator in human learning as well as in nonhuman-animal fear conditioning or punishment learning paradigms [45–46]. However, in a reward context, negative affect associated with regret and reward-related outcome experiences, while well-characterized in humans, is far less understood in animal learning models of positive reinforcement, reward-seeking learning.
The relatively straightforward view of reward maximization–driven reinforcement learning is challenged by the decision-making phenomena made tractable in these economic decision-making paradigms [33]. Postdecision regret is a well-known example that poses issues for traditional reinforcement learning algorithms dependent on updating stimuli or actions associated with actual experienced reward outcomes [33]. Hypothetical outcomes of forgone alternatives processed during counterfactual thinking that turn out to be better than chosen actions—key in regret—are indeed capable of driving long-term changes in future decision strategies through fictive learning, but it is a process that has been sparsely studied in nonhuman animals [3–7,13–15]. Mapping counterfactual outcomes onto corrective actions that could have been taken aids in the development of new decision strategies aimed to avoid regret in the future, yet this is a poorly understood behavioral and neural process.
Change-of-mind behaviors present unique decision-making scenarios that, when assessed on an economic task, can capture the economic advantageous versus disadvantageous nature of principal valuations and subsequent reevaluative choices. On this novel variant of the Restaurant Row task, we separate principal valuations (offer zone) from reevaluative choices (wait zone) across space and time within a single trial. Furthermore, change-of-mind behaviors present a powerful means of studying counterfactual decision processes [47–49]. In the context of the neuroeconomics of regret, a few questions arise: what drives individuals to change their minds? Which decisions might be economically fallible: the original choice, the delayed reconsideration, neither, or both? Why might individuals be reluctant to change their minds, how is this related to regret, and how might this interact with subjective valuation reinforcement learning algorithms?
Change-of-mind decisions occur every day in the real world, yet there is the general consensus that many individuals find this unpleasant and are often reluctant to do so, even when its utility is apparent [50–53]. Imagine the common scenario of a person in a food court during a 1h lunch break deciding which line to wait in—a direct analogue of what we test here in the Restaurant Row task. The decision to back out of waiting in any given line often comes with a sore feeling, even if doing so was an advantageous decision. Conversely, “going down with the ship” describes the sometimes-irrational motivation to refuse overturning a principal judgement and abandoning a partial investment. This is thought to be motivated by a desire to avoid being wasteful, admitting mistakes, or challenging one’s own beliefs. Thus, following an investment history, it is reasonable to appreciate that progress made toward a goal may be difficult to abandon, doing so may generate a source of cognitive dissonance, and thus, the decision to override a principal judgement when reevaluating continued investment errs on the side of perseveration, however economically irrational that may be. This describes a well-known decision-making phenomenon termed the sunk cost fallacy, in which the value of continued investment toward reward receipt is inflated as a function of irrecoverable past investments [54]. Mice, rats, and humans all demonstrate sensitivity to sunk costs in the wait zone when making quit decisions as a function of investment history on translated variants of the Restaurant Row task [55]. Thus, quit-induced regret and sunk cost–driven perseveration appear to be intimately related here. That is, after making a principal judgement in the offer zone to accept an offer at a cost higher than subjective value indicates one should (i.e., an initial economic violation of wait zone threshold), subjects are faced with a change-of-mind dilemma, torn between irrationally waiting out the expensive offer versus rationally backtracking and changing their plans, when affective contributions appear to weigh these options against one another.
In our food court example, the economically rational decision would be to select a line immediately and to make one’s decision while waiting in line. However, this is not what is typically observed—instead, it is far more common for people to deliberate before choosing and investing in any one option, despite the fact that this wastes time planning. Despite reevaluating an ongoing investment being the economically efficient and rational strategy, this hinges on a high frequency of change-of-mind decisions. After prolonged training in the Restaurant Row task, mice show a shift from the select-and-reevaluate foraging strategy to the deliberate-first strategy, even though it produces no change in reinforcement rate or energy expenditure. Thus, we conclude that mice are capable of learning from regret-related experiences induced by change-of-mind decisions and that they develop a forward-looking deliberative strategy that, although expensive in time and in computational resources, is economically advantageous because regret itself induces a negative utility. Rather than learning to deal with regret, sometimes mice take the time to plan ahead and learn to just avoid regret altogether.
31-C57BL/J6 male mice, 13 wk old, were trained in Restaurant Row. Mice were single-housed (beginning at 11 wk of age) in a temperature- and humidity-controlled environment with a 12 h light/12 h dark cycle with water ad libitum. Mice were food restricted to a maximum of 85% free-feeding body weight and trained to earn their entire day’s food ration during their 1 h Restaurant Row session. Experiments were approved by the University of Minnesota Institutional Animal Care and Use Committee (IACUC; protocol number 1412A-32172) and adhered to NIH guidelines. Mice were tested at the same time every day during their light phase in a dimly lit room, were weighed before and after every testing session, and were fed a small postsession ration in a separate waiting chamber on rare occasions to prevent extremely low weights according to IACUC standards (not <85% free-feeding weights). Previous studies using this task yielded reliable behavioral findings with minimal variability in at least sample sizes of n = 7.
Mice underwent 1 wk of pellet training prior to the start of being introduced to the Restaurant Row maze. During this period, mice were taken off of regular rodent chow and introduced to a single daily serving of BioServ full-nutrition 20 mg dustless precision pellets in excess (5 g). This serving consisted of a mixture of chocolate-, banana-, grape-, and plain-flavored pellets. Next, mice (hungry, before being fed their daily ration) were introduced to the Restaurant Row maze 1 d prior to the start of training and were allowed to roam freely for 15 min to explore, get comfortable with the maze, and familiarize themselves with the feeding sites. Restaurants were marked with unique spatial cues. Restaurant location remained fixed throughout the entire experiment. Feeding bowls in each restaurant were filled with excess food on this introduction day.
Task training was broken into 4 stages. Each daily session lasted for 1 h. At test start, one restaurant was randomly selected to be the starting restaurant where an offer was made if mice entered that restaurant’s T-shaped offer zone from the appropriate direction in a counterclockwise manner. During the first stage (days 1–7), mice were trained for 1 wk being given only 1 s offers. Brief low-pitch tones (4,000 Hz, 500 ms) sounded upon entry into the offer zone and repeated every second until mice skipped or until mice entered the wait zone, after which a pellet was dispensed. To discourage mice from leaving earned pellets uneaten, motorized feeding bowls cleared an uneaten pellet upon restaurant exit. Leftover pellets were counted after each session, and mice quickly learned to not leave the reward site without consuming earned pellets. The next restaurant in the counterclockwise sequence was always and only the next available restaurant where an offer could be made, such that mice learned to run laps encountering offers across all 4 restaurants in a fixed order serially in a single lap. During the second stage (day 8–12), mice were given offers that ranged from 1 s to 5 s (4,000 Hz to 5,548 Hz, in 387 Hz steps) for 5 d. Offers were pseudorandomly selected, such that all 5 offer lengths were encountered in 5 consecutive trials before being reshuffled, selected independently between restaurants. Again, offer tones repeated every second in the offer zone indefinitely until either a skip or enter decision was made. In this stage and subsequent stages, in the wait zone, 500 ms tones descended in pitch every second by 387 Hz steps, counting down to pellet delivery. If the wait zone was exited at any point during the countdown, the tone ceased, and the trial ended, forcing mice to proceed to the next restaurant. Stage 3 (days 13–17) consisted of offers from 1 s to 15 s (4,000–9,418 Hz) for another 5 d. Stage 4 (days 18–70) offers ranged from 1 s to 30 s (4,000–15,223 Hz) and lasted until mice showed stable economic behaviors. We used 4 Audiotek tweeters positioned next to each restaurant, powered by Lepy amplifiers, to play local tones at 70 dB in each restaurant. We recorded speaker quality to verify frequency playback fidelity. We used Med Associates 20 mg feeder pellet dispensers and 3D-printed feeding bowl receptacles fashioned with mini-servos to control automated clearance of uneaten pellets. Animal tracking, task programming, and maze operation were powered by AnyMaze (Stoelting). Mice were tested at the same time every day in a dimly lit room, were weighed before and after every testing session, and were fed a small postsession ration in a separate waiting chamber on rare occasions as needed to prevent extremely low weights according to IACUC standards (not <85% free-feeding weights).
All data were processed in Matlab, and statistical analyses were carried out using JMP Pro 13 Statistical Discovery software package from SAS. All data are expressed as mean +/− 1 SE. Sample size is included in each figure. No data were lost to outliers. Offer zone thresholds were calculated by fitting a sigmoid function to offer zone choice outcome (skip versus enter) as a function offer length for all trials in a single restaurant for a single session and measuring the inflection point. Wait zone thresholds were calculated by fitting a sigmoid function to wait zone choice outcomes (quit versus earn) as a function of offer length for all entered trials in a single restaurant for a single session. For dynamic analyses that depend on thresholds (e.g., Fig 4), analyses at each timepoint used that timepoint’s threshold information. Statistical significance was assessed using Student t tests, one-way, two-way, and repeated-measures ANOVAs, using mouse as a random effect in a mixed model, with post-hoc Tukey t tests correcting for multiple comparisons. Significance testing of immediate changes at block transitions were tested using a repeated-measures ANOVA between 1 d pre- and 1 d posttransition. These are indicated by significance annotations below the x-axis on relevant figures. Significance testing of gradual changes within block were tested using a repeated-measures ANOVA across all days within a given block or epoch. These are indicated by significance annotations within the plot either directly above or below the data centered within the epoch of interest. If significant interactions between factors were found (e.g., x rank), these are reflected by multiple significance annotations either below the x-axis or within the plot, respectively. The period of renormalization was estimated based on animal self-driven performance improvements in the 1–30 s block and not imposed on the animals by experimenters nor the protocol design. Renormalization was characterized by identifying the number of days in the 1–30 s block, after which total pellet earnings and reinforcement rate reliably stabilized (within a sliding 5 d window) and was no different from performance in relatively reward-rich environments collapsing across the first 3 training blocks. This was estimated to be approximately by day 30 of the experiment.
|
10.1371/journal.pmed.1002250 | Early diagnosis of mild cognitive impairment and mild dementia through basic and instrumental activities of daily living: Development of a new evaluation tool | Assessment of activities of daily living (ADL) is paramount to determine impairment in everyday functioning and to ensure accurate early diagnosis of neurocognitive disorders. Unfortunately, most common ADL tools are limited in their use in a diagnostic process. This study developed a new evaluation by adopting the items of the Katz Index (basic [b-] ADL) and Lawton Scale (instrumental [i-] ADL), defining them with the terminology of the International Classification of Human Functioning, Disability and Health (ICF), adding the scoring system of the ICF, and adding the possibility to identify underlying causes of limitations in ADL.
The construct validity, interrater reliability, and discriminative validity of this new evaluation were determined. From 2015 until 2016, older persons (65–93 y) with normal cognitive ageing (healthy comparison [HC]) (n = 79), mild cognitive impairment (MCI) (n = 73), and Alzheimer disease (AD) (n = 71) underwent a diagnostic procedure for neurocognitive disorders at the geriatric day hospital of the Universitair Ziekenhuis Brussel (Brussels, Belgium). Additionally, the ICF-based evaluation for b- and i-ADL was carried out. A global disability index (DI), a cognitive DI (CDI), and a physical DI (PDI) were calculated. The i-ADL-CDI showed high accuracy and higher discriminative power than the Lawton Scale in differentiating HC and MCI (area under the curve [AUC] = 0.895, 95% CI .840–.950, p = .002), MCI and AD (AUC = 0.805, 95% CI .805–.734, p = .010), and HC and AD (AUC = 0.990, 95% CI .978–1.000, p < .001). The b-ADL-DI showed significantly better discriminative accuracy than the Katz Index in differentiating HC and AD (AUC = 0.828, 95% CI .759–.897, p = .039). This study was conducted in a clinically relevant sample. However, heterogeneity between HC, MCI, and AD and the use of different methods of reporting ADL might limit this study.
This evaluation of b- and i-ADL can contribute to the diagnostic differentiation between cognitively healthy ageing and neurocognitive disorders in older age. This evaluation provides more clarity and nuance in assessing everyday functioning by using an ICF-based terminology and scoring system. Also, the possibility to take underlying causes of limitations into account seems to be valuable since it is crucial to determine the extent to which cognitive decline is responsible for functional impairment in diagnosing neurocognitive disorders. Though further prospective validation is still required, the i-ADL-CDI might be useful in clinical practice since it identifies impairment in i-ADL exclusively because of cognitive limitations.
| Mild cognitive impairment (MCI) is seen as a transitional zone between normal aging and dementia.
Assessment of activities of daily living (ADL) is paramount to underpin accurate diagnostic classification in MCI and dementia.
Unfortunately, most common report-based ADL tools have limitations for diagnostic purposes.
We set out to develop and validate a new tool to evaluate basic (b-; activities including personal hygiene, dressing, and eating) and instrumental (i-; cooking, shopping, and managing medication) ADL in an older population with cognitive disorders.
We developed a tool based on the framework of the International Classification of Functioning, Disability and Health (ICF).
A global disability index (DI), a Cognitive DI (CDI) (a disability index taking into account solely activities impaired because of cognitive reasons), and a Physical DI (PDI) (a disability index taking into account solely activities impaired because of physical reasons) were calculated for both b- and i-ADL, based on the number of activities performed and the severity and causes of the functional problem.
223 community-dwelling older persons diagnosed as (1) cognitively healthy participants (n = 79), (2) patients with MCI (n = 73), or (3) mild to moderate dementia (n = 71) were included and underwent the new ICF-based evaluation of b- and i-ADL.
The i-ADL-CDI showed high accuracy and discriminative power in differentiating healthy comparisons (HC), MCI, and AD, and the b-ADL-DI showed high discriminative accuracy in differentiating HC and AD.
The new ICF-based evaluation of b- and i-ADL showed a good ability to distinguish normal and pathological cognitive aging and might enable the diagnosis of MCI and mild dementia.
The discriminative power of the ICF-based evaluation of b- and i-ADL for underlying causes of limitations is an advantage.
Since an evaluation of ADL is experienced as less invasive by older persons, this evaluation offers directions for clinical use in diagnosing cognitive disorders and might offer possibilities for clinical treatment, rehabilitation, advising, and coaching.
This study was conducted in a clinically relevant sample at a geriatric day hospital. However, heterogeneity in the participants might limit this study.
| Health services are dealing with an increasing number of older patients. Although most seniors are in reasonably good health and living an active life, a considerable number of them are at risk of developing major chronic conditions and mental disorders such as dementia. Worldwide, it is estimated that dementia affects 46.8 million persons, which causes great stress to medical, social, and informal care [1,2]. Several interventions have already proven efficient in reducing caregiver strain, psychological morbidity, and delaying or avoiding admissions in residential care. Since such interventions may be more effective early in the disease course, early diagnosis of dementia is pivotal [3,4]. In this regard, the concept of mild cognitive impairment (MCI) is interesting since it is seen as a transitional zone between normal aging and dementia. However, MCI is a heterogeneous concept in its clinical presentation and its progression to dementia; mainly, amnestic MCI (a-MCI) has high risk of dementia, but some persons remain stable or even revert to normal cognition [5–8]. Boundaries between normal aging, MCI, and mild dementia are vague, and discussion about the MCI criteria and their operationalization is ongoing [6,9]. The differentiation between mild and major neurocognitive disorders (NCD)—referring to the new version of the Diagnostic Statistical Manual of Mental Disorders (DSM-5) [10]—may be a step in a good direction since this entails a stronger emphasis on “independence in activities of daily living (ADL)” [11–14]. The distinction between mild and major NCD is determined by the extent to which cognitive decline interferes with everyday functioning [12,15]. In major NCD or dementia, cognitive impairment influences independence in everyday functioning in a negative way. In mild NCD or MCI, individuals remain autonomous [15,16], although subtle problems may already occur in complex activities [12,17–21]. The process of functional decline shows a typical and distinctive progression [22,23]. Instrumental ADL (i-ADL) such as cooking, shopping, and managing medication will become slightly limited in mild NCD and will require support in major NCD [18,23–26]. Basic ADL (b-ADL), which includes personal hygiene, dressing, and eating, remain stable the longest [27]. Only in major NCD does one need the support of others in performing b-ADL [23,28,29]. Consequently, assessment of ADL is paramount to determine the degree of impairment in everyday functioning and to underpin accurate diagnostic classification in NCD [9,12,15,30]. Besides, ADL disability might increase the risk for incident dementia. In that way, an evaluation of ADL might be useful not just as diagnostic tool but also as an indicator of the risk for future dementia [12,30].
In clinical practice, information about ADL is most commonly ascertained by asking a patient or his or her caregiver to report about the everyday functioning [31]. Report-based ADL scales are usually quick and easy to administer [9,32,33]. Unfortunately, most common report-based ADL tools have limitations for diagnostic purposes. Firstly, they often use scoring systems solely assessing the success or failure of completing a task [17,34]. They do not reflect the process of performing the task, although this could be meaningful for diagnostic purposes, particularly in mild cognitive disorders [35–39]. Secondly, evaluations of ADL often entail gender-dependent tasks, tasks that a person does not perform, or tasks that become subject to family support, which commonly comes into play among an older population. Clear-cut guidelines on how to deal with tasks that a person does not perform are lacking [40]. Thirdly, ADL evaluations have a poor sensitivity to detect mild functional impairments and are mostly unresponsive to detect changes in a person’s ability level [41–44]. The discriminative power of existing tools is insufficient, and psychometric properties are either unavailable or do not meet standards of quality [9,43,45]. Finally, assessment tools do not differentiate between underlying causes of limitations [46]. Nevertheless, in diagnosing NCD in a geriatric population, it is crucial to determine the extent to which cognitive decline is responsible for functional impairment, since comorbidities, physical limitations, or other noncognitive causes of decline in ADL are often seen in old age [12,47].
Over the years, multiple report-based ADL scales have been developed in order to contribute to the early diagnosis of NCD [31]. Tools such as the Functional Activities Questionnaire (FAQ) [48], the Everyday Technology Use Questionnaire (ETUQ) [49], and the Everyday Cognition (ECog) [50] have already targeted some shortcomings of current evaluations by including “new” items specific to everyday technologies and using scoring systems that only take activities into account that are relevant to an individual. These evaluations showed promising results in assessing individuals with NCD [31]. However, they do not solely assess performance in b- or i-ADL but rather evaluate a mixed spectrum of self-care, household, and other activities or assess everyday abilities such as memory, language, or divided attention.
To address the concerns of report-based ADL scales, performance-based scales such as the Assessment of Motor and Process Skills [51,52] and the Naturalistic Action Test [53,54] have been developed. These evaluations examine the process of task performance, detect changes in everyday functioning, and address causation in observable behaviors. However, these assessments also have limitations, such as being more time-consuming and needing a high degree of training of the assessors, which often limit its use in clinical practice [9,31]. Furthermore, most performance-based ADL scales are not freely available and are mostly not yet validated for use in MCI [55].
Currently, the most commonly used tools for assessing b- and i-ADL are respectively the Katz Index [22] and the Lawton Scale [56] [31,57,58]. Although in widespread use, both scales have shortcomings as mentioned above: they have poorly described psychometric properties, the scoring systems are not sensitive enough to detect subtle deficits, and they do not identify causes of limitations in ADL [9,43,58–61]. Many studies have attempted to improve the potential use of the Katz Index and Lawton Scale, including using item response theory methods [34], providing short versions of these scales [62], or by combining both scales in new evaluations [63,64]. However, these improvements could not overcome all mentioned shortcomings.
Therefore, this study set out to develop a new tool to evaluate b- and i-ADL for diagnostic purposes in a geriatric population with NCD. This evaluation is based on the International Classification of Functioning, Disability and Health (ICF) developed by the World Health Organization (WHO) [65]. The ICF provides a framework for describing everyday functioning and advances the understanding and measurement of disability [58]. It is increasingly being applied in clinical practice and research and has gained acceptance as the worldwide framework of assessing human functioning [66,67]. The new evaluation adopted the activities of the Katz Index and Lawton Scale—since they are considered sound as items for describing functioning in b- and i-ADL [68]—and were defined with the ICF terminology. Besides, the new evaluation took over the scoring system of the ICF and added the possibility to determine underlying causes of limitations. This evaluation might be useful in clinical and research settings to evaluate everyday functioning in NCD since it has an advantage over currently used report-based scales by applying the ICF terminology and scoring system. This offers a more standardized evaluation of ADL, which might benefit a more reliable and accurate diagnosis and treatment of NCD. In this study, the construct validity, interrater reliability, and discriminative validity of this new evaluation were determined. We hypothesized that the ICF-based evaluation of b- and i-ADL will have a good construct validity and interrater reliability and will be able to discriminate between cognitively healthy comparisons (HC), MCI, and AD.
The study protocol was based on the STARD criteria, developed to improve the completeness and transparency of reporting of studies of diagnostic accuracy [69].
The Ethical Committee of the Universitair Ziekenhuis Brussel approved this study (B.U.N. 143201523678). All data were collected in accordance with the ICH-GCP guidance and the declaration of Helsinki. All participants and informants gave written informed consents.
Three groups of community-dwelling older persons (≥65 y) were recruited consecutively through the geriatric day hospital of an academic teaching hospital (UZ Brussel, Belgium): (1) HC, (2) patients with MCI, and (3) with Alzheimer disease (AD). Patients with MCI and AD underwent a procedure for the diagnosis of cognitive disorders that was performed by a multidisciplinary team and is considered as good clinical practice [70]. This procedure consisted of a physical and neurological examination, clinical history taking, and neuropsychological assessment using the Mini-Mental State Examination (MMSE) [71]; Cambrigde Examination for mental disorders of the elderly, cognitive part (CamCog) [72]; memory subscale of the Alzheimer’s Disease Assessment Scale [73]; Visual Association Test (VAT) [74]; Memory Impairment Screen (MIS) [75]; Trail Making Test, parts A and B [76,77]; Frontal Assessment Battery [78]; and Geriatric Depression Scale (GDS-15) [79]. The procedure was completed by an evaluation of ADL using the Katz Index and Lawton Scale, extensive laboratory blood testing, and imaging of the brain by CT or MRI scan. HC were recruited separately from the diagnostic process for MCI and AD. They represent a heterogeneous sample of community-dwelling volunteers and geriatric patients who visited the geriatric day hospital for the diagnosis or treatment of conditions other than cognitive disorders (e.g., osteoporosis). HC were evaluated by the researchers using the same neuropsychological assessment and evaluation of ADL as MCI and AD. For all groups, the number of comorbidities and medication use was inventoried. The number of comorbidities was determined by counting the active diseases listed in the medical records at the moment the participants visited the geriatric day hospital, whether they were being treated pharmaceutically or not. All active diseases were counted in HC or as co-occurring with MCI or AD.
This evaluation has been designed as a semistructured interview that takes 10 min to complete. For the HC, self-report was used. For MCI and AD, proxy report was conducted.
According to the linking rules of Cieza et al. (2005) [83], the content of each item included in the Katz Index [22] and the Lawton Scale [56,84] was linked to one or more definitions of the activities component of the ICF (Tables 1 and 2).
First, the participant or proxy is asked whether an activity was performed during the past years. It is expected that the interviewer uses the ICF definitions to clarify the content of an activity. Each activity is rated for its relevance, which means that it is currently performed or it was previously performed by the individual. If activities have not been carried out during the past years because they were not relevant for an individual, they are not taken into account. This is mainly important for i-ADL, since these activities may never have been performed before (e.g., gender relevant) and are consequently irrelevant for the individual. For b-ADL, all items are relevant for every individual since—according to the definition of Reuben [85]—these activities are necessary to survive. The sum of relevant activities leads to the Total Number of relevant Activities (TNA). There is no cutoff of how many items are allowed to be not relevant.
The participant or proxy is asked how the activities are currently being performed. Based on the description, the investigator assigns a score. The scoring system adopted the performance qualifiers of the ICF, consisting of a five-point scale ranging from 0 (no difficulty to perform) to 4 (complete difficulty or unable to perform) (Table 3). Each score describes how an activity is performed (ICFScoreAct). The qualifiers were operationalized based on the experience of this research team with the development of the advanced ADL tool (a-ADL tool) [35,86] and on a previous qualitative study [87]. The sum of activities with a limitation (score ≥ 1) leads to the total number of Limited Activities (LimAct).
If a score of 1 or higher is assigned, the interviewer determines the underlying cause of limitation by asking the participant what causes the limitations. The interviewer probes with the following questions: “Why do you/does (s)he performs this activity differently?” or “What causes the need for help to perform this activity?” In this way, the interviewer interprets the story of the participant and can distinguish cognitive reasons (e.g., global mental functions, memory, attention, etc.), physical reasons (e.g., sensorial functions, mobility, stability, etc.), intrapersonal reasons (e.g., switch in field of interest), social reasons (e.g., loss of partner), and environmental reasons (e.g., car sold, moving to a new place, etc.) of limitations. The assignment of a reason is dichotomous: “yes” when a reason is present and “no” when a reason is absent. It is possible to assign more than one reason of limitation. To clarify how ICF scores can be derived and physical or cognitive causes of limitations can be assigned, some examples are illustrated in Table 4.
A “global disability index” (DI) can be calculated for b-ADL (b-ADL-DI) and i-ADL (i-ADL-DI) by taking into account a maximal disability (TNA multiplied by ICFScoreAct 4, which is equal to complete difficulty) and an absolute disability (LimAct multiplied by the severity of each limitation [ICFScoreAct]) (see Fig 1). Furthermore, for each reason of limitation, an index can be calculated. In this study, a “cognitive disability index” (CDI) and a “physical disability index” (PDI) for both b-ADL (b-ADL-CDI and b-ADL-PDI) and i-ADL (i-ADL-CDI and i-ADL-PDI) is computed, considering exclusively activities that are limited because of respectively cognitive and physical limitations (see Fig 1). When limitations are caused by multiple reasons (e.g., using transportation is limited to both physical and cognitive reasons), reasons can be assigned in both indices (e.g., i-ADL-CDI and i-ADL-PDI). All indices are expressed as percentages, with higher scores representing more disability.
Example (for b-ADL): A person previously performed 6 b-ADL (TNA = 6). ICFScore 0 is assigned to four activities, ICFScore 1 is assigned to one activity because of cognitive problems, and ICFScore 3 is assigned to one activity because of physical factors. This person has two limited activities, and the maximal disability is 24 (TNA*4). His b-ADL-DI is 16.6% (all limited activities are taken into account), the b-ADL-CDI is 4.2% (only the activities limited because of cognitive reasons are taken into account), and the b-ADL-PDI is 12.4% (only the activities limited because of physical reasons are taken into account).
From the diagnostic procedure, data regarding the MMSE, Katz Index, Lawton Scale, number of comorbidities, and medication use were extracted for this study. There were no missing data. Statistical analyses were performed with IBM SPSS for Mac (version 22.0) (SPSS Inc, Illinois, United States) with an α-level set two sided at p < 0.05 for all analysis. Demographic and clinical characteristics (i.e., age, education, gender, number of used medications, and number of comorbidities) and the MMSE, Katz Index, Lawton Scale were evaluated between groups by one-way ANOVA with Bonferonni post hoc tests or chi-square analysis. The construct validity was checked, in absence of a true golden standard, by determining the new evaluation’s ability to distinguish between HC, MCI, and AD. We hypothesised HC would show less disability than MCI and the latter less than AD, and that the CDI for both b- and i-ADL would differ more than DI and PDI. The indices for b- and i-ADL were compared across the groups using analysis of covariance (ANCOVA) in which age, number of used medications, number of comorbidities, level of education, and gender were included in the model as covariates. Secondly, in checking the construct validity, we calculated correlations between the indices and the MMSE. We hypothesised that (1) the CDI for both b- and i-ADL would show higher correlations with the MMSE than the DI and PDI since the CDI expresses solely deficits caused by cognitive disorders and (2) the i-ADL-CDI would show higher correlations than b-ADL-CDI because performing i-ADL is more vulnerable to cognitive disorders. Correlation analyses were performed using Pearson correlation between the MMSE, Katz Index, Lawton Scale, and the indices. To interpret the correlations, the guideline by Evans (1996) [88] was used: .00–.19 very weak, .20–.39 weak, .40–.59 moderate, .60–.79 strong, and .80–1.0 excellent. The interrater reliability was checked by comparing simultaneous observation of the interview by two independent raters in a sample of 25 participants and was evaluated by computing intraclass correlation coefficients (ICC) in a two-way mixed model and 95% confidence intervals (CI). Lastly, discriminative validity was evaluated by calculating receiver-operating-characteristics (ROC) curves and the AUC with cutoffs, sensitivity, and specificity for the new evaluation of b-ADL and i-ADL and for the Katz Index and Lawton Scale. The ROC curves and AUC for the new evaluation were compared with the Katz Index and Lawton Scale to determine the added diagnostic value of the new evaluation by using the method of DeLong et al. (1988) [89] in MedCalc (version 14.8.1.0) (MedCalc Software, Mariakerke, Belgium).
All participants reported to have enjoyed the assessment. The interviews lasted between 8 and 15 min for both b- and i-ADL; no adverse events occurred during the diagnostic procedure or the ICF-based evaluation of b- and i-ADL. Table 5 shows the demographic and clinical characteristics of the participants. In comparison to HC, patient groups had less years of education (F(2,220) = 13.7, p < .001) and reported more comorbidities (F(2,220) = 20.1, p < .001) and use of medications (F(2,220) = 15.3, p < .001). Between MCI and AD, no significant differences were found for age, education, medication use, and comorbidities. For MCI and AD, data about their everyday functioning were obtained by spouses (40.3%), children (50.0%), or close friends (9.7%). Almost half of them (48.6%) lived together with the person with MCI or AD. No significant differences between MCI and AD were found for relationship of the proxy (χ2(3) = 4.64, p < .199) and whether or not living together with the proxy (χ2(1) = 1.37, p < .241).
The interrater reliability (n = 25) was excellent for b-ADL-DI (ICC = .965, 95% CI [.920–.984]); b-ADL-CDI (ICC = .943, 95% CI [.872–.975]); b-ADL-PDI (ICC = .934, 95% CI [.850–.971]); i-ADL-DI (ICC = .986, 95% CI [.968–.994]); i-ADL-CDI (ICC = .986, 95% CI [.969–.994]); and i-ADL-PDI (ICC = .972, 95% CI [.973–.988]) (all p < .001). No significant differences between raters were observed.
Table 7 present the results of the ROC curves for the Katz Index, Lawton Scale, and the indices of the new evaluation.
This study developed and validated an evaluation of everyday functioning in b- and i-ADL by (1) adopting the activities of the Katz Index and Lawton Scale and linking them to the definitions and codes of the ICF, (2) by developing a scoring system based on the performance qualifiers of the ICF, and (3) by adding the possibility to take causes of limitations in performance into account. This new evaluation takes the person as his or her own reference. By doing so, it is possible to compute a set of indices. This study determined the construct validity, discriminative validity, and interrater reliability of this new evaluation in a geriatric population.
The new evaluation showed more accuracy in evaluating b- and i-ADL compared to the Katz Index and Lawton scale and subsequently has the potential to improve diagnostic differentiation between HC, mild NCD (e.g., MCI), and major NCD (e.g., AD). As hypothesised, this evaluation followed the hierarchical continuum of functional decline [19]; b-ADL-DI and i-ADL-DI showed significantly less disability in HC than in MCI and the latter less than in AD. The i-ADL-DI showed more disability than b-ADL-DI and had a significantly better accuracy than the Lawton Scale to differentiate HC from MCI and AD. The b-ADL-DI, in its turn, had a significantly better accuracy than the Katz Index in differentiating HC from AD. Other promising results were seen in the i-ADL-CDI. Although the original Lawton Scale cannot detect mildly affected quality of performance in i-ADL [90], the i-ADL-CDI could detect subtle functional deficits of persons with MCI and AD. The i-ADL-CDI demonstrated a significantly better accuracy than the Lawton Scale and is able to distuinguish between HC, MCI, and AD. This illustrates that it is important to make a distinction in causes, especially in older patients in whom physical limitations are commonly seen and also affect everyday functioning. When considering the diagnosis of NCD, it is of utmost importance to determine to what extent functional limitation is due to cognitive limitations and not due to other causes [17]. The b-ADL-CDI showed, as hypothesized, significantly more severe deficits in AD than in HC and MCI but had no better accuracy than the Katz Index. This can be explained by the fact that performing b-ADL is less vulnerable for cognitive decline and is often largely spared until later stages of the disease (i.e., moderate or severe dementia) [91]. If limitations are observed in MCI or mild AD, it will rather be caused by other reasons such as physical limitations, as illustrated in the b-ADL-PDI. The b-ADL-PDI showed significantly more severe limitations in AD than in HC and had similar accuracy to the Katz Index. Although the differentiation between normal cognition and mild AD is usually not much of a diagnostic dilemma in clinical context, the results of this study clearly state that b-ADL distinguish well when reasons for limitations are taken into account.
Until now, self- and informant-report scales did not show sound psychometrical properties [9,43] and were not considered as the best methods to evaluate everyday functioning since they might over- and underestimate functional ability [32,92]. There is growing evidence that performance-based evaluations might have more advantages over other assessment approaches [9,32,33]. However, only few of them are developed to assess MCI or mild dementia [55,93], and they are mostly too time- and cost-consuming to be administered [9,31]. Two recent performance-based instruments, the Erlangen Test of Activities of Daily Living in Mild Dementia or Mild Cognitive Impairment (ETAM) [55] and the Sydney Test of Activities of Daily Living in Memory Disorders (STAM) [94] have been developed with the aim to assess everyday activities in a time-efficient and reliable way for persons with MCI or mild dementia. Both evaluations show good psychometric characteristics, are easy to administer, and seem to be valuable in clinical practice and research. However, our i-ADL-DI and i-ADL-CDI show similar accurate validity to discriminate between HC, MCI, and AD. So, although this study developed a report-based measure—which may not be as accurate in detecting functional difficulties in persons with mild cognitive decline—the results of this study indicate that the ICF-based evaluation of b- and i-ADL might compete with the recently developed performance-based tools as the standard for classifying functional status and decline [32,33,55,94]. Ongoing research will clarify this and is already assessing the convergent validity between the ICF-based evaluation of b- and i-ADL and a performance-based measure.
Although many studies have already tried to improve the use of the Katz Index and the Lawton Scale, not all improvements were relevant for the diagnosis of cognitive disorders. In this study, we attempted to achieve more clarity, transparency, and nuance by maintaining the activities of the Katz Index and the Lawton Scale but by adopting the terminology and the scoring system of the ICF. A first advantage is that each activity is clearly defined by definitions according to the ICF. In contrast to the content of the original Katz Index and Lawton Scale—which varies depending on setting and circumstances [58,95]—the ICF definitions provide clear descriptions of the content of activities. In this way, no more doubt can arise about the exact content of activities such as, e.g., doing laundry (should ironing also be considered?) or using transportation (should driving a car also be considered?). Furthermore, since the ICF definitions do not impose a manner of performing, this evaluation will remain useful for future generations and might also have advantages across cultures since b- and i-ADL will always be applicable. Secondly, this evaluation only considers activities that are relevant for a person. In contrast with other scales, activities that are gender-dependent or a person has never performed in his or her life are not be taken into account. In this way, each person is considered as his or her own reference and is compared to his or her own previous level of everyday functioning, as suggested by Ganguli (2013) [12]. This might also be considered as an advantage for use in other generations and cultures. Thirdly, by using the detailed ICF qualifiers—ranging from 0 to 4—this evaluation provides a more sensitive scoring system, as recommended by Jekel et al. (2015) [9]. This new evaluation makes it possible to calculate indices and showed an excellent interrater reliability in this study. Lastly, this evaluation has the advantage to discriminate between reasons of limitations. Although other tools such as the FAQ [48], ETUQ [49], and ECog [50] are also valuable instruments in assessing individuals with NCD, they do not make a distinction between reasons of functional decline.
Although the results of our study are promising and may imply a change in the evaluation of everyday functioning in clinical practice, some considerations need to be made. First, a measurement bias might have occurred by using different methods of reporting ADL in HC and patients with MCI and AD. Although a report-based method has the clinical advantages of being easy to obtain, minimally disturbing, and of low cost, proxy and patient-based measures can be biased by mood status, social desirability, diminished awareness, denial, and other cognitive deficits [96,97]. But since informant-reports are generally preferred to self-report in evaluating everyday functioning in clinical practice and research settings, a reliable proxy was questioned about the everyday functioning of participants with MCI and AD in this study [97]. This closely resembles clinical reality, in which health care professionals have to work with the information that is available. Nevertheless, we could not rule out that the informants were not mildly cognitively impaired themselves. However, 50.0% of the informants were children of the persons with MCI and AD and had an estimated age range of 45 to 65 y. Although it is known that children of persons with AD are at high risk of cognitive disorders as they age, it seems unlikely that this would have influenced the results at this point of time. For the HC, only self-report was used because prior research in cognitively healthy older persons suggested that self-report evaluations are generally accurate indicators of ADL for older persons who demonstrate insight into their functional abilities [96,98]. Additionally, a second reflection must be made about the participants in this study. The patients with AD and MCI represent a clinically relevant sample but were significantly older, had more comorbidities, and took more medications than the HC. This suggests that the patient groups were frailer and might have experienced more functional problems. However, not all medications and comorbidities would be expected to contribute equally to functional impairment. Furthermore, this study did not report any measures of current depressive symptoms. The presence of a major depression was ruled out prior to the diagnosis in MCI and AD. However, mild to moderate depressive symptoms are an important comorbidity of cognitive disorders and may have an impact on everyday activities [99–102].
As a result, the contrast between groups might be larger than would be expected in a clear clinical sample. However, in the statistical analysis, our data was controlled for possible confounders such as age, medication use, number of comorbidities, level of education, and gender. Lastly, another consideration is that HC were—apparently—cognitively healthy persons. Yet, it is still possible that mild cognitive problems were present in some of them. However, we used strict cutoffs of MMSE—which can be considered as a valuable instrument for cognitive screening—in order to rule out cognitive deficits.
Based on the results of this study, we argue that this evaluation can contribute to the diagnostic differentiation between cognitively healthy ageing, mild NCD (e.g., MCI), and major NCD (e.g., AD). Particularly, the i-ADL-CDI might be useful. Since it is likely that decline in everyday functioning occurs over time, and this change leads to a conversion from mild to major NCD, further research—a longitudinal prospective follow up study—should address the predictive validity of this evaluation as follow-up assessment [14].
In conclusion, this new ICF-based evaluation for b- and i-ADL addresses important issues in assessing everyday functioning by (1) providing an operationalization of the evaluated activities by ICF codes and definitions, (2) providing a detailed scoring system that is based on the ICF qualifiers, and (3) by making a differentiation in causes of limitations. With validation in longitudinal prospective cohorts, this evaluation might offer a useful addition to the common diagnostic process and be of added value in a multidisciplinary approach with established cognitive and mood measures and biomarkers.
|
10.1371/journal.pntd.0006181 | Household expenditure on leprosy outpatient services in the Indian health system: A comparative study | Leprosy is a major public health problem in many low and middle income countries, especially in India, and contributes considerably to the global burden of the disease. Leprosy and poverty are closely associated, and therefore the economic burden of leprosy is a concern. However, evidence on patient’s expenditure is scarce. In this study, we estimate the expenditure in primary care (outpatient) by leprosy households in two different public health settings.
We performed a cross-sectional study, comparing the Union Territory of Dadra and Nagar Haveli with the Umbergaon block of Valsad, Gujrat, India. A household (HH) survey was conducted between May and October, 2016. We calculated direct and indirect expenditure by zero inflated negative binomial and negative binomial regression. The sampled households were comparable on socioeconomic indicators. The mean direct expenditure was USD 6.5 (95% CI: 2.4–17.9) in Dadra and Nagar Haveli and USD 5.4 (95% CI: 3.8–7.9) per visit in Umbergaon. The mean indirect expenditure was USD 8.7 (95% CI: 7.2–10.6) in Dadra and Nagar Haveli and USD 12.4 (95% CI: 7.0–21.9) in Umbergaon. The age of the leprosy patients and type of health facilities were the major predictors of total expenditure on leprosy primary care. The higher the age, the higher the expenditure at both sites. The private facilities are more expensive than the government facilities at both sites. If the public health system is enhanced, government facilities are the first preference for patients.
An enhanced public health system reduces the patient’s expenditure and improves the health seeking behaviour. We recommend investing in health system strengthening to reduce the economic burden of leprosy.
| Leprosy leads to low quality of life even after cure. The anaesthetic hands and feet leading to ulcers and deformities, stigma and poor mental health are just a few challenges. After declaration of leprosy elimination at global level, the research activities reduced significantly, and the health economics aspect was not an exception. The knowledge on economic burden of a disease helps in prioritization, policy making and advocacy. Our study is a step towards quantifying the economic burden of leprosy. Currently the aim is to eliminate leprosy at national level, therefore the countries need more information to plan high impact activities. Moreover, the patient’s perspective is important as they are the end-point recipients. Our study explores the patient’s financial burden due to leprosy (outpatient services), which is a significant indicator of a public health program’s success. If invested properly, the public health system has potential to reduce the economic burden of public health diseases. Our study is an attempt to link the patient’s perspective with the health system performance. This will help to encourage health systems strengthening.
| Leprosy is caused by Mycobacterium leprae, affecting the peripheral skin, nerve and nasal mucosa [1]. The adverse impact of leprosy on human lives is serious due to nerve function impairment and disabilities. Moreover, the early manifestation of disability in the form of sensory loss of hands or feet, often fails to seize attention of clinicians and patients, resulting into detection delay and further transmission of M. leprae [2, 3]. Therefore, the annual new case detection rate (NCDR) of leprosy is stagnant since many years [4]. The expectation to permanently eradicate leprosy, also referred as zero transmission [5] is now reflected into new WHO targets i.e. zero grade 2 disabilities among children, and new cases with grade 2 disability <1 case/million population [6]. However, the targets are difficult to achieve in the near future [7, 8], which means that leprosy will keep on imposing burden in many endemic countries.
Leprosy and poor socioeconomic status are in a vicious cycle, characterized by inequality [9–11], poor education [12], poverty [13, 14], stigma, etc. [15, 16]. A broad spectrum of evidence confirms the strength of the relationship between leprosy and poverty [17–21]. Evidence from Bangladesh shows that leprosy affected households have a poor nutritional level due to lower food expenditure per capita and household food stocks. This in fact increases the risk of acquiring leprosy in healthy household members [22]. Another study revealed that “people affected by leprosy are less likely to be stigmatized because of leprosy impairments than for their incapacity to contribute to family/community finances” [23]. Furthermore, leprosy incidence is high in the productive age group, resulting in long term financial loss [17]. Therefore, we suspect that the economic burden of leprosy is higher than perceived so far.
Household expenditure represents the patient’s perspective and is critical in estimating the economic burden. It is now routinely done across diseases [24], revealing underlying expenditure like income loss, which can sometimes be significant. Unfortunately, the cost evidence in leprosy is limited [25]. A literature search on PubMed using a broad search builder with ‘leprosy’ as MeSH term and ‘economics’ as sub-MeSH heading (year 2001 onwards), resulted in 51 records. Only 6 records presented some cost estimates: three studies focused on a particular event (ENL reaction and ulceration) in hospital settings [26–28]; two cost-effectiveness analysis (CEA) studies on provider’s perspective [29, 30]; and one study on human resource cost of a project [31]. No study was found exclusively on primary care in a general public health setting, covering the patient’s perspective.
Leprosy is a chronic infectious disease with long treatment duration, therefore needs long term care and support, mainly in an outpatient setting. Therefore, the primary objective of our study is to estimate the expenditure in primary (outpatient) care incurred by leprosy patients in two different health system settings in India. The secondary objective is to compare the effect of the health systems on consumer behaviour and practices. The results will help in understanding the economic burden of leprosy in primary care, and eventually contribute in building an investment case for leprosy elimination [25].
The study was conducted under the Leprosy Post Exposure Prophylaxis (LPEP) program, approved in India by the Institutional Human Ethics Committees of the National Institute of Epidemiology (NIE/IHEC201407-01). Written informed consent was received from the respondents and necessary permission was taken from the concerned departments.
India contributes almost 60% to the global leprosy burden [4]. The LPEP program was launched in March 2015 in the Union Territory of Dadra and Nagar Haveli (DNH), located on the western coast of India. The program aims to assess impact and feasibility of contact tracing and administration of single dose of rifampicin (SDR) to asymptomatic contacts of leprosy cases. LPEP is implemented by the National Leprosy Elimination Program (NLEP) of India [32].
The study followed a cross-sectional design, where a cohort from the Union Territory of DNH was compared with a cohort from Umbergaon block of Valsad district, Gujarat, India. A union territory is an administrative division, ruled directly by the federal government, whereas a block is the smallest administrative unit under a district. The cohorts were leprosy cases detected between April 2015 and March, 2016. A sample of 120 participants from each group was selected randomly from the annual leprosy case detection list. In the financial year of 2015–16, DNH reported 425 and Umbergaon reported 287 cases.
DNH and Umbergaon share boundaries and are comparable with regard to demographic, epidemiological, and socioeconomic indicators (Table 1), but not to public health facilities due to the different governmental arrangement (see below).
Both study sites are mainly tribal areas, but there is a remarkable difference in the public health system of both sites. The public health system in DNH is enhanced because it falls directly under the federal government by bypassing provincial bureaucracy, and receives a higher health budget per capita [33–35] than the provinces. In comparison to DNH, Umbergaon has more PHCs per population covered; the average population screened for leprosy by a Primary Health Center (PHC) in Umbergaon was 43% more than DNH PHC (Table 1). The actual screening (active and passive) coverage was reported to be very high in both sites, approximating the total population of these areas. In the year 2015–16, the leprosy program performed two active case detection surveys in both sites. Currently both sites fall under the Leprosy Case Detection Campaign (LCDC), which was launched in early 2016 under the NLEP [36]. Furthermore, the population screened by Umbergaon PHCs is far more than the public health norms for tribal PHCs, i.e. 86% more in Umbergaon and 26% in DNH [37]. Typically, a PHC should cover a population of 20,000 in hilly, tribal, or difficult areas and 30,000 populations in plain areas [37]. Both sites provide free of charge leprosy outpatient department (OPD) services at all public health facilities, but the health systems vary with regard to infrastructure, availability, accessibility, and quality of services.
A household survey was conducted between June and October, 2016 by means of a structured questionnaire. The data were collected by two experienced staff members, post-graduates in public health. The patient, or head of the household, or most knowledgeable person in the household was asked to report on patient demographics, HH socioeconomic status, accessibility of health services, treatment seeking history and OPD expenditure. Respondents were asked to report on the last three OPD visits, either in a public or private facility, in the last 6 months. The database was created in Excel. The analysis included only those patients who mentioned at least 1 OPD visit out of 3.
The costs were categorized as direct and indirect expenditure. The direct part included the expenditure on consultation, investigations and medicines & supplies. The indirect part constituted expenditure on transport, food, and days lost during illness of the patient and attendant. We calculated the transportation expenditure by multiplying to-and-fro distance from house to the nearest health facility, using the government transportation rate [38]. The wage loss was analysed by means of the human capital approach [39]. The wage losses for patients and attendants per illness episode were calculated by using government minimum wage rates [40]. There were 20 (8%) patients who paid at least 1 OPD visit, but failed to report any loss of productive days. For these, we imputed half a day wage loss per visit under the assumption that at least half a day (4 hours) is required to travel and avail services for each illness episode. But attendant’s productive day loss could be zero, as not all patients required attendants. We reported separately the days lost by child patients (age < 16 years) as ‘school days lost’, but while calculating indirect expenditure, all patients and attendants were assumed to be 16 years and older. The results are presented in US dollars (USD) using the conversion rate of INR 67 for 1 dollar for the year 2016 [41]. The analyzed expenditure was exclusively of outpatient services.
In order to answer our objectives, i.e. expenditure and patient’s health seeking behaviour differences in DNH and Umbergaon, we used an integrated analytical approach. The data distribution was evaluated by observing normality plots. The distribution of the direct expenditure variables were not normally distributed due to abundance of zeros and highly skewed for non-zero values, which is common in cost data [42]. The indirect expenditure variables were skewed, but not zero inflated. We compared four different distribution models, i.e. Poisson, negative binomial, zero inflated Poisson, and zero inflated negative binomial distribution [43]. The ‘zero inflated negative binomial regression’ was selected for direct expenditure variables, and ‘negative binomial regression’ for indirect and total expenditure variables. We estimated the mean expenditure for each variable, followed by association measurement between expenditure and patient’s household characteristics. Only significant (p <0.05) variables were modelled together for multivariate regression analysis (Generalized Linear Model). The magnitude of total expenditure was compared against the individual’s monthly income. The total per visit expenditure was defined catastrophic for an individual, if it exceeded 10% of the quarterly income [44, 45]. We assumed that at least one visit to the health centre in a quarter is necessary for regular check-up of leprosy. However as per NLEP norms, patients should visit the health center every month, which rarely happens. In practice, monthly MDT is delivered by staff at the patient’s doorstep and health facility visits occur only during severe illnesses to avoid any wage loss.
A total of 240 patient households (120 in each group) were approached to capture their characteristics and OPD visit details in the last 6 months. The area-wise household characteristics are summarized in Table 2. The mean age (DNH: 25, Umbergaon: 24) showed a young and comparable population in both sites. The average monthly income (DNH: USD 81, Umbergaon: USD 97), expenditure (DNH: USD 73, Umbergaon: USD 83) and saving (DNH: USD 1 Umbergaon: USD 1) per earning member showed a poor economic status in both sites. The respondents differed prominently on characteristics such as distance to the nearest health facility, type of housing, OPD frequency and type of facility visited. Paucibacillary (PB) leprosy was more prevalent in both sites than multibacillary (MB) leprosy. Collectively in the three visits, 69% of the respondents in Umbergaon and 14% of the respondents in DNH had not paid any visit, and were therefore dropped for further analysis.
The three visits expenditure was aggregated to obtain an average per visit. The details of direct and indirect expenditure are shown in Table 3. DNH and Umbergaon were comparable on demographic and socioeconomic parameters, however, they statistically significantly differed with regard to health seeking behaviour. As a behaviour, OPD visit frequency is higher, and a government facility is more preferred in DNH as compared to Umbergaon.
All the presented expenditure estimates are per visit. The mean consultation fee in DNH and Umbergaon was comparable (DNH: USD 1.2, Umbergaon: USD 1.6). The mean expenditure on medicines and supplies (USD 7) was 80% higher in DNH than Umbergaon (USD 4). Only 2 respondents reported investigation expenditure in Umbergaon and none in DNH. Only 1 respondent in Umbergaon and 2 respondents in DNH reported expenditure on food. The mean medical direct expenditure per visit (DNH: USD 6.5, Umbergaon: USD 5.4) was not statistically significantly different between the sites. In indirect expenditure, the mean wage loss for patients was the highest item (DNH: USD 5.2, Umbergaon: USD 7.3), followed by attendant wage loss (DNH: USD 2.7, Umbergaon: USD 3.7). Transportation expenditure (DNH: USD 0.8, Umbergaon: USD 1.4) differed significantly (p ≤ 0.01) in the two groups.
The details on association of expenditures with patient’s household characteristics are shown in Table 4. The proportion of patients with catastrophic expenditure in DNH was 88% less than in Umbergaon. If catastrophic expenditure occurred, then direct expenditure rose three-fold in DNH and two-fold in Umbergaon, (DNH: coef. 2.92, 95% CI: 1.86–3.98; Umbergaon: coef. 1.00, 95% CI: 0.23–1.77). In DNH, the direct expenditure decreased statistically significantly more than two-fold (coef. -2.49, 95% CI: -3.74 to -1.24) with the increase in age groups, whereas a decrease in indirect expenditure against age was not statistically significant (coef. -0.40, 95% CI: -0.92 to 0.12). Umbergaon’s indirect expenditure decreased statistically significantly more than half (coef. -0.79, 95% CI: -1.49 to -0.09) among patients who visited both (government and private) facilities in comparison to those who visited only private facilities. For total expenditure, age and type of facility remained statistically significant factors, whereas catastrophic expenditure remained statistically significant only in DNH. Therefore these factors were considered for the next level of analysis, i.e. multivariate regression.
Table 5 presents the association when only statistically significant variables (p < 0.05) are modelled together with total expenditure (direct + indirect). When modelled separately for both sites, all the variables in Umbergaon turned statistically not-significant. Age however, remained a statistically significant factor (p = 0.03) in DNH. The overall model (Omnibus Test) was statistically significant in DNH (p = 0.001), but not in Umbergaon (p = 0.06). Furthermore, the same model was applied jointly for DNH and Umbergaon (n = 140), which was overall highly significant (p ≤ 0.001). The age (p = 0.019) and type of facility (p = 0.002) were statistically significant, but catastrophic expenditure became statistically not-significant. Catastrophic coefficients however, indicated that catastrophic expenditure groups (in both the areas) had risk of spending (total expenditure) almost twice, compared to non-catastrophic groups.
Our study explored the leprosy patient’s financial burden due to primary care outpatient services. Primary care is an important aspect of disease control under a public health program, therefore costs at this level are important for policy and planning. Moreover, a high out of pocket expenditure indicates public health systems inefficiency, and act as barrier to access services [46]. The results show that the sampled patients were mainly in their economically productive lifetime, indicating leprosy imposing a high economic burden. The leprosy patients of DNH went more frequently to the OPD, and preferred a government facility as compared to Umbergaon. Furthermore, the total expenditure (direct + indirect) was statistically significantly lower in DNH than Umbergaon. The age of the leprosy patients and type of health facilities were the major predictors of total expenditure. The higher the age, the higher the expenditure, and private health facilities were more expensive than government facilities, at both sites.
As a limitation, our study only considered direct and indirect costs, however skin anesthesia (a common phenomenon), neuropathic pain [47, 48], poor mental health [49] and stigma [49, 50] can be significant factors, which can elevate the total expenditure further. We could not focus on these parameters under patient characteristics, and recommend to explore this in detail in future. Next, the households belong to poor socioeconomic groups, which correlates with other studies [9, 13, 22], but we drew the sample from government records, which often caters mainly to poor. Also, adequate representation of patients who are diagnosed and treated completely in private facilities cannot be ascertained. The relatively small sample size is also a limitation of this study. The sample size turned out to be low (reduced power) because of high zero visits, meaning that patients often did not visit the outpatient clinics according to the official schedule. Moreover, to minimize recall bias, we only included the patients of the most recent one year, which was a small cohort. Many patients were not traceable due to migration. Furthermore, we computed catastrophic expenditure based on the income, rather than consumption pattern, which is a more rigorous method. The study is cross-sectional and there is no insight on how patients adapt over time. We recommend to repeat the survey after an appropriate time gap. Also, OPD expenditure is not as high as hospitalization, therefore often failed to be recalled. We do not reject the possibility of recall bias, but we further reduced this by averaging the expenditure from last three visits. Although we have quantified health seeking behaviour, this study does not identify the underlying reasons for these patterns, which would further necessitate qualitative studies.
So far, sound evidence is lacking on the private sector uptake of leprosy cases, therefore we compared the patient’s selection of health facilities for primary leprosy care. We observed that the government is mostly preferred over private health facilities (government 80.8% vs. private 1.7%) in an enhanced health system (DNH). In a non-enhanced health system (Umbergaon) however, private is equally preferred (private 15% vs. government 11.7%). Moreover, in a non-enhanced health system (Umbergaon) patients have poor health seeking behaviour (zero OPD visits in last 6 months: Umbergaon 69% vs. DNH 14%). Contrary to the high number of subjects reporting zero visits, the predicted probability of zero direct medical expenditure (Umbergaon 0.35 vs. DNH 0.88) is lower in Umbergaon, and vice versa in DNH. It means that patients in Umbergaon avoid visiting any health facility, but if they visit then end up paying more than in DNH, therefore out of pocket direct medical expenditure acts as a potential barrier to access leprosy health care. The indirect expenditure is the largest cost impoverishing component for patients. Next, the indirect expenditure with transportation and total expenditure in an enhanced health system (DNH) is lower than non-enhanced health system (Umbergaon). Usually, a high variation is expected in indirect expenditure and transportation, because in many instances they are not paid out of pocket and are presumptive e.g. wage loss. This can lead to over or under reporting. For example, many people use their own vehicle or are supported by others, and often fail to report this. This in turn leads to unrealistic and non-comparable estimates, which are of low utility for policy purposes. Therefore, we used standard government labour market and transportation rates in both areas for comparable results, which are appropriate for the sampled socioeconomic groups. Our study identifies the linkage between socioeconomic factors and expenditure increase. The total expenditure peaked at the 19–35 age category, which correlates with the human capital approach, i.e. the productive age group is more weighted than early or old age [39, 51]. Next, private health facilities are significantly more expensive than government facilities, therefore one of the reasons for higher total expenditure in Umbergaon than DNH.
We conclude that the condition of public health systems has a direct relationship with the patient’s expenditure, and the better the public health system, the lesser the expenditure from the leprosy patient’s pocket. Next, the condition of public health system has a major effect on the patient’s health seeking behaviour, i.e. selection of health facility and services uptake. If a health system is weak, then leprosy patients are forced to seek private health care, which is more expensive and imposes a significant financial burden on the leprosy affected population, proven to be catastrophic. If a public health system is enhanced, then patients prefer to avail government health facility services. We recommend to invest in health system strengthening to reduce the economic burden of leprosy.
|
10.1371/journal.pcbi.1003569 | A Coarse-Grained Elastic Network Atom Contact Model and Its Use in the Simulation of Protein Dynamics and the Prediction of the Effect of Mutations | Normal mode analysis (NMA) methods are widely used to study dynamic aspects of protein structures. Two critical components of NMA methods are coarse-graining in the level of simplification used to represent protein structures and the choice of potential energy functional form. There is a trade-off between speed and accuracy in different choices. In one extreme one finds accurate but slow molecular-dynamics based methods with all-atom representations and detailed atom potentials. On the other extreme, fast elastic network model (ENM) methods with Cα−only representations and simplified potentials that based on geometry alone, thus oblivious to protein sequence. Here we present ENCoM, an Elastic Network Contact Model that employs a potential energy function that includes a pairwise atom-type non-bonded interaction term and thus makes it possible to consider the effect of the specific nature of amino-acids on dynamics within the context of NMA. ENCoM is as fast as existing ENM methods and outperforms such methods in the generation of conformational ensembles. Here we introduce a new application for NMA methods with the use of ENCoM in the prediction of the effect of mutations on protein stability. While existing methods are based on machine learning or enthalpic considerations, the use of ENCoM, based on vibrational normal modes, is based on entropic considerations. This represents a novel area of application for NMA methods and a novel approach for the prediction of the effect of mutations. We compare ENCoM to a large number of methods in terms of accuracy and self-consistency. We show that the accuracy of ENCoM is comparable to that of the best existing methods. We show that existing methods are biased towards the prediction of destabilizing mutations and that ENCoM is less biased at predicting stabilizing mutations.
| Normal mode analysis (NMA) methods can be used to explore potential movements around an equilibrium conformation by mean of calculating the eigenvectors and eigenvalues associated to different normal modes. Each normal mode represents a global collective, correlated and complex, form of motion of the entire protein. Any conformation around equilibrium can be represented as a weighted combination of normal modes. Differences in the magnitudes of the set of eigenvalues between two structures can be used to calculate differences in entropy. We introduce ENCoM the first coarse-grained NMA method to consider atom-specific side-chain interactions and thus account for the effect of mutations on eigenvectors and eigenvalues. ENCoM performs better than existing NMA methods with respect to traditional applications of NMA methods but is the first to predict the effect of mutations on protein stability and function. Comparing ENCoM to a large set of dedicated methods for the prediction of the effect of mutations on protein stability shows that ENCoM performs better than existing methods particularly on stabilizing mutations. ENCoM is the first entropy-based method developed to predict the effect of mutations on protein stability.
| Biological macromolecules are dynamic objects. In the case of proteins, such movements form a continuum ranging from bond and angle vibrations, sub-rotameric and rotameric side-chain rearrangements [1], loop or domain movements through to folding. Such movements are closely related to function and play important roles in most processes such as enzyme catalysis [2], signal transduction [3] and molecular recognition [4] among others. While the number of proteins with known structure is vast with around 85K structures for over 35K protein chains (at 90% sequence identity) in the PDB database [5], our view of protein structure tends to be somewhat biased, even if unconsciously, towards considering such macromolecules as rigid objects. This is due in part to the static nature of images used in publications to guide our interpretations of how structural details influence protein function. However, the main reason is that most known structures were solved using X-ray crystallography [6] where dynamic proprieties are limited to b-factors and the observation of alternative locations. Despite this, it is common to analyze larger conformational changes using X-ray structures with the comparison of different crystal structures for the same protein obtained in different conditions or bound to different partners (protein, ligand, nucleic acid). It is particularly necessary to consider the potential effect of crystal packing [7], [8] when studying dynamic properties using X-ray structures.
Nuclear magnetic resonance (NMR) is a powerful technique that gives more direct information regarding protein dynamics [9], [10]. Different NMR methodologies probe distinct timescales covering 15 orders of magnitude from 10−12 s side chain rotations via nuclear spin relaxation to 103 s using real time NMR [10]. In practice, there is a limitation on the size of proteins that can be studied (between 50–100 kDa) although this boundary is being continuously pushed [11] providing at least partial dynamic information on extremely large systems [12]. However, only a small portion (around 10%) of the available proteins structures in the Protein Data Bank (PDB) are the result of NMR experiments [5].
Molecular dynamic simulations numerically solve the classical equations of motion for an ensemble of atoms whose interactions are modeled using empirical potential energy functions [13]–[15]. At each time step the positions and velocities of each atom are calculated based on their current position and velocity as a result of the forces exerted by the rest of the system. The first MD simulation of a protein (Bovine Pancreatic Trypsin Inhibitor, BPTI) ran for a total 8.8 ps [16] followed by slightly longer simulations (up to 56 ps) [17]. A film of the latter can be seen online (http://youtu.be/_hMa6G0ZoPQ). Despite using simplified potentials and structure representations (implicit hydrogen atoms) as well as ignoring the solvent, these first simulations showed large oscillations around the equilibrium structure, concerted loop motions and hydrogen bond fluctuations that correlate with experimental observations. Nowadays, the latest breakthroughs in molecular dynamics simulations deal with biological processes that take place over longer timescales. For example, protein folding [18], transmembrane receptor activation [19], [20] and ligand binding [21]. These simulations require substantial computer power or purpose built hardware such as Anton that pushes the current limit of MD simulations to the millisecond range [22]. Despite powerful freely available programs like NAMD [23] and GROMACS [24] and the raise of computational power over the last decade, longer simulations reaching timescales where most biological processes take place are still state-of-the-art.
Normal modes are long established in the analysis of the vibrational properties of molecules in physical chemistry [25]. Their application to the study of proteins dates back to just over 30 years [26]–[30]. These earlier Normal Mode Analysis (NMA) methods utilized either internal or Cartesian coordinates and complex potentials (at times the same ones used in MD). As with earlier MD methods their application was restricted to relatively small proteins. Size limitations notwithstanding, these early studies were sufficient to demonstrate the existence of modes representing concerted delocalized motions, showing a facet of protein dynamics that is difficult to access with MD methods. Some simplifications were later introduced and shown to have little effect on the slowest vibrational modes and their utility to predict certain molecular properties such as crystallographic b-factors. These simplifications included the use of a single-parameter potential [31], blocks of consecutive amino acids considered as units (nodes) [32] and the assumptions of isotropic [33] fluctuations in the Gaussian Network Model (GNM) or anisotropic fluctuations [34]. These approximations have drastically reduced the computational time required, thus permitting a much broader exploration of conformational space using conventional desktop computers in a matter of minutes. Of these, the most amply used method is the Anisotropic Network Model (ANM) [35], [36]. ANM is often referred simply as an elastic network model; one should however bear in mind that all normal mode analysis methods are examples of elastic network models. ANM uses a simple Hook potential that connects every node (a point mass defined at the position of an alpha carbon), within a predetermined cut-off distance (usually 18 Å). More recently, a simplified model, called Spring generalized Tensor Model (STeM), that uses a potential function with four terms (covalent bond stretching, angle bending, dihedral angle torsion and non-bonded interaction) has been proposed [37]. The normal mode analysis of a macromolecule produces a set of modes (eigenvectors and their respective eigenvalues) that represent possible movements. Any conformation of the macromolecule can in principle be reached from any other using a linear combination of amplitudes associated to eigenvectors.
It is essential however to not loose sight of the limitations of normal mode analysis methods. Namely, normal modes tell us absolutely nothing about the actual dynamics of a protein in the sense of the evolution in time of atomic coordinates. Plainly speaking, normal mode analysis is informative about the possible movements but not actual movements. Additionally, normal modes tell us of the possible movements around equilibrium. These two caveats clearly place normal mode analysis and molecular dynamics apart. First, molecular dynamics gives an actual dynamics (insofar as the potential is realistic and quantum effects can be ignored). Second, while the equilibrium state (or the starting conformation) affects the dynamics, one can explore biologically relevant timescales given sufficient computational resources to perform long simulations.
The vast majority of coarse-grained NMA models only use the geometry of the protein backbone (via Cα Cartesian position) disregarding the nature of the corresponding amino acid, in doing so a lot of information is lost. To our knowledge, there have been three independent attempts at expanding coarse-grained NMA models over the years to include extra information based on backbone and side-chain atoms. Micheletti et al. [38] developed the βGM model in which the protein is represented by Cβ atoms for all residues except Glycine in addition to the Cα atoms. The Hamiltonian is a function exclusively of Cα and Cβ distances. As a Gaussian model, βGM does not give information about directions of movement but only their magnitude and can be used solely to predict b-factors. As Cβ atoms do not change position, this model cannot be used by definition to predict the effect of mutations, either on dynamics or stability. The authors report results on b-factor prediction comparable to GNM. Lopéz-Blanco et al. [39] developed a NMA model in internal coordinates with three different levels of representation: 1. Heavy-atoms, 2. five pseudo-atoms (backbone: NH, Cα, CO and side-chain: Cβ and one at the center of mass of the remaining side chain atoms) and 3. Cα representation. While the potential is customizable, the default potential uses a force constant that is distance dependent but atom type independent. The method is validated through overlap analysis on a dataset of 23 cases. The authors report no significant differences in overlap for the different representations. Lastly, Kurkcuoglu et al. [40] developed a method that mixes different levels of coarse-graining. The authors test the method on a single protein, Triosephosphate Isomerase with higher atomic representation for a loop and achieve better overlap values compared to Cα only representation. All the methods above, while adding more detail to the representation utilize force constants that are not atom-type dependent. Therefore, while less coarse-grained, all the methods above are still atom-type and amino-acid type agnostic.
By definition, irrespective of the level of coarse-graining, such models cannot account for the effect of mutations on protein dynamics or stability. It has been shown that different amino acids interact differently and that single mutations can have a high impact on protein function and stability [41]–[43]. Mutations on non-catalytic residues that participate into concerted (correlated) movements have been shown to disrupt protein function in NMR relaxation experiments [44]–[46]. Several cases have been documented of mutations that don't affect the global fold of the protein, but affect protein dynamics and disrupt enzyme function [47].
To overcome this limitation of coarse-grained NMA methods while maintaining the advantages of simplified elastic network models, we developed a new mixed coarse-grained NMA model called Elastic Network Contact Model (ENCoM). ENCoM employs a potential function based on the four bodies potential of STeM with an addition to take in consideration the nature and the orientation of side chains. Side-chain atomic contacts are used to modulate the long range interaction term with a factor based on the surface area in contact [48] and the type of each atom in contact. Additionally, we introduce a non-specific version of ENCoM (ENCoMns) where all interactions between atom types are the same. ENCoM and ENCoMns were validated trough comparison to ANM, GNM and STeM with respect to the prediction of crystallographic b-factors and conformational changes, two properties conventionally used to test ENM methods. Moreover, we test the ability of ENCoM and ENCoMns to predict the effect of mutations with respect to protein stability and compare the ability of ENCoM and ENCoMns to a large number of existing methods specifically designed for the prediction of the effect of single point mutations on protein stability. Finally, we use ENCoM to predict the effect of mutations on protein function in the absence of any effects on protein stability.
We utilized a dataset of 113 non-redundant high-resolution crystal structures [49] to predict b-factors using the calculated ENCoM eigenvectors and eigenvalues as described previously [35] (Equation 4). We compared the predicted b-factors using ENCoM, ENCoMns, ANM, STeM and GNM to the experimental Cα b-factors for the above dataset (Supplementary Table S1). For each protein we calculate the Pearson correlation between experimental and predicted values. The results in Figure 1 represent the bootstrapping average of 10000 iterations. We observe that while comparable, ENCoM (median = 0.54) and ENCoMns (median = 0.56) have lower median values than STeM (median = 0.60) and GNM (median = 0.59) but similar or higher than ANM (median = 0.54). It should be noted that it is possible to find specific parameter sets that maximize b-factor correlations beyond the values obtained with STeM and GNM (see methods). However we observe a trade-off between the prediction of b-factors on one side and overlap and the effect of mutations on the other (see methods). Ultimately we opted for a parameter set that maximizes overlap and the prediction of mutations with complete disregard to b-factor predictions. Nonetheless, as shown below, even the lower correlations obtained with ENCoM are sufficiently high to detect functionally relevant local variations in b-factors as a result of mutations. As GNM does not provide information on the direction of movements or the effect of mutations, it is not considered further in the present study.
By definition, any conformation of a protein can be described as a linear combination of amplitudes associated to the eigenvectors representing normal modes. It should be stressed that such conformations are as precise as the choice of structure representation used and correct within the quadratic approximation of the potential around equilibrium. Those limitations notwithstanding, one application of NMA is to explore the conformational space of macromolecules using such linear combinations of amplitudes. Pairs of distinct protein conformations, often obtained by X-ray crystallography are used to assess the extent to which the eigenvectors calculated from a starting conformation could generate movements that could lead to conformational changes in the direction of a target conformation. Rather than an optimization to determine the amplitudes for a linear combination of eigenvectors, this is often simplified to the analysis of the overlap (Equation 5), i.e., the determination of the single largest contribution from a single eigenvector towards the target conformation. In a sense the overlap represents a lower bound on the ability to predict conformational changes without requiring the use of an optimization process.
The analysis of overlap for ANM, STeM, ENCoM and ENCoMns was performed using the Protein Structural Change Database (PSCDB) [50], which contains 839 pairs of protein structures undergoing conformational change upon ligand binding. The authors classify those changes into seven types: coupled domain motions (59 entries), independent domain motions (70 entries), coupled local motions (125 entries), independent local motions (135 entries), burying ligand motions (104 entries), no significant motion (311 entries) and other type of motions (35 entries). The independent movements are movements that don't affect the binding pocket, while dependent movements are necessary to accommodate ligands in the pose found in the bound (holo) form. Burying movements are associated with a significant change of the solvent accessible surfaces of the ligand, but with small structural changes (backbone RMSD variation lower than 1 Å). Despite differentiating between types of movements based on the ligands, the ligands were not used as part of the normal mode analysis. Since side-chain movements associated to the burying movements cannot be predicted with coarse-grained NMA methods, we restrict the analysis to domain and loop movements [51] as these involve backbone movements amenable to analysis using coarse grained NMA methods. For practical purposes, in order to simplify the calculations in this large-scale analysis, NMR structures were not considered. It is worth stressing however that all NMA methods presented here don't have any restriction with respect to the structure determination method and can also be used with modeled structures. A total of 736 conformational changes, half representing apo to holo changes and the other half holo to apo (in total 368 entries from PSCDB) are used in this study (Supplementary Table S2).
Overlap calculations were performed from the unbound (apo) form to the bound form (holo) and from the bound form to the unbound form. Bootstrapped results based onto the best overlap found within the first 10 slowest modes [52], [53] for the different types of conformational changes, domain or loop are shown in figures 2 and 3 respectively. In each case a set of box-plots represent the performance of the four methods being compared, namely STeM, ANM, ENCoM and ENCoMns. The left-most set of box-plots represents the average over all data while subsequent sets represent distinct subsets of the dataset as labeled. The first observation (comparing Figures 2 and 3) is that all tested NMA models show higher average overlaps for domain movements (Figure 2) than loop movements (Figure 3). This confirms earlier observations that NMA methods capture essential cooperative global (delocalized) movements associated with domain movements [51]. Loop movements on the other hand are likely to come about from a more fine tuned combination of normal mode amplitudes than what can be adequately described with a single eigenvector as measured by the overlap.
The second observation is that STeM performs quite poorly compared to other methods irrespective of the type of movement (domain or loop). This is somewhat surprising when one compares with ENCoM or ENCoMns considering how similar the potentials are. This suggests that the modulation of interactions by the surface area in contact (the βij terms in Equation 1) of the corresponding side-chains as well as the specific parameters used are crucial.
Focusing for a moment on domain movements (Figure 2), ENCoM/ENCoMns outperform all other methods for domain movements in general as well as for every sub category of types of motions therein. Independent movements show lower overlaps than coupled ones, a fraction of those movements may not be biologically relevant due to crystal packing. Interestingly, while there are no differences between the overlap for independent movements starting from the apo or holo forms, this is not the case for coupled movements. In this case (right-most two sets in Figure 2), it is easier to use the apo (unbound) form to predict the holo (bound) form, suggesting that the lower packing in the apo form (as this are frequently more open) generates eigenvectors that favor a more comprehensive exploration of conformational space.
Lastly, with respect to loop movements (Figure 3), while it is more difficult to obtain good overlaps irrespective of the method or type of structure used, overall ENCoM/ENCoMns again outperforms ANM. Some of the same patterns observed for domain movements are repeated here. For example, the higher overlap for coupled apo versus holo movements.
We observe that ENCoMns consistently performs almost as well as ENCoM irrespective of the type of motion used (all sets in Figures 2 and 3). As side-chain conformations in crystal structures tent to minimize unfavourable interactions, the modulation of interactions by atom types that differentiate ENCoM from ENCoMns plays as minor but still positive role.
Normal mode resonance frequencies (eigenvalues) are related to vibrational entropy [54], [55] (see methods). Therefore, it is reasonable to assume that the information contained in the eigenvectors can be used to infer differences in protein stability between two structures differing by a mutation under the assumption that the mutation does not drastically affect the equilibrium structure. A mutation may affect stability due to an increase in the entropy of the folded state by lowering its resonance frequencies, thus making more microstates accessible around the equilibrium.
We utilize experimental data from the ProTherm database [56] on the thermodynamic effect of mutations to validate the use of ENCoM to predict protein stability. Here we benefit from the manual curation efforts previously performed to generate a non-redundant subset of ProTherm comprising 303 mutations used for the validation of the PoPMuSiC-2.0 [57]. The dataset contains 45 stabilizing mutations (ΔΔG<−0.5 kcal/mol), 84 neutral mutations (ΔΔG [−0.5,0.5] kcal/mol) and 174 destabilizing mutations (ΔΔG>0.5 kcal/mol) (Supplementary Table S2). Each protein in the dataset have at least one structure in the PDB database [58]. As we calculate the eigenvectors in the mutated form we require model structures of the mutants. We generate such models using Modeller [59] and are thus assume that the mutation does not drastically affect the structure. Mutations were generated using the mutated.py script from the standard Modeller software distribution. Modeller utilizes a two-pass minimization. The first one optimizes only the mutated residue, with the rest of the protein fixed. The second pass optimizes the non-mutated neighboring atoms. It is important to stress that our goal is to model the mutated protein as accurately as possible and thus using any method that unrealistically holds the backbone fixed to model the mutant form would be an unnecessary simplification. We observe a backbone RMSD for the whole protein of 0.01+/−0.01 Å on average. Considering that RMSD is a global measure that could mask more drastic local backbone rearrangements, we also calculated the average maximum Cα displacement but with a value of 0.13+/−0.12 Å we are confident that while not fixed, backbone rearrangements are indeed minimal.
In the present work we predict the effect of mutations (Equation 6) for ENCoM, ENCoMns, ANM and STeM and compare the results to existing methods for the prediction of the effect of mutations using the PoPMuSiC-2.0 dataset above. We compare our results to those reported by Dehouck et al. [57] for different existing techniques: CUPSAT, a Boltzman mean-force potential [60]; DMutant, an atom-based distance potential [61]; PoPMuSiC-2.0, a neural network based method [57]; Eris, a force field based approach [62]; I-Mutant 2.0, a support vector machine method [63]; and AUTO-MUTE, a knowledge based four body potential [64]. We used the same dataset to generate the data for FoldX 3.0, an empirical full atom force-field [65] and Rosetta [66], based on the knowledge based Rosetta energy function. A negative control model was build with a randomized reshuffling of the experimental data. Figure 4 presents RMSE results for each model. The raw data for the 303 mutations is available in Supplementary Table S3.
ANM and STeM are as good as the random model when considering all types of mutations together (Figure 4). This is not surprising as the potentials used ANM and STeM are exclusively geometry-based and are thus agnostic to sequence. ENCoMns, ENCoM, Eris, CUPSAT, DMutant, I-Mutant 2.0 and give similar results and predict significantly better than the random model. AUTO-MUTE, FoldX 3.0, Rosetta and in particular PoPMuSiC-2.0 outperform all of the other models.
The RMSE values for the subset of 174 destabilizing mutations (Figure 5) shows similar trends as the whole dataset with the exceptions of DMutant losing performance and PoPMuSiC-2.0 as well as AUTO-MUTE gaining performance compared to the others. It is important to stress that the low RMSE of PoPMuSiC-2.0 on the overall dataset is to a great extent due to its ability to predict destabilizing mutations.
The subset of 45 stabilizing mutations (Figure 6) gives completely different results as those obtained for destabilizing mutations. AUTO-MUTE, Rosetta, FoldX 3.0 and PoPMuSiC-2.0 that outperformed all of the models on the whole dataset or the destabilizing mutations dataset cannot predict better than the random model. This is also true for CUPSAT, I-Mutant 2.0 and Eris. ENCoM and DMutant are the only models with significantly better than random RMSE values for the prediction of stabilizing mutations.
ANM and STeM outperform all models on the neutral mutations (Figure 7). All other models fail to predict neutral mutations any better than random. While the accuracy of ANM and STeM to predict neutral mutations may seem surprising at first, it is in fact an artifact of the methodology. As the wild type or mutated structures are assumed to maintain the same general backbone structure, the eigenvectors/eigenvalues calculated with ANM or STeM will always be extremely similar for wild type and mutant forms. Any differences will arise as a result of small variations in backbone conformation produced by Modeller. As such, ANM and STeM predict almost every mutation as neutral, explaining their high success in this case.
At first glance, the comparison of ENCoMns and ENCoM could suggest that a large part of the effect observed come from a consideration of the total area in contact and not the specific types of amino acids in contact. However, the side chains in contact are already in conformations that minimize unfavourable contacts to the extent that is acceptable in reality (in the experimental structure) or as a result of the energy minimization performed by Modeller for the mutant form given the local environments. The fact that ENCoM is able to improve on ENCoMns is the actual surprising result and points to the existence of frustration in molecular interactions [67].
Considering that none of the existing models can reasonably predict neutral mutations, the only models that achieve a certain balance in predicting both destabilizing as well as stabilizing mutations better than random and with low bias are ENCoM and DMutant.
The analysis of the performance of ENCoM in the prediction of different types of mutations in terms of amino acid properties shows that mutations from small (ANDCGPSV) to big (others) residues are the most accurately predicted followed by mutations between non-polar or aromatic residues (ACGILMFPWV). ENCoM performs poorly on exposed residues (defined as having more than 30% of the surface area exposed to solvent) (Figure 8).
It may in principle be possible to find particular linear combinations of ENCoM and other methods that further improve predictions given the widely different (and potentially complementary) nature of the various approaches with respect to ENCoM. We performed linear regressions to find parameters involving ENCoM and each of the other methods in turn that maximize the RMSE difference between the combined models and random predictions (Eq. 8). When considering all types of mutations together, all mixed models perform better than either model individually (left-most column in Figure 8). By definition, a mixed model cannot perform worst than the better of the two models individually. The contribution of ENCoM to the improved performance of the combined model varies according to the model. The ratio of the relative contributions (in parenthesis), broadly classifies the methods into three categories: 1. Methods where ENCoM contributes highly, including I-Mutant (1.18±0.25), DMutant (1.07±0.31), CUPSAT (1.02±0.09) and Eris (1.02±0.11); 2. Methods where the contribution of ENCoM is smaller than that of the other method but still significant, including Rosetta (0.90±0.11) and FoldX3 (0.89±0.08); and finally methods where the addition of ENCoM have a small beneficial effect, including in this class Automute (0.69±0.13) and especially PoPMuSIC (0.18±0.10). The left-hand side dendrogram in Figure 8 clusters the methods according to their overall accuracy relative to random based on the entire profile of predictions (Eq. 8) for different subsets of the data (columns) according to the type of mutations being predicted. This clustering of methods shows that the relative position of the methods is maintained throughout except for a small rearrangement due to changes in the predictions for CUPSAT. This result suggests that the contribution from the combination of ENCoM to other methods is uniform irrespective of the type of mutations studied.
One basic requirement for a system that predicts the effect of mutations on stability is that it should be self-consistent, both unbiased and with small error with respect to the prediction of the forward or back mutations as reported by Thiltgen et al. [68]. The authors built a non-redundant set of 65 pairs of PDB structures containing single mutations (called form A and form B) and utilized different models to predict the effect of each mutation going from the form A to form B and back. From a thermodynamic point of view, the predicted variation in free energy variation should be of the same magnitude for the forward or back mutations, ΔΔGA→B = −ΔΔGB→A.
Using the Thiltgen dataset we performed a similar analysis for ENCoM, ENCoMns, ANM, STeM, CUPSAT, DMutant, PoPMuSiC-2.0 and a random model (Gaussian prediction with unitary standard deviation). For the remaining methods (Rosetta, Eris and I-Mutant) we utilize the data provided by Thiltgen. We removed three cases involving prolines as such cases produce backbone alterations. Furthermore, PoPMuSiC-2.0 failed to return results for five cases. The final dataset therefore contains 57 pairs (Supplementary Table S4). The CUPSAT and AUTO-MUTE servers failed to predict 25 and 32 cases respectively. As these failure rates are significant considering the size of the dataset, we prefer to not include these two methods in figures 9 and 10 (the remaining cases appear however in Supplementary Table S4).
The results in figure 9 show that compared to the random model (a positive control in this case), Rosetta and FoldX 3.0 show moderate bias while PoPMuSiC-2.0 and I-Mutant show significant bias. All biased methods are biased toward the prediction of destabilizing mutations (data not shown) in agreement with the results in Figure 3. DMutant, Eris, ENCoM and ENCoMns are the only models with bias comparable to that of the random model (the positive control in this experiment).
ENCoM, ENCoMns, and to a lesser extent Rosetta and DMutant have lower errors than the random model (Figure 10). All other methods display an error equal or higher than that of the random model. ENCoM and ENCoMns vastly outperform all the others models in terms of error.
Lastly, STeM and ANM show low and moderate biases respectively and errors equivalent to random (data not shown) but as mentioned, these methods cannot be used for the prediction of mutations (other than neutral mutations as an artefact).
Mutations may not only affect protein stability but also protein function. While experimental data is less abundant, one protein in particular, dihydrofolate reductase (DHFR) from E. coli, has been widely used experimentally to understand this relationship [69], [70]. Recently, Boher et al. [47] have analyzed the effect of the G121V mutation on protein dynamics in DHFR by NMR spectroscopy. This mutation is located 15 Å away from the binding site but reduces enzyme catalysis by 200 fold with negligible effect on protein stability (0.70 kcal/mol). The authors evaluated, among many other parameters, the S2 parameter of the folic acid bound form for the wild type and mutated forms and identify the regions where the mutation affects flexibility. We calculated b-factor differences (Equation 4) between the folate-bound wild type (PDB ID 1RX7) and the G121V mutant (modeled with Modeller) forms of DHFR (Supplementary Table S5). We obtain a good agreement (Pearson correlation = 0.61) between our predicted b-factor difference and S2 differences (Figure 11). As mentioned earlier, the overall correlation of 0.54 in the prediction of b-factors (Figure 1) appears at least in this case to be sufficient to capture essential functional information.
Our results show that a small modification of the long-range interaction term in the potential energy function of STeM had an important positive impact on the model. This small change improves the method in comparison to existing NMA methods in the traditional areas such as the prediction of b-factors and conformational sampling (overlap) where coarse-grained normal mode analysis are applied. More importantly however, it opens an entire new area of application to coarse-grained normal mode analysis methods. Specifically ENCoM is the first coarse-grained normal-mode analysis method that permits to take in consideration the specific sequence of the protein in addition to the geometry. This is introduced through a modification in the long-range interactions to account for types of atoms in contact modulated by their surface in contact. As a validation of the approach we explored the ability of the method to predict the effect of mutations in protein stability. In doing so we created the first entropy-based methodology to predict the effect of mutations on the thermodynamic stability of proteins. This methodology is entirely orthogonal to existing methods that are either machine learning or enthalpy based. Not only the approach is novel but also the method performs extremely favourably compared to other methods when viewed in terms of both error and bias.
As the approach taken in ENCoM is completely different from existing methods for the prediction of the effect of mutations on protein stability, a new opportunity arises to combine ENCoM with enthalpy and machine-learning methods. Unfortunately, we tried to create a naïve method based on linear combinations of the predictions of ENCoM and the different methods presented without success, perhaps due to the large bias characteristic to the different methods.
To assess the relative importance of contact area and the modulation of interactions with atom types, we tested a model that has non-specific atom-type interactions (ENCoMns), this model is atom type insensitive, but is sensitive the orientation of side-chain atoms. While a large fraction of the observed effect can be attributed to surfaces in contact only, ENCoM is consistently better than ENCoMns, particularly at predicting destabilizing mutations where the possibility to accommodate unfavourable interactions is more restricted. We cannot however exclude the effect of the intrinsic difficulty in modeling destabilizing mutations. For stabilizing mutations, the near equivalence of ENCoM and ENCoMns may be explained in part by the successful energy minimization of the mutated side-chain performed by Modeller. ANM and STeM failed to predict the effect of mutations on the whole dataset. They were not expected to perform well because their respective potentials only take in account the position of alpha carbons (backbone geometry). As such ANM and STeM tend to predict mutations as neutral, explaining their excellent performance onto the neutral subset and failure otherwise. Our results suggest that surfaces in contact are essential in a coarse-grained NMA model to predict the effect of mutation and that the specific interactions between atom types is necessary to get more subtle results, particularly stabilizing mutations.
ENCoM is consistently better than ENCoMns in the prediction of loop or domain movements irrespective of the dependency of the coupling of this movement to ligand binding or the starting structure (apo or holo form) and both outperform ANM and STeM. Our results corroborate previous work on a mix coarse-grained method adding a atomistic resolution to loops capable of improving the prediction of loop movements [40]. ENCoM performs considerably better than STeM throughout despite having very similar potentials, showing the importance of surfaces in contact in the prediction of movements. There is little difference between ENCoMns and ENCoM in the prediction of b-factors, but both perform worst than ANM, STeM and GNM. At least in the case of DHFR b-factor differences capture some essential characteristics of the system as calculated by NMR. However, one should be careful in placing too much emphasis on the validation of b-factor predictions using experimental data derived from crystals as these are affected to a great extent by rigid body motions within the crystal [71].
PoPMuSiC-2.0, AUTO-MUTE, FoldX 3.0 and Rosetta perform better than other models in the whole test dataset of mutations. However, the dataset consists of 15% stabilizing mutation, 57% of destabilizing and 28% of neutral mutations. When looking at each subset, machine learning or enthalpy based models failed to predict better than random on the stabilizing mutations subset. Biases in the dataset may have affected the training of machine-learning methods. For example the training set of PoPMuSiC-2.0 contains 2648 mutations in proportions that are similar to those in the testing set with 60%, 29% and 11% destabilizing, neutral and stabilizing mutations respectively. While it is true that most mutations tend to be destabilizing, if one is interested in detecting stabilizing mutations, a method over trained on destabilizing mutations will not meet expectations. Indeed, PoPMuSiC-2.0 and I-mutant the two machine learning based methods, have larger biases and errors than other methods in their predictions. Our method relies on a model structure of the mutant. As the modeling may fail to find the most stable side-chain conformation, it could have a bias toward giving slightly higher energies to the mutant. Notwithstanding this potential bias, ENCoM have the lowest error and bias. This may be a case where less is more as the coarse-grained nature of the method makes it also less sensitive to errors in modeling that may affect enthalpy-based methods to a greater extent. Finally, there is one more advantage in the approach taken in ENCoM. As the network model is a global connected model it considers indirectly the entire protein, while in existing enthalpy or machine-learning methods the effect of a mutation is calculated mostly from a local point of view.
The prediction of the thermodynamic effect of mutations is very important to understand disease-causing mutations as well as in protein engineering. With respect to human diseases, and particularly speaking of cancer mutations, one of the factors that may lead to tumour suppressor or oncogenic mutations is their effect on stability (the authors thank Gaddy Getz from the Broad Institute for first introducing us to this hypothesis). Specifically, destabilizing mutations in tumour suppressor genes or alternatively stabilizing mutations in oncogenes may be driver mutations in cancer. Therefore the prediction of stabilizing mutations may be very important to predict driver mutations in oncogenes. Likewise, in protein engineering, one major goal is that of improving protein stability with the prediction of stabilizing mutations. Such mutations may be useful not as the final goal (for purification or industrial purposes) but also to create a ‘stability buffer’ that permits the introduction of potentially destabilizing additional mutations that may be relevant to create the intended new function.
The work presented here is to our knowledge also the most extensive test of existing methods for the prediction of the effect of mutations in protein stability. The majority of methods tested in the present work fail to predict stabilizing mutations. However, we are aware that the random reshuffled model used may be too stringent given the excessive number of destabilizing mutations in the dataset. The only models that predict stabilizing as well as destabilizing mutations are ENCoM and DMutant, however ENCoM is the only method with low self-consistency bias and error.
While the contribution of side chain entropy to stability is well established [72], [73], here we use backbone normal modes to predict stability. As a consequence of the relationship between normal modes and entropy, our results attest to the importance of backbone entropy to stability and increase our understanding of the overall importance of entropy to stability. The strong trend observed on the behaviour of different parameters sets with respect to the α4 parameter is very interesting. Lower values are associated with better predictions of conformational changes while higher values are associated with better b-factor predictions. One way to rationalize this observation is to consider that higher α4 values lead to a rigidification of the structure, adding constraints and restricting overall motion. Likewise, lower α4 values remove constraints and thus lead to higher overlap.
We used ENCoM to predict the functional effects of the G121V mutant of the E. coli DHFR compared to NMR data. This position is part of a correlated network of residues that play a role in enzyme catalysis but with little effect on stability. The mutation affects this network by disrupting the movement of residues that are far from the binding site. We can predict the local changes in S2 with ENCoM. As these predictions are based on b-factor calculations, this result shows that at least in this case, even with b-factor prediction correlation lower than ANM, STeM and GNM we can detect functionally relevant variations. Clearly, despite the greater performance of GNM, ANM or STeM in the calculation of b-factors, these methods cannot predict b-factor differences as a consequence of mutations, as their predictions are the same for the two forms. While a more extensive study is necessary involving S2 NMR parameters, our results serve as an example against relying too heavily on crystallographic b-factors for the evaluation of normal mode analysis methods.
The fundamentals of Normal Mode Analysis (NMA) have been extensively reviewed [74], [75]. The key assumption in NMA is that the protein is in an equilibrium state around which fluctuations can be described using a quadratic approximation of the potential via a Taylor series approximation. In equilibrium the force constants are summarized in the Hessian matrix H that contains the elements of the partial second derivatives of the terms of the potential with respect to the coordinates. The potential used in ENCoM is similar to that of STeM, a Go-like potential where the closer a conformation is to the reference (in this case equilibrium) conformation , the lower the energy.(1)The principal difference between the potential above and that of STeM are the terms that modulate non-bonded interactions between amino acid pairs according to the surface area in contact. Specifically,(2)where represents a pairwise interaction energy between atom types and of atom and respectively of amino acids and containing and atoms each. Finally, represent the surface area in contact between atoms and calculated analytically [48]. We utilize the atom types classification of Sobolev et al. [76] containing 8 atom types. A matrix with all the interaction between atom types set at the value 1 is used in the non-specific ENCoMns model. In Figure 12 we illustrate with the concrete case of a set of 3 amino acids (D11, D45 and R141) in M. tuberculosis ribose-5-phosphate isomerase (PDB ID 2VVO), the differences between ENCoM, STeM and ANM in terms of spring strengths associated to different amino acids pairs. D45 is equally distant from R141 and D11 (around 6.0 Å) and interacts with R141 but does not with D11. Likewise, D11 does not interact with R141 with a Cα distance of 11.6 Å. ANM assigns equal strength to all three pairwise Cα springs (as their distances fall within the 18.0 Å threshold). STeM assigns equal spring strengths to the D11–D45 and D45–R141 pairs. Among the three methods, ENCoM is the only one to properly assign an extremely weak strength to the D11–R141 pair, a still weak but slightly stronger strength to the D11–D45 pair (due to their closer distance) and a very strong strength to the spring representing the D45–R141 interactions.
The hessian matrix can be decomposed into eigenvectors and their associated eigenvalues . For a system with N amino acids (each represented by one node in the elastic network), there are 3N eigenvectors. Each eigenvector describes a mode of vibration in the resonance frequency defined by the corresponding eigenvalue of all nodes, in other words the simultaneous movement in distinct individual directions for each node (Cα atoms in this case). The first 6 eigenvectors represent rigid body translations and rotations of the entire system. The remaining eigenvectors represent internal motions. The eigenvectors associated to lower eigenvalues (lower modes) represent more global or cooperative movements while the eigenvectors associated to higher eigenvalues (fastest modes) represent more local movements. Any conformation of the protein can be described by a linear combination of different amplitudes of eigenvectors :(3)The source code for ENCoM is freely available at http://bcb.med.usherbrooke.ca/encom.
In terms of running time, the computational cost of running ENCoM is only slightly higher than that of other methods representing the protein structure with one node per amino acid. The main bottleneck in terms of computational time is the diagonalization of the Hessian matrix. As this matrix is the same size for ENCoM, ANM, STeM and GNM by virtue of considering a single node per amino acid, all methods should in principle run equivalently. Differences occur due to pre-processing, in particular with ENCoM where this step is more involved due to the more detailed calculations involved in the measurement of surface areas in contact. Taking the dataset used for the prediction of b-factors as an example, we obtain an average running time of 23.8, 30.2 and 34 seconds on average for STeM, ANM and ENCoM respectively on an Intel Core i7 CPU Q 740 @ 1.73GHz laptop.
In order to obtain a set of parameters to be used with ENCoM we performed a sparse exhaustive integer search of the logarithm of parameters with for to maximize the prediction ability of the algorithm in terms of overlap and prediction of mutations. In other words, we searched all combinations of 13 distinct relative orders of magnitude for the set of 4 parameters . For each parameter set, we calculated the bootstrapped median RMSE (see below) Z-score sum for the prediction of stabilizing and destabilizing mutations, . Keeping in mind that lower RMSE values represent better predictions, the 2000 parameter sets (out of 28561 combinations) with highest were then used to calculate Z-scores for overlap in domain and loop movements. As our goal is to obtain a parameter set that combines low RMSE and high overlap, we ranked the 1000 parameter sets according to . The parameter set with highest is (solid black line in Figure 13). The optimization of the bootstrapped median is equivalent to a training procedure with leave-many-out testing. Given the dichotomy in predicting the effect of mutations and overlap on the one hand and b-factors on the other, we provide, the following is the best parameter set observed for the prediction of b-factors with average b-factor correlation of .
The exploration of parameter space shows that there is a clear trade-off between the prediction of mutations (low RMSE), conformational sampling (high overlap) and b-factors (high correlations). Parameter sets that improve the prediction of b-factors are invariably associated with poor conformational changes (low overlap) associated to both domain and loop movements and variable RMSE for the prediction of mutations (red lines in Figure 13). On the other hand, parameter sets that predict poorly b-factors, perform better in the prediction of conformational changes and the effect of mutations (blue lines in Figure 13). The parameters used in STeM, are arbitrary, taken without modifications from a previous study focusing on folding [77]. As expected, this set of parameters can be considerably improved upon as can be observed in Figure 13 (dashed line).
The four right-most variables in the parallel coordinates plot in Figure 13 show the logarithm of the α parameters for each parameter set. Either class of parameter sets, better for b-factors (in red) or better for overlap/RMSE (in blue) come about from widely diverging values for each parameter across several orders of magnitude. There are however some patterns. Most notably for α4, where there is an almost perfect separation of parameter sets around α4 = 1. Interestingly, higher values of α1 and α2, associated with stronger constraints on distances and angles tend also to be associated to better overlap values. While it is likely that a better-performing set of parameters can be found, the wide variation of values across many orders of magnitude show that within certain limits, the method is robust with respect to the choice of parameters. This result justifies the sparse search employed.
Bootstrapping is a simple and general statistical technique to estimate standard errors, p-values, and other quantities associated with finite samples of unknown distributions. In particular, bootstrapping help mitigate the effect of outliers and offers better estimates in small samples. Bootstrapping is a process by which the replicates (here 10000 replicates) of the sample points are stochastically generated (with repetitions) and used to measure statistical quantities. In particular, bootstrapping allows the quantification of error of the mean [78]–[82]. Explained in simple terms, two extreme bootstrapping samples would be one in which the estimations of the real distribution of values is entirely made of replicates of the best case and another entirely of the worst case. Some more realistic combination of cases in fact better describes the real distribution. Thus, bootstrapping, while still affected by any biases present in the sample of cases, helps alleviate them to some extent.
One of the most common types of experimental data used to validate normal mode models is the calculation of predicted b-factors and their correlation to experimentally determined b-factors. B-factors measure how much each atom oscillates around its equilibrium position [6]. Predicted b-factors are calculated as previously described [35]. Namely, for a given Cα node (i), one calculates the sum over all eigenvectors representing internal movements (n = 7 to 3N) of the sum of the squared ith component of each eigenvector in the spatial coordinates x,y and z normalized by the corresponding eigenvalues:(4)We calculate the Pearson correlation between predicted and experimental b-factors for each protein and average random samples according to the bootstrapping protocol described above.
The overlap is a measure that quantifies the similarity between the direction of movements described by eigenvectors calculated from a starting structure and the variations in coordinates observed between that conformation and a target conformation [52], [53]. In other words, the goal of overlap is to quantify to what extent movements based on particular eigenvectors can describe another conformation. The overlap between the nth mode, , described by the eigenvector is given by(5)where represent the vector of displacements of coordinates between the starting and target conformations. The larger the overlap, the closer one can get to the target conformation from the starting one though the movements defined entirely and by the nth eigenvector. We calculate the best overlap among the first 10 slowest modes representing internal motions.
Insofar as the simplified elastic network model captures essential characteristics of the dynamics of proteins around their equilibrium structures, the eigenvalues obtained from the normal mode analysis can be directly used to define entropy differences around equilibrium. Following earlier work [54], [55], the vibrational entropy difference between two conformations in terms of their respective sets of eigenvalues is given by:(6)In the present work the enthalpic contributions to the free energy are completely ignored. Therefore, in the present work we directly compare experimental values of ΔG to predicted ΔS values. In order to use the same nomenclature as the existing published methods, we utilize ΔΔG to calculate the variation of free energy variation as a measure of conferred stability of a mutation.
A linear regression going through the origin is build between predicted ΔΔG and experimental ΔΔG values to evaluate the prediction ability of the different models. The use of this type of regression is justified by the fact that a comparison of a protein to itself (in the absence of any mutation) should not have any impact on the energy of the model and the model should always predict an experimental variation of zero. However, a linear regression that is not going through the origin would predict a value different from zero equal to the intercept term. In other words, the effect of two consecutive mutations, going from the wild type to a mutated form back to the wild type form (WT→M→WT) would not end with the expected net null change.
The accuracy of the different methods was evaluated using a bootstrapped average root mean square error of a linear regression going through the origin between the predicted and experimental values. We refer to this as RMSE for short and use it to describe the strength of the relationship between experimental and predicted data.
If one was to plot the predicted energies variation of ΔΔGA→B versus ΔΔGB→A and trace a line y = −x, the bias would represent a tendency of a model to have points not equally distributed above or below that line while the error would represent how far away a point is from this line. In other words, considering a dataset of forward and back predictions, the error is a measure of how the predicted ΔΔG differ and the bias how skewed the predictions are towards the forward or back predictions [68]. A perfect model, both self-consistent and unbiased, would have all the points in the line. Statistically, the measures of bias and error are positively correlated. The higher the error for a particular method, the higher the chance of bias.
We determined the efficacy of linear combinations involving ENCoM and any of the other models for the prediction of the effect of mutations on thermostability as follows. For a given bootstrap sample of the data, we rescaled the predictions of each model as follows:(7)where the vector notation signifies all data points in the particular bootstrap sample and the index represents each model as well as the random model. We then use singular value decomposition to determine the best parameter the normalized predicted values to calculate the parameters that maximize the RMSE difference between the linear combination model and the random predictions as follows(8)whereand(9)The bootstrapped average is then calculated from the 10000 bootstrap iterations. The relative contribution of ENCoM and the model under consideration is given by the ratio of the parameters . It is interesting to note that this ratio could be seen as an effective temperature factor, particularly considering that predicted values are primarily enthalpic in nature for certain methods and entropy based in ENCoM.
|
10.1371/journal.pcbi.1005607 | Advances in using Internet searches to track dengue | Dengue is a mosquito-borne disease that threatens over half of the world’s population. Despite being endemic to more than 100 countries, government-led efforts and tools for timely identification and tracking of new infections are still lacking in many affected areas. Multiple methodologies that leverage the use of Internet-based data sources have been proposed as a way to complement dengue surveillance efforts. Among these, dengue-related Google search trends have been shown to correlate with dengue activity. We extend a methodological framework, initially proposed and validated for flu surveillance, to produce near real-time estimates of dengue cases in five countries/states: Mexico, Brazil, Thailand, Singapore and Taiwan. Our result shows that our modeling framework can be used to improve the tracking of dengue activity in multiple locations around the world.
| As communicable diseases spread in our societies, people frequently turn to the Internet to search for medical information. In recent years, multiple research teams have investigated how to utilize Internet users’ search activity to track infectious diseases around our planet. In this article, we show that a methodology, originally developed to track flu in the US, can be extended to improve dengue surveillance in multiple countries/states where dengue has been observed in the last several years. Our result suggests that our methodology performs best in dengue-endemic areas with high number of yearly cases and with sustained seasonal incidence.
| Dengue fever poses a growing health and economic problem throughout the tropical and sub-tropical world. Dengue is today one of the fastest-growing and most important mosquito-borne viral diseases in the world, with an estimated 390 million infections each year and threatening an estimated 3.9 billion people in 128 countries [1]. Infection often causes high fever and joint pain, and severe cases can lead to hemorrhage, shock and death. Dengue epidemics strain health services and lead to massive economic losses.
Dengue transmission is subject to complex environmental factors influencing the Aedes aegypti and albopictus mosquitoes which spread the disease. A mosquito is able to transmit dengue within a few weeks of contracting the virus, and a person bitten by such a mosquito will usually fall ill within a week, with symptoms lasting for up to 10 days afterward [2, 3]. There is a 5-day window when another mosquito can pick up the virus from an infected person [3]. The time scale of these transmission dynamics lends itself to tracking patterns of infection at a weekly or monthly level. Seasonal conditions such as temperature and precipitation can affect mosquito feeding rate, development, and lifespan, contributing to annual seasonality in observed dengue cases [4–8]. Other factors affecting the local or regional level include human population density and mobility, mosquito control efforts, and the distribution of the four dengue virus serotypes, adding complications to efforts to model transmission dynamics [9, 10].
Dengue mortality and morbidity both need to be addressed to reduce this heavy burden. The World Health Organization has called for better early case detection among other tactics to reduce dengue mortality, and for the reduction of dengue morbidity through coordinating epidemiological and entomological surveillance. Timely identification of outbreaks can inform and help preventative measures to lower infection rates, including mosquito population control and providing supplies such as screens and nets for mosquito bite prevention. Thorough, data-informed implementations of these vector control methods have been found effective in reducing case counts in many locations, but require sustainable investment to prevent resurgence [3, 11], highlighting the need for accurate and timely dengue surveillance tools.
However, such a comprehensive, effective and reliable disease surveillance system for dengue is not yet available. Governments traditionally rely on hospital-based reporting, a method that is often lagged and limited with frequent post-hoc revisions, due to communication inefficiencies across local and national agencies and the time needed to aggregate information from the clinical to the state level [12, 13]. This lack of timely information limits the identification and optimization of effective interventions. Measurement difficulties are compounded by the fact that a majority of dengue cases are asymptomatic [14].
In this context, building an effective disease surveillance tool is essential to being able to identify and assess the severity of dengue outbreaks and to enable better assessment of the effectiveness of ongoing interventions. Such tools should provide accurate and consistent measures of regional or national infection levels, be updated in near real-time, and be immune to bureaucratic or resource-related delays. To improve accuracy, these tools should use and link together multiple sources of information, using both traditional and non-traditional sources.
Over the years, a broad range of traditional epidemiological methods have been proposed by research teams to fill this time gap of information by supplementing official case counts with now-cast estimation using dengue incidence data from previous seasons. Autoregressive models, such as Seasonal Autoregressive Integrated Moving Average (SARIMA) model, that take into account recent and seasonal patterns, have been shown to produce useful disease estimates, some including additional variables such as concurrent weather information [15–19]. Other studies have further examined climate-driven models, finding associations of seasonal and long-term weather patterns such as El Niño with dengue levels in various countries [5, 20–22]. In addition, various mechanistic models on the dynamics of dengue transmission have also been explored, with some recent promise [23]. A comprehensive survey of these methods are given in Andraud et al. [24]
In parallel and complementary to the aforementioned methodologies, the global spread of the Internet has opened up the opportunity to investigate whether users’ activity patterns on Internet search-engines and social media platforms may lead to reasonable estimates of dengue infection levels [25–27]. In theory, Internet search tracking is consistent, efficient, and reflects real-time population trends, giving it strong potential to supplement current epidemiological methods [13, 28]. Studies have previously demonstrated the feasibility of using Internet search data to track dengue case counts [25, 27]. Google Dengue Trends (GDT), launched in 2011, was one of the first tools to quantitatively track dengue activity in multiple regions throughout the world by leveraging the aggregate Google search patterns of millions of users [25]. Since its start, the methodology behind GDT has been updated to address flaws found in its sister effort, Google Flu Trends [29–37], before finally being discontinued in August 2015. An assessment of GDT in Mexico showed mixed prediction accuracy compared to official case counts, with strong correlation in some states [38].
Despite progress in the use of both dengue time-series information (time series approaches [15]) and real-time Internet searches for dengue tracking [25], an approach for accurate tracking of dengue by combining the respective strengths of each data source has not been documented to the best of our knowledge. We extend a methodology recently introduced in the flu surveillance literature to combine dengue-related Google searches with dengue case count time-series to track dengue activity. Specifically, we evaluate the performance of the ARGO model (AutoRegressive model with GOogle search queries as exogenous variables), as introduced in [35], in tracking dengue in five countries/states around the globe: Mexico, Brazil, Thailand, Singapore, and Taiwan. These countries were chosen to explore the applicability of our approach in a diverse set of ecological situations where dengue has been identified as an important local threat. Our contribution shows that the lessons learned to track influenza in data-rich environments, like the United States, can be used to develop methodologies to track an often-neglected tropical disease, dengue, in data-poor environments.
We used two kinds of data sets for our study: (a) historical dengue incidence from government-led health agencies and (b) Google search fractions of dengue-related queries, aggregated at the national-level.
We used the multivariate linear regression modeling framework ARGO (AutoRegressive model with GOogle search queries as exogenous variables) [35], previously used to track flu incidence using flu-related Google searches, to combine information from historical dengue case counts and dengue-related Google search frequencies with the goal of estimating dengue activity one month ahead of the publication of official local health reports. ARGO uses a training set that consists of a two-year moving time window (immediately prior to the month of estimation) and an L1 regularization approach, to identify the best performing parsimonious model [39]. This moving window approach allows the model to constantly improve its predictive ability by capturing the changing relationship between Internet search behavior and dengue activity.
For comparison with ARGO, we included estimation results from five alternative methods. These are:
All benchmark models (except the naive method) were trained by linear regression with sliding two year windows for fair comparison.
We used five accuracy metrics to compare model performance: root mean squared error (RMSE), mean absolute error (MAE), root mean squared percentage error (RMSPE), mean absolute percentage error (MAPE), and Pearson correlation.
Mathematically, these accuracy metrics of estimator c ^ for target dengue case count c are defined as, RMSE = [ 1 / n ∑ t = 1 n ( c ^ t - c t ) 2 ] 1 / 2, MAE = 1 / n ∑ t = 1 n | c ^ t - c t |, RMSPE = { 1 / n ∑ t = 1 n [ ( c ^ t - c t ) / c t ] 2 } 1 / 2, MAPE = 1 / n ∑ t = 1 n | c ^ t - c t | / c t.
Retrospective out-of-sample estimates of dengue case counts were generated for each country using ARGO and the five benchmark models, assuming we only had access to information available at the time of estimation. The time windows in which we assessed the performance of our dengue estimates for each country were chosen based on the availability of official and GDT benchmark data.
These time windows are: Brazil from Mar 2006–Dec 2012, Mexico from Mar 2006–Aug 2015, Thailand from Oct 2010–Aug 2015, Singapore from Feb 2008–Aug 2015, and Taiwan from Jan 2013–Mar 2016.
In four of the five countries/states, Brazil, Mexico, Thailand and Singapore, ARGO outperformed all benchmark models across essentially all accuracy metrics (RMSE, MAE, RMSPE, MAPE, correlation). See Table 1. In particular, by incorporating information from the Internet searches and the dengue time-series, ARGO achieved better results than using either information alone. This accuracy improvement is reflected in the decreased errors during both peaks of dengue activity and off-season/periods with low levels of infection. See Fig 1. Unlike the seasonal autoregression with GDT model (SAR+GDT), ARGO avoided the significant overshooting problem that has been previously noted in Google Trends data ([35], [40]). This is especially notable between 2006–2008 and 2012–2014 in Mexico, and 2006–2010 in Brazil.
Taiwan shows notably different results. Of all the available models, the naive and seasonal autoregressive models have the best performance, but neither is clearly effective. The naive model has the lowest RMSE and MAE, but the worst RMSPE, MAPE and correlation, while the seasonal model shows the best RMSPE, MAPE and correlation. In comparison, the other models have a much greater RMSE to MAE ratio, indicating worse performance during high prevalence relative to the naive model. ARGO does not outperform the benchmarks in this case.
This result seems to reflect the distinct case count pattern in Taiwan compared to the other countries. Taiwan experienced little to no dengue prevalence for years until two epidemic spikes occurred in 2014 and 2015. In contrast, the other countries experience seasonal fluctuations of dengue over their entire estimation windows. This lack of predictability may be the reason that both seasonal and Google Trends-based models have greater error than the naive model, significantly over-estimating the 2015 peak for example. Yet overall, these methods show greater correlation than the naive method, perhaps because they are more responsive. Because ARGO over-estimates to a greater extent than the autoregressive methodology, this again reflects previous observations on the tendency of Google data to overshoot.
ARGO dynamically adjusts weights of dengue time-series and Google Trends data to best fit the most recent dengue behavior (See Fig. A, B, C, D, and E in S1 Text).
Our findings confirm that combining historical dengue incidence information with dengue-related Google search data, in a self-adjusting manner, leads to better near real-time dengue activity estimates than those obtained with previous methodologies that exploit the information separately. This also confirms that the hidden Markov model framework used by ARGO is appropriate in this context [35].
ARGO’s uniform out-performance of other benchmark methods for Mexico, Brazil, Thailand, and Singapore demonstrates its robustness and broad applicability. ARGO achieves this by balancing the influence of Internet search data, which quickly change in the face of outbreaks, and auto-regressive information, which tempers the estimations to mitigate the problem of overshooting. The application of an L1 regularization approach [34, 35, 39] helps identify the query terms most relevant to estimation at any given time, providing easy-to-interpret information as shown in the heatmaps in Figures A, B, C, D, and E in S1 Text. ARGO dynamically trains on a two-year rolling window, allowing model parameters to adjust over time to account for changes in Internet users’ behavior. The success of our methodology is based on the intuition that the more people are affected by dengue, the higher the number of dengue-related searches will be during an outbreak, and therefore the more likely Google query information will be useful at detecting dengue activity. This is observed in our findings, where the median yearly dengue case counts are strongly associated with the performance of our methodology (i.e. the higher the median yearly cases the higher the correlation of ARGO), as shown in Table 2. This is consistent with earlier findings that dengue virus prevalence is correlated with model performance in sub-regions of Mexico [38]. In addition, in Brazil, Mexico, and Thailand, the countries where our methodology works best, a clear seasonal pattern is observed in the disease incidence trends over time, as shown in Table 2.
On the other hand, the results from Taiwan illustrate the limitations of our approach. Taiwan does not present either an observable seasonal trend or a high number of dengue cases. As a result, neither ARGO nor the model using only Google search terms reliably track dengue. Low dengue-related Internet search activity during most years and sudden public interest during the outbreaks of 2014 and 2015, causing mis-calibration of the Google Trends data, may be another contributor. Other unique characteristics of the Taiwan outbreaks are that they were largely localized in South Taiwan, where Aedes aegypti is resident, and featured viral strains from neighboring countries rather than endemic strains [45, 46]. Also of interest is that the increased case counts occurred during periods of significantly increased temperature and rainfall [46]. The unpredictable character of these outbreaks present challenges for the performance of ARGO, and generally of all the methods considered in our comparison, but also highlight the potential of incorporating environmental predictors such as temperature and precipitation in our approaches.
While Internet penetration may seem to be an important factor in assessing the quality of Google Trends data, the statistics from Table 2 show that it alone is not as effective as dengue prevalence or seasonality in predicting the overall performance of our methodology. As an example, although Taiwan has high Internet penetration, the dengue case count may be low enough over most years that dengue-related searches motivated by other medical or educational purposes may introduce significant noise in the Google-query data. On the other hand, ARGO shows strong improvement over the seasonal autoregressive model in Brazil and Singapore, two countries with moderate to high Internet access, compared to Mexico and Thailand, which have low Internet access, suggesting that web penetration is nevertheless still an important factor. Finally, the proportion of the population within a country using Google as a search engine also provides some insight into the performance of ARGO (Table 2). ARGO shows the lowest correlation in Taiwan, which happens to have the lowest Google market share among the countries studied here [42].
Despite dengue and flu having very different biological transmission patterns, the fact that modifications to the ARGO methodology yield robust and accurate dengue estimates indicates the strength of our methodological framework. Although the monthly time scale chosen for this study was originally chosen based on data availability, inspection shows that a monthly surveillance approach is better suited for the 2-week serial interval of dengue [47].
The dengue activity estimates obtained with our methodology, like estimates from any novel digital disease detection tool, are not meant to replace dengue information obtained from traditional healthcare-based disease surveillance; instead, they can help decision-makers confirm (or deny) suspected disease trends ahead of traditional disease surveillance systems. Ultimately, the goal of this effort is to take a step closer to the development of an accurate, real-time modeling platform, where dengue case estimates can be constantly updated to provide authorities and non-governmental organizations with potentially actionable and close to real-time data on which they can make informed decisions, as well as providing travelers visiting high-risk areas with warnings. Such a platform could bring multiple information sources together, including but not limited to traditional epidemiological case reports, Google searches, crowd-sourced data, and climate and transportation information, creating a rapid response and alert system for users based on their specific location. Timely and precise detection may turn out to play a large role in reducing infections in the near future by influencing the timing of vector control efforts, hospital and clinical preparation, and providing public and individual alerts.
The platform would also enable users to verify dengue risk information with their own observations, creating a positive feedback loop that would continuously improve the accuracy of the tool. We are currently implementing two building-blocks that could help shape such a platform. The first one consists of a webpage Healthmap.org/denguetrends where dengue estimates produced with the methodology introduced in this manuscript are continuously displayed, and the second one is a crowd-sourced tool (currently in beta) that offers a user-friendly online chat system which maps dengue cases worldwide, and gives the public free access to toolkits that help reduce their risk of infection. This second effort is led by Break Dengue’s “Dengue Track” initiative www.breakdengue.org/dengue-track/. The potential impact may be far reaching, as the same models could also be used to track and map other infectious and mosquito-borne diseases, like Zika, malaria, yellow fever or Chikungunya.
Real-time implementation of our methods requires robust responses to changes in data quality, availability, and format. For example, Google correlate data shows internal variability attributed to re-sampling when the tool is accessed at different times. In addition, epidemiological data is not always published consistently by countries, creating lags in reporting that would make our methodology (which assumes having access to last month’s dengue case counts) not applicable.
In order to understand the impact of these data limitations, we performed two robustness studies of ARGO with respect to (1) the variations in Google Trends data, and (2) the availability of the most recent dengue case count data. For the first, we obtained multiple data sets containing the search frequencies of the query terms displayed on Table A in S1 Text by accessing Google Trends 10 different times during a week. We then produced Dengue activity estimates with ARGO using these 10 data sets as input. Table B in S1 Text shows that ARGO still outperforms all other methods in Brazil, Mexico, Thailand and Singapore, despite the random variations observed in Google Trends data. For the second, we retrained all the models under the assumption that the dengue case count from the past month was never available due to reporting delays. Table C in S1 Text shows that despite the unavailability of the last month dengue counts, ARGO had competitive predictive performance in the five countries/states when compared to other models (similar to the full data case), suggesting that our methodology is robust to the time delays in reporting in addition to variations in the input variables.
While our methods are designed to self-correct over time, the introduction of an intervention to curb dengue activity that could lead to a reduction in dengue cases, such as vector control or behavioral education (e.g. use of bed nets), may potentially lead our models to temporarily over-predict incidence. However, once such an intervention has been established and remains active in a given location, our models will self-correct over time to predict the new levels of dengue activity. Sporadic, nation-wide mosquito control methods would provide a bigger challenge to dengue case count predictability and, therefore, our model’s usability.
In light of ARGO’s strengths and limitations, future work should analyze the feasibility of applying our methodology to other countries, finer spatial resolutions, and temporal resolutions. This will be followed by routine reassessments of our methods to identify changes in information or potential improvements, including new search terms. As an example of such a change, Brazil has started publishing weekly dengue case counts since 2014. While our work used only the monthly resolution for fair comparison among all countries, adapting our methods to shorter time horizons for regions that provide such information would be useful.
Information on national-level dengue activity may not be ideal for decision-making at the local level since this information has been aggregated over a wide variety of potentially heterogeneous spatial environments. Future work should explore finer spatial resolution estimations to identify whether region-specific factors may improve or worsen results, similar to what has been done in [15, 38]. The five countries/states explored in this study vary on orders of magnitudes of size; for example, Brazil, Mexico, and Thailand each spans over 100 million square miles. As a result, these three countries contain wide ecological diversity and potentially varying patterns of dengue transmission among different sub-regions. It may be expected, for example, that Brazil would show different levels of seasonality in tropical compared to temperate areas. The success of finer spatial resolutions would depend on the quality of local case count and Google Trends data; the former can be affected by reporting efficiency, and the latter can be subject to Internet availability and Google use in a given region. Using national level data, on the other hand, has the advantage of smoother incidence curves for extraction and extrapolation of signal at the cost of more granular information. This is reflected in the observation that ARGO performed best in the three large countries despite the inherent heterogeneity within each country. This fits with our previous observation that a combination of higher dengue prevalence at the national level, seasonality and Google use in these countries leads to better results. We believe that these strengths and limitations also apply to extending our methodology to other countries/states besides those studied in the paper.
Producing short-term forecasts of dengue activity, in addition to the nowcast presented here should also be pursued (See [48] for such an extension for flu forecasting). Our approach may help produce dengue activity estimates in higher spatial resolutions that can lead to alert systems for people with an increased risk of exposure to the dengue virus at any given point in time. It is important to keep in mind that state-level or city-level spatial scales with low dengue activity may present similar challenges to the applicability of our approach as seen in Taiwan. The incorporation of other Internet-based data sources [48, 49] and cross-country spatial relationships should also be exploited in order to improve the accuracy in predictions.
|
10.1371/journal.pntd.0006762 | Contact tracing performance during the Ebola epidemic in Liberia, 2014-2015 | During the Ebola virus disease (EVD) epidemic in Liberia, contact tracing was implemented to rapidly detect new cases and prevent further transmission. We describe the scope and characteristics of contact tracing in Liberia and assess its performance during the 2014–2015 EVD epidemic.
We performed a retrospective descriptive analysis of data collection forms for contact tracing conducted in six counties during June 2014–July 2015. EVD case counts from situation reports in the same counties were used to assess contact tracing coverage and sensitivity. Contacts who presented with symptoms and/or died, and monitoring was stopped, were classified as “potential cases”. Positive predictive value (PPV) was defined as the proportion of traced contacts who were identified as potential cases. Bivariate and multivariate logistic regression models were used to identify characteristics among potential cases.
We analyzed 25,830 contact tracing records for contacts who had monitoring initiated or were last exposed between June 4, 2014 and July 13, 2015. Contact tracing was initiated for 26.7% of total EVD cases and detected 3.6% of all new cases during this period. Eighty-eight percent of contacts completed monitoring, and 334 contacts were identified as potential cases (PPV = 1.4%). Potential cases were more likely to be detected early in the outbreak; hail from rural areas; report multiple exposures and symptoms; have household contact or direct bodily or fluid contact; and report nausea, fever, or weakness compared to contacts who completed monitoring.
Contact tracing was a critical intervention in Liberia and represented one of the largest contact tracing efforts during an epidemic in history. While there were notable improvements in implementation over time, these data suggest there were limitations to its performance—particularly in urban districts and during peak transmission. Recommendations for improving performance include integrated surveillance, decentralized management of multidisciplinary teams, comprehensive protocols, and community-led strategies.
| Contact tracing is comprised of three main steps: identifying, listing, and monitoring persons who have been exposed to infected individuals, with the goal of rapidly diagnosing and treating new cases and preventing further spread of infection. This approach has been used to control transmission of infectious diseases including smallpox, tuberculosis, HIV, and syphilis, and while contact tracing has been used in prior outbreaks of hemorrhagic fever, these outbreaks were small in scale. During the 2014–2015 Ebola virus disease (EVD) epidemic in Liberia, contact tracing was implemented in all 15 counties on a scale that was unprecedented, particularly within both rural and crowded urban settings. This work provides insight into the magnitude that which contact tracing was implemented, its characteristics, as well as an assessment on its performance. Given that contract tracing is a critical tool for controlling disease spread, these findings aid in informing future planning and decision making for its implementation.
| In March 2014, Liberia detected its first cases of Ebola virus disease (EVD) in Lofa, a northern county bordering Guinea and Sierra Leone [1]. The Liberian Ministry of Health (MOH) (formerly Ministry of Health and Social Welfare) established a national task force and initiated control efforts, including contact tracing [1, 2]. As the epidemic grew, the task force developed into an Incident Management System, which oversaw contact tracing in all 15 counties with support from international partners including World Health Organization (WHO), U.S. Centers for Disease Control and Prevention (CDC), and Action Contre la Faim [3, 4, 5]. Continuous, widespread transmission continued until February 2015 [6], and 42 days after the last confirmed case had two negative samples in March 2015 [7], Liberia was declared free of Ebola on May 9, 2015—marking an end to the epidemic [8].
Contact tracing is comprised of three main steps: identifying, listing, and monitoring persons who have been exposed to infected individuals, with the goal of rapidly diagnosing and treating new cases and preventing further spread of infection. This approach has been used to control transmission of infectious diseases including smallpox, tuberculosis, HIV, and syphilis [9, 10, 11, 12]. Although contact tracing has been used in prior outbreaks of hemorrhagic fever, these outbreaks were small in scale [13, 14]. Contact tracing is most efficient for diseases with low incidence, limited transmissibility [15, 16], tight networks, and an incubation period long enough to allow intervention. Conversely, the effectiveness and optimal levels of investment for contact tracing, particularly for emerging diseases and for acute epidemics, are subjects of ongoing research and debate [15, 16, 17, 18, 19].
The 2014 EVD epidemic was the largest EVD outbreak in history and the first known EVD outbreak in West Africa [20]. The scale of the control efforts including contact tracing was unprecedented, particularly within both rural and crowded urban settings, which burdened existing surveillance capabilities and required immense commitment and cooperation on the part of government and the affected communities themselves. Furthermore, the strategies and implementation of contact tracing in Liberia evolved—from establishing operations to scaling them up—in order to respond to the changing phases of the epidemic. These aspects warrant the need to further examine contact tracing within this unique context. Here, we describe the scope and characteristics of contact tracing in Liberia and explore its performance during the 2014–2015 EVD epidemic in order to inform future contact tracing strategies in large-scale epidemics.
We performed a retrospective descriptive analysis of data collection forms for contact tracing that was conducted for the EVD epidemic in six of Liberia’s 15 counties during June 2014–July 2015. The six counties consisted of both rural and urban areas and represented 72% of the population of Liberia [21]. Three of the counties (Lofa, Bong, and Nimba) are at the border with Cote d’Ivoire, Guinea, or Sierra Leone, while the other three (Montserrado, Margibi, and Sinoe) extend from central areas to the coast. Additionally, both formal and informal sources of information regarding contact tracing organizational structures and implementation within these counties were reviewed to help provide context for the data analysis.
A contact was defined as a person who had direct or indirect exposure to any confirmed, probable, or suspect EVD case, or bodily fluids of a case, within the past 21 days [22, 23]. This definition also included any persons who had been discharged from an Ebola Treatment Unit (ETU) as not a case, due to their potential exposure to the virus while in the ETU.
National contact tracing guidelines and forms, which were initially adapted from existing WHO and CDC materials and finalized during the waning days of the epidemic, were used as the foundation for implementing the three steps of contact tracing: contact identification, listing, and monitoring. Once a case was detected, contact identification and listing were conducted by interviewing the case and/or family members to gather an initial list of potentially exposed persons. In most instances, this process was conducted by case investigation teams, which were distinct from contact tracing teams, and any of the following six types of exposure were added: (1) sleeping or eating in the same household; (2) direct physical contact with the body; (3) touching bodily fluids; (4) manipulating clothes or other objects; (5) through breastfeeding; and (6) attending a case’s funeral.
Contact tracers, chosen from within the community, located the listed contacts and identified any additional contacts missed in the initial investigation. Contact tracers transferred the information collected by case investigation teams to paper forms, including the contact’s name, county, district, town, and exposure(s). The name, age, location, and unique case identifier of the case for which contacts were listed, i.e. the “source case”, were also recorded.
During contact monitoring, contact tracers were expected to visit contacts twice daily (morning and afternoon) for 21 days post-exposure in order to identify and record whether the contact had EVD symptoms. This was determined initially through self-reports and physical observation, and eventually temperature readings were added for more objective monitoring. Contacts were monitored for nine symptoms: joint pain, fever (>38° Celsius), weakness, nausea, diarrhea, headache, throat pain, red eyes, and mucosal bleeding.
Following the outbreak, paper contact tracing forms were requested from all County Health Teams. Forms were received from six counties and the data were entered into a Microsoft Access database. Data were analyzed using Microsoft Access and Epi Info. Each form was considered a unique contact record and unit of analysis, though it was possible for individual contacts to be monitored more than once if re-exposed.
Source cases were identified using unique case identifiers, name, age, county, and district. To assess the coverage of contact tracing, or the percentage of cases for which contacts were monitored, we calculated the ratio of source cases in the database to the total number of suspected, probable, and confirmed EVD cases in MOH situation reports for the same counties using the closest approximate dates [24, 25, 26, 27, 28]. The mean number of contacts per source case was presented as contact-to-case ratios.
We analyzed records by county and district. Urban or rural classifications were assigned based on districts; districts that hold the county headquarters or that have settlements with a population of 5,000 or more persons were classified as urban [21]. Two districts, one each in Nimba and Montserrado counties, were divided into urban and rural sub-districts. Each of the six exposure categories and nine symptoms were analyzed, and medians and interquartile ranges (IQR) were calculated.
We divided the timeframe into four phases based on the observed epidemic trends of cases within Liberia [5], per epidemiologic week (EW): “Phase 1”: the initial increase of cases, from June to mid-August 2014 (EW 22–33); “Phase 2”: the peak, from mid-August to mid-November 2014 (EW 34–46); “Phase 3”: a decline in the epidemic, from mid-November 2014 through February 2015 (EW 47–9); and “Phase 4”: sporadic clusters, from March through July 2015 (EW 10–31). The first date of contact monitoring or last date of exposure was used to categorize records by phase. Medians and IQRs for timeliness, determined by the difference between the last date of exposure and first date of follow-up, were calculated and stratified by urban-rural and phases.
Each record was assigned one of seven outcomes of monitoring, either designated on the form or imputed using supplemental information: (1) “completed” the monitoring period of 21 days post-exposure; (2) “dropped” if the source was determined to be not a case; (3) “lost to follow-up” if the contact could not be located after three consecutive days; (4) “potential cases” if the contact presented with symptoms and/or died and monitoring was stopped; (5) “restarted” if monitoring was reinitiated due to a new exposure; (6) “transferred” if the contact moved to another jurisdiction; or (7) “unknown” for all remaining contacts with no outcome information. Contacts who presented with symptoms could be referred for medical evaluation without meeting EVD case definitions; hence, we use the terminology “potential cases”.
We calculated the positive predictive value (PPV) defined as the proportion of traced contacts—excluding those with dropped and unknown outcomes—who were potential cases. Sensitivity was defined as the ratio of potential cases identified during monitoring to the number of new cases in situation reports in the same counties [24, 25, 26, 27, 28]. This analysis assumes all potential cases were infected with EVD, and that all source cases and potential cases in the database were included in the total counts from situation reports. Therefore, to the extent that these assumptions are overstated, the calculations serve as upper limit estimates. PPVs were stratified by urban-rural and epidemic phases, whereas sensitivity and coverage were stratified by phases.
We used odds ratios and 95% confidence intervals to examine exposure types, symptom types, phases, and urban-rural amongst potential cases compared with contacts who completed monitoring; for ordinal variables, the lowest category was used as a reference group. Chi-square tests with p-values <0.05 were statistically significant. Two multivariate logistic regression models were used: (1) urban-rural, phase, and exposure type covariates, limited to records with ≥1 exposures, and (2) urban-rural, phase, and symptom type covariates, limited to records with ≥1 symptoms. Only statistically significant variables in bivariate analysis were included in the models. Nonparametric tests were used for continuous variables.
This assessment is included under Johns Hopkins School of Public Health Institutional Review Board no. 6296 with DHP as principal investigator. A letter of agreement was signed with the Liberia MOH concerning the publication of contact tracing analyses. This assessment used retrospective data collected for public health surveillance purposes so informed consent was deemed unnecessary according to the U.S. Common Rule. We followed the Declaration of Helsinki, aiming to provide assurance that the rights, integrity, and confidentiality of participants were protected.
We analyzed 25,830 records for contacts who had monitoring initiated or were last exposed between June 4, 2014 and July 13, 2015 in the six counties. Of these, 25,651 contacts were listed for 2,465 source cases; an additional 179 contacts had no source case provided. The overall contact-to-case ratio was 10:1 (median = 7, range 1–424). The contact-to-case ratio increased with each subsequent phase and was higher in urban than rural districts. There were 9,241 EVD cases in situation reports in the six counties. The upper limit estimate of coverage, or the maximum percentage of cases for which contacts were monitored, was 26.7%, and was lowest during Phase 1. (Table 1)
In the six counties providing data, 89.0% of the records were identified in Montserrado County, 8.6% in Margibi, 1.6% in Bong, 0.4% in Lofa, 0.4% in Sinoe, and 0.1% in Nimba (Fig 1) (Table 2). Records pertained to 22 of Liberia’s 136 districts (Table 2); data from the remaining districts was unavailable due to no contact tracing records or no reported EVD cases. In total, 21,500 (83.2%) contacts were in seven urban districts/sub-districts, mainly in the Monrovia capital district in Montserrado, while 4,327 (16.8%) contacts were in 17 rural districts/sub-districts. Potential cases were less likely to be from urban districts (Table 3).
Temporal trends for contact tracing aligned with disease transmission trends (Fig 2). For 25,690 records grouped by phase, 61.7% were monitored during Phase 2 and 32.9% during Phase 3. Only contacts in Montserrado were monitored during Phase 4. No contacts were monitored during May–June 2015, corresponding to the Ebola-free period. Based on 25,300 records, contact tracing was timelier in rural districts; overall, the median difference was 1 day (IQR 0–4) (Table 1).
Of 25,830 total contacts, 17,876 (69.2%) contacts reported 34,284 exposure types, and 7,954 (30.8%) had zero exposures recorded. Among the 17,876 contacts reporting any exposure, direct physical contact with the body was the most common (73.1%), while funeral attendance (2.1%) and breastfeeding (0.2%) were the least common (Table 3). Two or more exposure types were reported in 54.9% of 17,876 contacts; the median was 2 (IQR 1–3). Multivariate analysis showed the odds of sleeping or eating in the same household, direct physical contact, or touching bodily fluids were higher amongst potential cases than contacts completing monitoring (Table 3).
Of 25,569 contacts with an assigned outcome, 22,680 (87.8%) completed monitoring, 1,768 (6.8%) were dropped, 637 (2.5%) restarted, 334 (1.3%) were potential cases, 136 (0.5%) were lost to follow-up, and 14 (0.1%) were transferred. Most contacts completed monitoring during each phase except during Phase 4, when 53.6% of contacts were dropped (Fig 2). More contacts restarted during phases 2 and 3 than other phases. Potential cases were less likely to be monitored during phases 2 or 3 compared to Phase 1 (Table 3). Twenty-two contacts were not located prior to monitoring. Of 46 recorded contact deaths, 56.5% were in urban districts and 33 occurred during monitoring (15 after taken to an ETU).
The PPV was 1.4% overall, and was higher in rural (3.0%) than urban (1.1%) districts and highest during Phase 1 (4.7%), after which it decreased for subsequent phases. The sensitivity of monitoring, or the maximum proportion of new cases detected, was 3.6%, and was highest during phases 1 and 2. (Table 1)
Table 4 shows the distribution of reported symptoms. Overall, 326 contacts reported 1,299 symptom types and 3,732 symptom-days; the median symptom types per contact was 4 (IQR 2–5). Contacts of all outcomes reported symptoms except transferred contacts; 218 (66.9%) of 326 contacts reporting symptoms were potential cases, 92 (28.2%) completed monitoring, 6 (1.8%) restarted, 6 (1.8%) were unknown, 3 (0.9%) were dropped, and 1 (0.3%) was lost to follow-up. In multivariate analysis, potential cases were more likely to report fever, nausea, or weakness compared with contacts who completed monitoring.
During 2014–2015, more than 25,000 persons in six of Liberia’s 15 counties were identified, listed, and monitored for EVD, representing one of the largest contact tracing efforts during an epidemic in history. Nationwide, these efforts were even more substantial and required the dedication of responders, including the Government of Liberia, counties, and contact tracing teams. As a result, 334 contacts were identified as potential cases with the intention of providing earlier treatment and preventing hundreds of new infections.
Relative to the scale of these efforts, however, these data suggest there were limitations to the performance of contact tracing within Liberia. Overall, there was a small proportion of monitored contacts that were identified as potential cases, and more than 97% of reported EVD cases from the six counties were not detected through contact monitoring. This is greater than expected, especially compared to other examples in West Africa where approximately 69% to 78% of cases were not being traced prior to case identification [29, 30]. This measure is dependent upon the level of contact tracing coverage, and based on our database, though admittedly not comprehensive, coverage only accounted for a maximum of one-quarter of all EVD cases reported for these six counties. While this ratio is aligned with similar findings in two Guinea prefectures (32% and 39%) and in Sierra Leone (19%) [29, 30], it is possible that contact tracing was not initiated for up to three-quarters of the remaining EVD cases in Liberia, potentially due to a combination of factors discussed below.
Potential cases were more likely to be identified in rural districts and early in the epidemic, despite intensified efforts as the epidemic progressed. Possible explanations for why contact tracing was less effective in urban areas could include the following: higher population density and complex social networks making it more difficult to identify all contacts; less cooperation within urban settings; higher burden and strained resources; or a combination of these factors. These results support the concept that contact tracing is most successful when transmission is low, and models have shown that expanding implementation of contact tracing yields diminishing reductions in disease prevalence [15, 16]. Therefore, it is critical to conduct contact tracing rigorously and comprehensively as soon as an outbreak is identified, and to achieve higher sensitivity and coverage during this phase.
There were, however, notable improvements in implementation over time; specifically, greater coverage, fewer contacts lost to follow-up, and higher contact-to-case ratios. During Phase 4, Liberia was able to focus more resources on eliminating the last transmission chains, including expanding the inclusion criteria to ensure no new cases went undetected [6]. This would have resulted in a larger contact-to-case ratio during Phase 4 compared to all other phases.
The dynamics of contact tracing are complex, and its success is related to characteristics of the disease and etiologic agent, resources, and socio-political factors that influence its acceptability and implementation. Additionally, the approaches to contact tracing may differ depending on whether there is a vaccine or therapy available. Given that contact tracing remains one of the critical public health tools during outbreaks involving person-person transmission, optimizing its performance is paramount. While not exhaustive, we focus on four key challenges that may have limited the performance of contact tracing for EVD within Liberia, and propose recommendations for future efforts.
First, an integrated surveillance and data management system was lacking and had to be established for reporting between the national laboratory, healthcare facilities and ETUs, and contact tracing and case investigation field teams [5]. Consequently, contact tracing was less functional at the beginning of the epidemic when it could have been most effective in slowing the epidemic. Initially, an insufficiently integrated system resulted in missed source cases and contacts, and led to delays in monitoring; this is reflected in that 25% of contacts with available information started monitoring four days after their last exposure. Additionally, contacts were listed and needlessly traced because of delays in receiving negative laboratory results, thereby lowering the PPV. Although mobile applications had the potential to improve reporting and data management, these were not piloted until after the peak of the outbreak. In contrast, contact tracing in urban Nigeria successfully and rapidly contained EVD transmission, largely thanks to robust surveillance systems and leveraging mobile applications for real-time monitoring [31, 32]. Strengthening integrated surveillance and electronic data systems, and the early adoption of mobile technology, could help improve timely reporting for listing and monitoring contacts.
Secondly, the organizational structure for contact tracing likely led to inefficiencies in its implementation and management, particularly in urban districts. For instance, case investigation teams, who conducted contact listing, were often distinct from contact tracing teams who conducted contact monitoring. In some rural areas, teams responded in tandem thereby reducing gaps, yet this was more difficult in dense urban areas such as in Montserrado County. Additionally, the county level coordinated all aspects of the response—not just contact tracing. In January 2015, Montserrado created decentralized sub-county sectors to oversee and synchronize all operations—a change previously recognized as a critical step for halting transmission [6]. Particularly in urban areas and in the absence of a robust surveillance system, using a decentralized management approach and multidisciplinary teams may improve contact tracing performance.
Thirdly, there were challenges with adapting and implementing contact tracing protocols, which had to be used by novice teams during the epidemic. For instance, the number of contacts per source case ranged widely in our analysis, from 1 to 424, and nearly one-third of contacts had no exposure documented, indicating that some contacts may not have met the inclusion criteria, thereby straining resources. Also, written guidance for identifying potential cases during monitoring did not specify how contact tracers should determine when to refer a contact for medical evaluation [23]. Eighteen contacts, who presumably would have shown symptoms prior to death, died during monitoring without being referred for medical evaluation. Among contacts who reported symptoms, including multiple symptoms and symptom-days, 33.1% continued under monitoring without being referred for further evaluation, indicating that triggers for identifying potential cases was subjective. During future outbreaks, clear and comprehensive protocols need to be initiated early in the epidemic and reinforced throughout implementation. Furthermore, if resources are limited, inclusion criteria could prioritize contacts with multiple exposures, and/or those with household contact or direct contact with the body or bodily fluids. Triggers for identifying potential cases could include contacts reporting multiple symptoms types, fever, nausea, and weakness.
Finally, community perceptions, stigma, and mistrust reportedly led to challenges in obtaining complete and reliable information, to delays or an inability to trace contacts due to evasion, and even to violence [5, 33]. Underreporting of symptoms due to fear or due to fever-reducing drugs may explain why relatively few symptoms were captured in our database. Also, contacts were instructed to self-isolate within their home, which disrupted normal routines and the ability to maintain jobs; without adequate support from the community or organizations, contacts are less likely to cooperate. These aspects stress the importance of community cooperation, trust, and engagement. Overall, less than 1% of contacts were lost to follow-up, and this improved during each phase along with more contacts listed per source case, suggesting that this cooperation probably improved as the outbreak progressed. For future outbreaks, community-led strategies for contact tracing should be an early priority to foster cooperation, trust, and ownership of the control efforts.
This analysis represented both urban and rural settings, and Montserrado specifically, where the response was most intense. However, we were unable to collect forms from all 15 counties nor all forms from the six inclusive counties; for example, no forms were available for the EVD cluster that occurred in Margibi in July 2015. Despite commendable efforts, counties reported that paper forms were lost or destroyed due to perceived contamination risks. Using paper forms also led to variability in data quality, including illegible writing, misspellings, inversed source case and contact information, and difficulty in interpreting marks for visits and the presence of symptoms. Falsifying information on forms was a concern [33], such as documenting visits when the contact had not been seen, and this was an issue early in the epidemic.
These factors, combined with the lack of information to ascertain the final status of EVD infection amongst source cases and potential cases, constrained our analysis. Likewise, we could not conclude whether symptoms reported amongst potential cases evidenced EVD infection. Our data primarily represented contact monitoring, as we did not have a comprehensive contact listing. Finally, EVD case counts from situation reports were unavailable to stratify coverage and sensitivity by urban-rural districts and/or phase, and these aggregated totals could not be linked to our individual-level database.
Our findings suggest that despite the unprecedented scale of contact tracing for EVD in Liberia, there were limitations in its ability to detect new cases, especially in urban areas and during the peak case load. Since contact tracing remains a critical intervention for controlling outbreaks, we suggest rigorous implementation early in the outbreak and focusing on four key areas to optimize its performance within similar contexts: (1) strengthening integrated surveillance and electronic data systems, (2) decentralizing management of multidisciplinary teams for improved coordination and oversight, (3) instituting and reinforcing clear and comprehensive protocols, and (4) adapting community-led strategies to foster cooperation, trust, and ownership.
|
10.1371/journal.ppat.0030159 | Small-Molecule Inhibition of HIV pre-mRNA Splicing as a Novel Antiretroviral Therapy to Overcome Drug Resistance | The development of multidrug-resistant viruses compromises antiretroviral therapy efficacy and limits therapeutic options. Therefore, it is an ongoing task to identify new targets for antiretroviral therapy and to develop new drugs. Here, we show that an indole derivative (IDC16) that interferes with exonic splicing enhancer activity of the SR protein splicing factor SF2/ASF suppresses the production of key viral proteins, thereby compromising subsequent synthesis of full-length HIV-1 pre-mRNA and assembly of infectious particles. IDC16 inhibits replication of macrophage- and T cell–tropic laboratory strains, clinical isolates, and strains with high-level resistance to inhibitors of viral protease and reverse transcriptase. Importantly, drug treatment of primary blood cells did not alter splicing profiles of endogenous genes involved in cell cycle transition and apoptosis. Thus, human splicing factors represent novel and promising drug targets for the development of antiretroviral therapies, particularly for the inhibition of multidrug-resistant viruses.
| Over the two decades highly active antiretroviral therapy (HAART) for the treatment of HIV infection has led to a significant decline in morbidity and mortality rates among HIV-infected individuals. HAART uses a combination of molecules that target the virus itself. However, naturally occurring and extensive genetic variation found in the virus allow the emergence of drug-resistant viruses, which rapidly render individuals untreatable. An alternative approach for effective anti-HIV-1 chemotherapy should be targeting cellular factors required for HIV-1 replication. Here, we report the development of a successful therapeutic agent for HIV-1 infection based on inhibition of HIV-1 pre-mRNA splicing, a crucial step of the HIV-1 life cycle that allows production of key viral proteins. The splicing inhibitor IDC16 can efficiently block HIV-1 viral production in primary blood cells infected with different laboratory strains or clinical isolates from patients resistant to anti-HIV multitherapies. These findings may serve as the basis for a new strategy to develop a new class of anti-HIV drugs, the splicing inhibitors, and even of antiviral drugs in general, since any virus needing to splice its RNAs may be targeted.
| The increasing prevalence of drug-resistant human immunodeficiency virus type 1 (HIV-1) has highlighted the challenging issue of the optimal treatment of HIV-1-infected patients [1–3]. Current routine drug regimens, typically consisting of various combinations of compounds targeting the viral proteins reverse transcriptase, protease, and gp120, have revolutionized the treatment of HIV/AIDS [3–5]. However, HIV-1 can acquire resistance to all known inhibitors of these targets, and transmission of multidrug-resistant HIV strains is becoming a growing problem [1,2,6–9]. This, as well as other problems such as viral escape mutants [4,10], persistence of viral reservoirs [11–14], poor patient compliance due to complicated regimens [15,16], and toxic side effects [17], have emphasized the need for the development of new drugs with novel mechanisms of action. In addition to virus-specific enzymes, such as reverse transcriptase and protease, several cellular factors are required for replication of HIV-1 [10]. The identification of these critical host cell factors may provide novel cellular targets for the development of compounds that are potentially capable of inhibiting HIV-1, thereby decreasing the burden of viral replication in cases of transmitted multidrug-resistant HIV-1 infection.
To express key viral proteins, HIV-1 uses a combination of several alternative 5′ and 3′ splice sites to generate more than 40 different mRNAs from its full-length genomic pre-mRNA [10]. The choice of these alternative splice sites strongly depends on specific interactions between HIV pre-mRNA sequences and non-spliceosomal nuclear RNA-binding proteins (trans-acting factors). These trans-acting factors can be classified as SR proteins (serine-arginine-rich proteins) [18–21] and hnRNPs (heterogenous nuclear ribonucleoproteins) [22–25]. Binding of the SR proteins to exonic splicing enhancers (ESEs) downstream of the Tat-, Rev-, Vpr-, Env-, and Nef-specific 3′ splice sites promotes exon definition by recruiting constitutive factors and preventing the action of nearby splicing silencers [26]. Therefore, members of the SR protein family are thought to play a major role in the regulation of HIV-1 pre-mRNA splicing.
By screening a large collection of chemical compounds, our laboratory has discovered several benzopyridoindole and pyridocarbazole derivatives that selectively inhibit the ESE-dependent splicing activity of individual SR proteins [27–29]. Selective binding of these compounds to specific SR proteins prevents spliceosome assembly and splicing in an in vitro system. Given the fact that these proteins play an important role in regulating HIV-1 pre-mRNA splicing, we surmised that specific inhibition of SR proteins by benzopyridoindole and pyridocarbazole derivatives could block HIV-1 replication.
Among 220 indole derivatives that were screened for selective inhibition of ESE-dependent splicing events, one selected drug (IDC16) demonstrated a strong inhibitory effect on different substrates harboring SF2/ASF high-affinity binding site (Figure 1A). IDC16 had also no adverse effect on cell growth and viability of different cell lines, including primary cell lines (see below). Given that SF2/ASF plays a major role in HIV-1 pre-mRNA splicing, we decided in the present study to test the effect of IDC16 on HIV replication as well as on specific splicing events of the HIV-1 pre-mRNA. The efficiency of the drug was first assessed by using the pΔPSP plasmid containing the HIV-1 proviral genome deleted between nucleotides 1511 and 4550 (Figure 1B), which recapitulates all splicing events of HIV-1 pre-mRNA in transfected HeLa cells [30]. The mRNAs produced by splicing were then analyzed by reverse transcriptase (RT)-PCR using forward and reverse primers that amplify several splicing isoforms encoding the viral proteins Nef, Rev, and Tat. Compared to untreated cells, the synthesis of all splicing products was less efficient (compare lane 7 and lanes 1–6). The effect of IDC16 was dose dependent, especially for the synthesis of larger splicing isoforms, with a complete block at 2.5 μM. At this concentration of the drug neither global nor HIV-1 RNA synthesis were affected (unpublished data, see below), implying that IDC16 has a specific action on HIV-1 pre-mRNA splicing.
The dose-dependent profile of splicing inhibition indicated that IDC16 inhibits the use of several weak 3′ splice sites whose utilization is required for the production of key viral regulatory proteins. Utilization of these 3′ splice sites is known to critically depend upon the binding of SR proteins [31]. To examine the specificity of IDC16, we tested the effect of this drug in an in vitro splicing assay using a pre-mRNA substrate (HIV1-D1-A2) harboring the D1 and A2 HIV donor and acceptor splice sites, respectively (see Figure 2A, [32]). Splicing of this substrate is activated by the SR protein ASF/SF2, both in vitro and ex vivo, and therefore it constitutes an ideal target for IDC16, which inhibits most of SF2/ASF ESE-dependent splicing tested ([26] and unpublished data). Treatment with increasing concentrations of IDC16 inhibits splicing of HIV1-D1-A2 in a dose-dependent manner (Figure 2B). Complementation of the extract with recombinant SR protein ASF/SF2 strongly limited the IDC16 splicing inhibition (Figure 2C), whereas the addition of similar amounts of another SR protein, SC35, fails to do so (Figure 2D), demonstrating that IDC16 specifically impedes SF2/ASF-dependent splicing. Consistently, concentration up to 50 μM of IDC16 did not alter splicing of synthetic mRNA precursors derived from the adenovirus major late-transcription unit (Minx, Figure 2E), which is a single intron pre-mRNA not requiring ESE sequences in the second exon for efficient splicing.
Given the key role played by ASF/SF2 in the activation of several HIV-1 acceptor sites, it is expected that the treatment of infected cells with IDC16 might block HIV-1 replication. Initial experiments were designed to determine the concentration of IDC16 with minimal side effects on cell viability and cell cycle progression. Treatment of stimulated peripheral blood mononuclear cells (PBMCs) with 1 or 2.5 μM IDC16 next showed no effect on cell proliferation, as measured by tritiated thymidine incorporation into cellular DNA (Figure 3A). Since no adverse cellular effect was observed up to the 2.5 μM IDC16 concentration range, we then asked whether IDC16 could block HIV-1 replication in this potential therapeutic window. Stimulated PBMCs were infected with either NL4.3 (unpublished data) or Ada-M R5 (Figure 3B) HIV-1 strains and cultured for 14 d in the presence or absence of 0.1 μM, 0.5 μM, or 1 μM IDC16. At the indicated days, virus replication was determined by p24 antigen enzyme-linked immunosorbent assay (ELISA). Replication of both NL4.3 (unpublished data) and Ada-M R5 strains (Figure 3B) is very efficiently blocked in these primary cells following treatment with 1 μM of IDC16. Meanwhile, cell viability was not affected by the drug throughout the assay as shown in Figure 3C.
To generalize the effect of IDC16 on HIV-1 replication in other primary cells, the same protocol was repeated using infected macrophages, which act as viral reservoirs. Cells were treated with 0.1 μM, 0.5 μM, or 1 μM of IDC16 and p24 antigen levels were monitored both in culture supernatant (Figure 4A) and in cell lysates (Figure 4B). Again, IDC16 blocked virus replication efficiently and in a dose-dependent manner, reaching inhibition levels up to 90% in primary macrophages at 1 μM. However, cell viability was not decreased under IDC16 treatment (Figure 4C). It is important to note that macrophages survival was rather increased in the presence of IDC16 at day 7 and day 14 post-infection (Figure 4C), suggesting that IDC16 may protect these cells from deleterious effects induced by viral infection.
The previous experiments were all performed with primary human cells infected with either macrophage-tropic (R5) HIV-1, Ada-M, or T-cell-line-tropic (X4), NL4–3 HIV-1 laboratory strains, suggesting that IDC16 could be effective on a variety of viral strains detected in vivo. We next used an in vitro system that may be more relevant to the clinical situation, since it involves infecting primary cells with HIV-1 isolates from patients resistant to conventional antiretroviral therapies. We chose five extreme cases in which viruses harbored mutations in different regions of the viral genome, including those encoding the reverse transcriptase and protease domains. Using two strains that previously demonstrated robust resistance to different therapeutic agents in vitro, we show that IDC16 also very efficiently inhibits the replication of these clinical isolates, since no viral particles were detected 14 d post-infection (Figure 4D).
In order to provide further evidence that the anti-HIV activity we observed with IDC16 is actually the consequence of its inhibitory effect on viral RNA splicing, we examined various steps of the viral cycle in cells treated with the drug and submitted to one-round infection. For this purpose, we exposed HOS-CD4+-CCR5+ cells to defective virions obtained by cotransfecting 293T cells with a plasmid encoding the R5 envelope of the AD8 strain and another plasmid containing the entire HIV-1 genome mutated in the envelope gene and harboring a luciferase marker gene fused to nef (Figure 5A, [33]). The amounts of luciferase activity in cells infected with these virions reflect both the number of integrated proviruses and expression of multiply spliced species encoding nef/luc (Figure 5A). Two days post-infection, cells were lysed and the amount of early reverse transcriptase products (strong-stop fragment), of late reverse transcriptase products (LTR-gag fragment), and of integrated proviral DNA (Alu-LTR fragment) was measured by quantitative PCR. At all concentrations tested, the drug had no effect on early reverse transcription, late reverse transcription, or integration (Figure 5B). In contrast, Figure 5C shows the dose-dependent inhibition by IDC16 of luciferase activity in HOS-CD4+-CCR5+-infected cells. Of note, the inhibitory effect could be smaller in this one-round infection assay than in the previous assays, where several rounds of infection were carried out.
To evaluate the effect of IDC16 on the expression of unspliced and multiply spliced HIV-1 RNA species in HOS-CD4+-CCR5+-infected cells, we used real-time RT-PCR to analyze the changes in mRNA levels. The results (Figure 5D, left panel) show that treatment of infected cells with 0.1 μM and 1 μM concentrations of IDC16 did not change the level of unspliced mRNA species using GAPDH mRNA as a reference. A significant increase of these species was, however, observed when cells were treated with 0.5 μM of IDC16. In sharp contrast, IDC16 treatment induced a dose-dependent decrease of multiply spliced species (Figure 5D, right panel), confirming that the drug actually inhibited HIV RNA splicing. Also, consistent with the current view that the production of full-length HIV-1 RNA requires factors encoded by multiply spliced species, like Rev and tat, the large decrease of multiply spliced species induced by IDC16 treatment has not been compensated by a similar increase of unspliced HIV-1 RNA.
All drugs that inhibit viral production also inhibit cell–cell fusions; we therefore assessed the capacity of IDC16 to inhibit the fusion of HIV-1-infected cells compared to azidothymidine (AZT, 3′-azido-3′-deoxythymidine, zidovudine). AZT is the first nucleoside reverse transcriptase inhibitor approved for HIV-1 therapy [34]. Its antiretroviral activity is likely to involve at least two steps: incorporation into viral DNA and inhibition of the viral reverse transcriptase. While incorporation of the drug into host nuclear and mitochondrial DNA may be largely responsible for dose-limiting toxicities, AZT remains a potent and frequently prescribed antiretroviral therapy for HIV-positive individuals. We therefore subjected both AZT and ICD16 to HIV-1 inhibition tests in infected MT2 cultures to compare their relative antiretroviral effects. MT2 cells cultured in a 96-well plate were infected with pNL4.3 at 100 TCID50 in the absence or the presence of IDC16 or AZT for 18 h. Cells were then washed and changed to fresh medium with or without IDC16 or AZT. Half of the culture medium was refreshed each 2 or 4 d in the presence of drugs. The formation of syncytia was scored at the indicated time points. Results illustrated in Figure 6A and 6B show that viral infectivity is completely abolished using five times less IDC16 as opposed to AZT. In fact, at a concentration of 1 μM, IDC16 is able to reduce the percentage of infected cultures to nearly nil, whereas 5 μM of AZT were needed in order to observe the same effects, indicating that effective inhibitory concentration within cellular systems of compound IDC16 is in the range of that of AZT.
Since we are promoting SR proteins as an attractive intracellular target of anti-HIV therapies, it was necessary to demonstrate that the antiviral effect of compound IDC16 by inhibiting SR proteins did not have a global effect on the splicing of endogenous genes. To address this question and to determine the general impact of the compounds on alternative splicing, we selected a set of 96 alternative splicing units covering a variety of human apoptotic genes. RT-PCR analysis on RNA extracted from triplicate PBMCs treated or not with compound IDC16 or AZT revealed the existence of multiple amplification products for 81 of them. The relative abundance of these different products was not affected by compound IDC16, suggesting no large impact on alternative splicing (the full data can be accessed through http://www.lgfus.ca/Tazi/, username = Tazi, password = sc35). Among this set, however, three displayed reduced levels in the presence of AZT (Figure 6C): breast and ovarian cancer susceptibility (BRCA1, see Table 1), human homolog of double minute 2 (HDM2, see Table 1), and death inducer-obliterator 1 (DATF1, see Table 1). Thus, AZT, an antiretroviral agent with proven clinical benefit for the treatment of HIV/AIDS, might have a detrimental effect on cell survival since the genes whose expression is altered in AZT-treated cells play a critical role in maintaining genome integrity [35–37]. These results are unexpected because AZT specifically targets HIV-1 reverse transcriptase, and therefore it is not supposed to alter the expression of endogenous genes. Hence, through analysis of a small fraction of selected genes, it was possible to show that AZT but not IDC16 alters host cell gene expression.
Taken together, our data indicate that IDC16 can efficiently block HIV-1 viral production in PBMCs or macrophages infected with different laboratory strains or clinical isolates from patients resistant to anti-HIV multitherapies. As IDC16 acts on a cellular factor, the risk of resistance development should be reduced. Indeed, to escape from IDC16 inhibition, a strain should have to mutate its ESE into a new one, requiring another SR protein than SF2/AF. IDC16 could thus represent the first member of a new class of anti-HIV drugs, the splicing inhibitors, and even of antiviral drugs in general, since any virus needing to splice its RNAs may be targeted.
To help develop this novel class of anti-viral inhibitors, several aspects of SR functions need now to be considered. First, we will have to address the phosphorylation status and/or specific localization of SR proteins in response to drug treatment. Indeed, indole derivatives have been shown to bind directly to the RS domain of SR proteins and thereby impede its phosphorylation, at least in vitro, by SR kinases [27–29]. Since phosphorylation affects both splicing activity and subcellular trafficking of SR proteins [38], treatment of cells with IDC16 may modulate these two processes and act synergistically to modify HIV-1 RNA splicing and/or export. Second and most important, we will have to determine whether IDC16 will affect additional functions that SR proteins have during gene expression, like mRNA export [39–42], mRNA stability [43], stimulation of mRNA translation [44], or maintenance of genomic stability [45]. Our current studies are now aimed at confirming in vivo lack of deleterious side effects of this molecule by comparing the profiles of SF2/ASF depletion and inhibition by IDC16. The finding that IDC16, unlike SF2/ASF depletion, could exhibit a low cellular toxicity already indicates that this molecule is selective for some functions of SF2/ASF that are shared by other SR proteins. Along this line, it is noteworthy that most of the drugs selected in our initial in vitro splicing inhibition screen were not previously considered as good candidates for use in cancer therapy because of their low cytotoxicity [29]. Furthermore, IDC16 has more impact on HIV-1 alternative splicing than that of endogenous gene (present study). While IDC16-mediated splicing modulation has been tested only for few genes, a likely explanation for this difference of splicing inhibition behavior could be that the viral RNA has to escape the splicing machinery during later stages of infection to produce viral particles containing full-length unspliced pre-mRNA, whereas most cellular genes have constitutive exons that contain redundant binding sites for SR proteins. A robust and comprehensive exon microarray that can detect with high accuracy alteration of splicing events is needed to consolidate this hypothesis. Available tools are still at a validation stage and are not sensitive enough to monitor the transcripts variation generated by splicing at large-scale to cover the expression of whole human genome.
Our results are consistent with the notion that cellular targets, like the SR proteins, can be used as potent targets to overcome drug resistance resulting from highly active antiretroviral therapy. Moreover, genes encoding cellular proteins do not mutate under physiological conditions, and one could expect that HIV-1 resistance to IDC16 would occur far less frequently than resistance to a conventional drug targeting viral proteins. More importantly, enhancer sequences bound by SR proteins are essential for efficient splicing, a prerequisite step to viral replication. Viruses that harbor mutations in these enhancers due to reverse transcriptase errors would have very little or no chance of survival. Similarly, mutations that improve the binding of SR proteins will also be detrimental for viral replication, as they will impede the production of full-length HIV-1 mRNA.
The Institut Curie–CNRS chemical library contains 6,720 molecules kept in 96 well microplates at a concentration of 10 mg/ml in dimethyl sulfoxide (DMSO). Extemporaneous dilutions were made with 10% DMSO. Microplates were kept at −20 °C.
Recombinant SF2/ASF and SC35 were produced and purified from baculovirus-infected Sf9 cells as previously reported [46].
The HIV1-D1-A2 plasmid has already been described [32]. In vitro transcription to obtain radiolabeled transcripts and splicing reactions were performed under standard conditions for 1 h as described [32] in the presence of the indicated concentration of IDC16. Splicing products were analyzed by electrophoresis on denaturing 7% polyacrylamide gels and revealed by autoradiography.
HeLa cells (5 × 105 cells) were grown in RPMI 1640 (GIBCO BRL), supplemented with 10% fetal calf serum (FCS) on 3-cm diameter dishes (Nunc) to 70%–80% confluence. Transient transfections with splicing reporter constructs (1 μg) were performed with the LipofectAMINE Plus reagent (Invitrogen) according to the manufacturer's instructions.
Total cellular RNA was isolated from transfected HeLa cells 48 h post-transfection, and 3 μg of RNA was reversed transcribed and PCR amplified with forward primer BSS (5′-GGCTTGCTGAAGCGCGCACGGCAAGAGG-3′; nt 700–727) and reverse primer SJ4.7A, which spans sites D4 and A7 (5′-TTGGGAGGTGGGTTGCTTTGATAGAG-3′; nt 8369–8381 and 6032–6044) [30]. To normalize the signals, GAPDH was used as an internal control of the PCR reactions as described [27]. Amplification products were radiolabeled by performing a single round of PCR with the addition of 10 μCi of [α-32P]dCTP, and the products were analyzed by electrophoresis on 6% polyacrylamide 8 M urea gel as described [30].
Human PBMCs from healthy donors were isolated by Ficoll-Paque density centrifugation. They were then cultured in RPMI plus 1% heat-inactivated human serum AB at a concentration of 2.5 × 106 cells/ml and then incubated at 37 °C, 5% CO2 for 1 h in 24-well plates. After removing the non-adherent cells, the adherent cells were kept in complete RPMI with 10% FCS at 37 °C for another 2 h. Ten U/ml GM-CSF (Roche) was added to the culture medium and incubated for 4 d before viral infection.
For HIV-1 infection, human PBMCs and macrophages were infected with 100 TCID50 of Ada-M (R5 strain) for 18 h in the absence or presence of various concentrations of IDC16 and then washed intensively with PBS. The culture medium and cells were collected at day 4, 7, and 14. Viral production was measured by the quantification of viral capsid protein p24 using ELISA (Beckman Coulter). Cell viability was quantified by trypan blue exclusion.
MT2 cells cultured in quadriplicate in a 96-well plate were infected with NL4–3 at 100 TCID50 in the absence or presence of IDC16 or AZT for 18 h. Cells were then washed and changed to fresh medium with/without IDC16 or AZT. Half of the culture medium was refreshed each 2 or 4 d in the presence of drugs. The formation of syncytia was scored at the indicated time points.
The plasmid pNL4.3-env−-Luc+ harboring a luciferase gene was co-transfected with the envelop plasmid pCMV-Ad8-Env (NIH AIDS Research and Reference Reagent Program) into human embryonic kidney cells-293T to produce R5-pseudotyped virions. Human osteosarcoma (HOS)-CD4+-CCR5+ cells (from the NIH AIDS Research and Reference Reagent Program) were treated with or without drug and infected for 18 h, washed three times with PBS, and kept in fresh RPMI-1640 containing 10% FCS with or without drug. Luciferase activity was monitored 48 h post-infection using a luciferase assay kit and a luminometer according the manufacturer's instructions (Promega).
At 48 h post-infection with Δenv-defective viruses, HOS-CD4+-CCR5+ cells were resuspended in lysis buffer (10 mM Tris [pH 8.0]; 0.5 mM EDTA; 0.0001% SDS; 0.001% Triton; 100 μg/ml Proteinase K), incubated 3 h at 50 °C and 10 min at 95 °C. For strong-stop and late reverse transcripts, DNA was amplified with the appropriate primers at 70 °C in a LightCycler (Roche) with SYBR Green following the manufacturer's recommendation. Viral DNA was normalized by cellular genomic CCR5. Integration was measured using Alu-LTR-based real-time nested PCR procedure according to Brussel et al. [47], with the following modifications: the first amplification with primers L-M667 only (control) or with Alu1 and Alu2 (integrated) had an annealing temperature of 65 °C. To reduce unspecific background, 2 μl of the first amplification was digested with 20 U of Exonuclease I (New England Biolabs) in 20 μl for 2 h at 37 °C. The nuclease was heat inactivated at 80 °C for 20 min. Two μl of the digestion was then amplified with SYBR Green at 65 °C with primers Lambda T and AA55M. Primers sequence: Strong-stop: (1) agcctgggagctctctggcta and (2) ccagagtcacacaacagacgg; Late: (1) and (3) cgcttcagcaagccgagtcct; CCR5 gene: (4) gtgaagcaaatcgcagcccgc and (5) gcagcatagtgagcccagaag.
To quantify the unspliced and multiply spliced HIV-1 RNA, 5 μg of total RNA was extracted and reverse transcribed using first strand synthesis reverse transcription kit (Invitrogen) (5 μg of RNA were used without Reverse Transcription Enzyme as a negative control). Real-time PCR (Bio-Rad) was used to amplify both unspliced HIV-1 RNA using primers La8.1 (CTGAAGCGCGCACGGCAA) and L9 (GACGCTCTCGCACCCATCTC) and multiply spliced HIV-1 RNA using primers p659 (GACTCATCAAGTTTCTCTATCAAA) and p413MOD (AGTCTCTCAAGCGGTGGT) as described previously [48]. The amount of RNA was normalized to GAPDH mRNA.
Four ug of total RNA was reverse transcribed with Omniscript reverse transcriptase (QIAGEN) using random hexamers and oligo dT. The mixture was aliquoted in a 96-well plate and subjected to PCR amplification using 0.375 U/15 μl of hotStarTaq DNA Polymerase with specific primers (0.3–0.6 μM) using the buffer provided by the manufacturer (QIAGEN). The PCR reaction was carried out in a GeneAmp 9700 PCR system. Following an incubation of 15 min at 95 °C, and 35 cycles of 30 s at 94 °C, 30 s at 55 °C, and 1 min at 72 °C, the reaction was ended with an extension step of 10 min at 72 °C. PCR products were fractionated on a LabChip HT DNA assay station (Caliper) for quantitation and sizing. |
10.1371/journal.ppat.1005964 | HIV Cell-to-Cell Spread Results in Earlier Onset of Viral Gene Expression by Multiple Infections per Cell | Cell-to-cell spread of HIV, a directed mode of viral transmission, has been observed to be more rapid than cell-free infection. However, a mechanism for earlier onset of viral gene expression in cell-to-cell spread was previously uncharacterized. Here we used time-lapse microscopy combined with automated image analysis to quantify the timing of the onset of HIV gene expression in a fluorescent reporter cell line, as well as single cell staining for infection over time in primary cells. We compared cell-to-cell spread of HIV to cell-free infection, and limited both types of transmission to a two-hour window to minimize differences due to virus transit time to the cell. The mean time to detectable onset of viral gene expression in cell-to-cell spread was accelerated by 19% in the reporter cell line and by 35% in peripheral blood mononuclear cells relative to cell-free HIV infection. Neither factors secreted by infected cells, nor contact with infected cells in the absence of transmission, detectably changed onset. We recapitulated the earlier onset by infecting with multiple cell-free viruses per cell. Surprisingly, the acceleration in onset of viral gene expression was not explained by cooperativity between infecting virions. Instead, more rapid onset was consistent with a model where the fastest expressing virus out of the infecting virus pool sets the time for infection independently of the other co-infecting viruses.
| How quickly infection occurs should be an important determinant of viral fitness, but mechanisms which could accelerate the onset of viral gene expression were previously undefined. In this work we use time-lapse microscopy to quantify the timing of the HIV viral cycle and show that onset of viral gene expression can be substantially accelerated. This occurs during cell-to-cell spread of HIV, a mode of directed viral infection where multiple virions are transmitted between cells. Surprisingly, we found that neither cooperativity between infecting viruses, nor trans-acting factors from already infected cells, influence the timing of infection. Rather, we show experimentally that a more rapid onset of infection is explained by a first-past-the-post mechanism, where the fastest expressing virus out of the infecting virus pool sets the time for the onset of viral gene expression of an individual cell independently of other infections of the same cell. Fast onset of viral gene expression in cell-to-cell spread may play an important role in seeding the HIV reservoir, which rapidly makes infection irreversible.
| Cell-to-cell spread of HIV is a mechanism of viral transmission whereby interaction between an infected donor cell and an infectable target cell leads to the directed transmission of virions to the target cell. Such interactions can occur between donor and target cells by various mechanisms [1–12], all of which involve the directed delivery of virions very close to the target cell, minimizing the distance over which virions need to diffuse and the consequent loss of virions en route [1–9, 11–24]. Because of the resulting high efficiency of viral delivery, target cells in cell-to-cell spread are exposed to multiple virions per cell both in in vitro infections and in vivo [17, 18, 25–31]. Multiple infections per cell decrease the sensitivity of cell-to-cell spread to antiretroviral drugs [17, 25, 27, 32, 33] and neutralizing antibodies [18, 34–36], and can overcome low infectivity and cellular restriction factors [37], since they increase the chances that at least one of the transmitted virions will successfully infect the cell despite inhibitors or unfavorable infection conditions [27, 38]. Because the source of insensitivity to inhibitors in cell-to-cell spread of HIV derives from multiple infections per cell, it is expected that sufficiently high inhibitor concentrations, or inhibitors more adept at suppressing multiple infections, could overcome this barrier [32, 33]. Conversely, cell-to-cell spread would offer a window of opportunity for HIV to evolve resistance to antiviral inhibitors [35].
As well as decreasing sensitivity to inhibitors, cell-to-cell spread of HIV was observed to be more rapid than cell-free infection [2, 13, 39–41]. One explanation may be fusion between donor and target cells. Fusion is insufficient for infection, as nucleic acids cannot directly infect a cell by translocating to the uninfected target cell [22]. However, the target cell would be scored as infected if a viral gene product or marker is used for detection, as fused cells share their protein pools and the marker would translocate to the target from the donor cell whether or not infection of the target cell took place. If fusion is excluded, acceleration of the viral cycle may be the result of several mechanisms: Shorter distance for the virus to transit before reaching a target cell, faster virus entry, faster pre- or post-integration dynamics due to cooperativity, and faster dynamics due to trans-acting factors secreted by the donor cells. Cooperativity would be expected to play a role in accelerating the virus cycle due to the Tat positive feedback loop [42–44], where Tat expressed from one provirus would trigger the transcript elongation of another provirus. Since the Tat protein can diffuse in and out of cells [43], such acceleration can also be potentially mediated in trans by the presence of nearby infected cells. Other HIV proteins, such as Nef, may also modify the physiology of yet uninfected cells upon cell-to-cell contact [45].
Another mechanism which can contribute to the acceleration of the viral cycle is probabilistic: since time to productive infection varies between virions due to integration site and stochastic gene expression [42, 44, 46, 47], cell-to-cell spread, which leads to multiple infections per cell, could increase the probability that at least one of the infecting viruses would have rapid infection dynamics.
Here we determined the timing of cell-to-cell spread and cell-free infection in a short infection time window, thereby limiting the role that the transit time to the target cell plays in infection timing. Despite this, we observed that cell-to-cell spread of HIV led to significantly earlier onset of viral gene expression. Surprisingly, we did not find evidence that factors secreted by donor cells, infected donor cell contact with target cells in the absence of transmission, or cooperativity between virions caused the earlier gene expression onset. We were able to replicate earlier onset in viral protein expression by increasing the multiplicity of infection with cell-free virus. This explains the observed rapid onset of viral gene expression of cell-to-cell spread by a mechanism where the fastest virus to be expressed sets the time of infection independently of other infections of the same cell.
In this study, we used the timing of the detectable onset of viral gene expression as a measure of the rate of the viral cycle. We used several ways to detect HIV gene expression, as summarized in S1 Table. Virus used for infection was produced from a molecular clone of the NL4-3 HIV strain to minimize any sequence differences between infecting virions. To compare the onset of cell-free infection to cell-to-cell spread, we infected target cells with either cell-free virus obtained from the filtered supernatant of virus producing cells, or by coculture with infected donor cells. In coculture, infection consists of a mix of cell-to-cell spread of HIV and cell-free infection. Hence, any observed difference between coculture and cell-free infection would be an underestimate of the difference between cell-to-cell spread and cell-free infection.
In order to quantify the onset of coculture versus cell-free infection by time-lapse microscopy, we imaged infection in the RevCEM cell line [48]. This cell line contains a GFP reporter that is responsive to the HIV splicing regulator protein Rev and hence reflects the timing of late HIV proteins [43, 49–51]. In order to efficiently detect infection, we subcloned the cell line to produce the reporter clone E7. This increased the maximum percentage of GFP positive cells from approximately 10% in the parental line to 70% in E7 (Fig 1A, left column). To enable the automated determination of the number of infected target cells (S1 Fig), we further stably expressed mCherry in these cells and derived the mCherry labelled G2 clone (Fig 1A, middle column). To exclude donor-target cell fusions, we labelled donor cells with the vital stain CellTrace Far Red (CTFR, Fig 1A, right column). CTFR and mCherry double positive cells were excluded from the analysis. In the absence of fusion exclusion, coculture infection showed a baseline from the earliest time points, which may not be real infection (S2 Fig).
We imaged infection over two days (S1 Movie). We used automated image analysis to determine the number of GFP+/mCherry+/CTFR- cells over the total number of mCherry+/CTFR- cells in each field of view at each frame of the movie (Fig 1B). In this experiment and the other time-lapse experiments performed in this study, we did not track individual cells, but rather measured the number of target cells with detectable viral gene expression at each time-point. We limited infection to the first two hours by washing away cell-free virus after that time window, and inhibiting additional infection cycles by addition of the protease inhibitor atazanavir (ATV), which has been described to effectively inhibit cell-to-cell transmission [33]. We imaged infection after washing and ATV addition. The protease inhibitor was used at a concentration that blocked over 99% of coculture infections (S3A Fig). This window for infection limited the time that the virus could transit to the target cell to no more than two hours in both coculture and cell-free infection. We calibrated the input of cell-free virus and infected cells so that the frequency of infected target cells after 48 hours was similar between the infection modes and did not saturate the available target cells (S4 Fig).
We quantified the fraction of infected cells over time and observed that both cell-free and coculture infection resulted in a variable time to Rev activity in individual infected cells, consistent with previous results showing heterogeneity in the length of the HIV replication cycle in cell-free infection [52]. In both infection modes, no Rev activity was detected before approximately 20 hours, corresponding to a period of intracellular delay [53–55]. On average, coculture infection showed more rapid HIV gene expression relative to cell-free infection (Fig 1C). We derived the mean and standard deviation for the timing of coculture and cell-free infections by parametrizing the number of infected cells over time with a best fit Gamma distribution, since Gamma distributions are a standard model for the timing of multi-step processes [56]. We obtained a time to detectable per cell Rev activity in coculture infection of 28±5.0 hours (mean±std). In contrast, mean time to per cell Rev activity in cell-free infection was 34.5±6.1 hours. This constituted an acceleration of 19% in the mean time to Rev activity in coculture infection. The difference between the two means was significant (p = 9x10-4, bootstrap).
We investigated the role of secreted factors acting in trans in the earlier onset of HIV gene expression by coculture with infected donor cells. To isolate the contribution of factors acting in trans, we separated infected donors from targets by a transwell membrane permeable to cell-free virus and soluble factors. We obtained no acceleration of time to detectable GFP expression using transwell infection (Fig 2A).
We considered the possibility that factors acting in trans may only operate over very short distances or that direct contact between donor and target cells, unrelated to viral transmission, may be required for an earlier onset of HIV gene expression. To test this, we took advantage of the fact that our reporter cell line could only be infected with HIV which uses the CXCR4 co-receptor. We therefore infected cells using the cell-free route with our CXCR4 tropic strain (NL4-3) in the presence of cocultured CD4+ cells infected with CCR5 tropic HIV (NL-AD8). This CCR5 tropic strain is identical to NL4-3, except for the Env protein, which is specific for the CCR5 co-receptor. We verified that NL-AD8 infected CD4+ cells could not infect the G2 target cells by coculture (S5 Fig). We did not observe a more rapid onset of HIV gene expression of cell-free infection cocultured with cells infected with the CCR5 tropic HIV compared to cell-free infection in the absence of these cells (Fig 2B), indicating that trans-acting factors are unlikely to induce an earlier onset of viral gene expression.
We asked whether the higher force of infection in cell-to-cell spread, manifesting as multiple infections per target cell, leads to earlier onset of HIV gene expression. We therefore used concentrated cell-free virus to mimic the higher infection levels per cell observed in cell-to-cell spread. We used the highly infection permissive MT4 cell line [27] to enable infection at a multiplicity greater than 1 within a two-hour infection window. As a reporter for infection, we used the NL4-3YFP strain of HIV [57] which substitutes YFP for the HIV early gene Nef. Therefore, YFP expression reflects the timing of HIV early genes (S1 Table). We infected MT4 cells with NL4-3YFP cell-free virus (S2 Movie) at increasing multiplicities of infection (MOI) per target cell, starting at an MOI of 0.1 infectious units per cell and up to an MOI of 4. After two hours, we removed the residual virus by washing and added sufficient ATV to prevent additional infections from coculture (S3B Fig).
We observed a more rapid onset of YFP expression with increasing MOI, accelerating mean expression time from 27.5±6.1 hours at an MOI of 0.1, which results almost exclusively in infections with one virus, to 22.6±5.5 hours at an MOI of 4 (Fig 3A). This acceleration in onset relative to the 0.1 MOI infection was significant (p = 1.7x10-3 for MOI = 0.5, p<10−4 for MOI = 2 and MOI = 4 using bootstrap).
We asked whether this earlier onset was mediated by cooperativity: pre- or post-integration interactions between virions that would lead to faster HIV gene expression. For this, we compared MT4 cells infected with NL4-3YFP alone to MT4 cells co-infected with NL4-3YFP and the unlabeled NL4-3 strain of HIV. The unlabeled HIV infection was at high multiplicity (MOI = 8) to ensure that the majority of cells infected with the YFP reporter HIV were also co-infected with the unlabeled HIV. If cooperativity has a role in the more rapid onset of viral gene expression, the unlabeled virus should accelerate the expression of labelled virus to the threshold of detection. However, we observed that co-infection did not lead to a more rapid onset of YFP expression (Fig 3B).
MT4 cells are known to be infected with HTLV-I [58] and hence any lack of cooperativity due to co-infection may be the result of saturating cooperativity with the endogenous virus. We therefore proceeded to investigate cooperativity in the onset of HIV gene expression between co-infecting viruses in primary CD4+ T cells. To investigate cooperativity in this system, we co-infected cells by the cell-free route with HIV expressing YFP and HIV expressing CFP. We detected the number of infected cells by flow cytometry at 6 hour intervals. We obtained CFP and YFP singly infected cells, as well as low but significant numbers of double infected cells (S6 Fig). We did not observe differences in timing of the onset of viral gene expression between singly infected and the CFP/YFP co-infected cells, indicating that co-infecting viruses did not show cooperativity in the onset of viral gene expression in primary CD4+ T cells and confirming our results in MT4 cells.
Since cooperativity between virions could not account for the earlier onset of HIV gene expression, we asked whether multiple infections per cell accelerated onset of gene expression by a first-past-the-post mechanism, where the earliest virus to express sets the time of infection (Fig 4A). This mechanism operates if: 1) Each integrated virus has a stochastically set time to viral protein expression. 2) Infections proceed independently. 3) A single expressed virion is sufficient to make use of target cell resources so that the target cell becomes infectious [26].
We reasoned that if a cell is infected by n>1 virions, the virion that first completes the replication cycle sets the time to infection. If each infection is independent, the distribution of the time to infection given n virions per cell is:
p(t,n)=n p(t)(1−Q(t))n−1.
(1)
Here p(t) is the distribution of the time to viral gene expression given a single virion per cell approximated by a Gamma distribution, and Q(t) is the corresponding cumulative distribution. In an infection with an average MOI m, the cells will be infected with a number of virions which is Poisson distributed around m and can be modelled by the average of Eq 1 over different n with Poisson weights (excluding n = 0). The distribution of the time to viral gene expression at m is then given by:
ρ(t,m)=e−m1−e−mΣn=1mnn!p(t,n),
(2)
where the pre-factor normalizes the distribution. We determined the shape and scale parameters of p(t) by jointly fitting the time series data for the multiple MOI infections to Eq 2. The model fits the data well for MOI 0.1, 0.5, 2, 4 with only the two parameters of the Gamma distribution, indicating that our model of independent stochastic infections can explain the acceleration at high MOI (Fig 4B).
To determine the effective MOI for coculture infections, we fitted the time course data to Eq 2 using the shape and scale parameters of the Gamma distribution determined by a fit to the cell-free data, approximating n = 1 for cell-free infections (Fig 4C). We obtained that the acceleration of viral gene expression with coculture was predicted by an effective MOI of 4.6 per cell.
To examine whether the earlier onset of HIV gene expression observed in the cell line also occurs in primary cells, we used coculture with autologous infected donor cells or cell-free virus to infect peripheral blood mononuclear cells (PBMCs) derived from healthy donors. As with the cell lines, donor cells were separated from target cells by labelling them with a vital stain. The fraction of infected target cells at different times post-infection was quantified by detection of the viral p24 protein, made as part of the HIV Gag polyprotein, using flow cytometry (Fig 5A). Virus was washed away after 2 hours in both cell-free and coculture infections, and ATV added to prevent additional infection cycles. ATV was used at a concentration sufficient to inhibit more than 99% of coculture infections (S3C Fig). Coculture dramatically accelerated the onset of HIV gene expression as measured by the detection of the HIV Gag protein relative to cell-free infection in primary human cells from 34.2±9.1 hours to 22.1±9.3 hours (mean±std). This constituted a decrease of 35% in the mean time to detectable HIV Gag expression (Fig 5B). Based on the cell-free distribution, the best-fit MOI per cell in coculture infection to recapitulate the difference in viral expression onset was 5.0 (Fig 5B, dashed red line).
PBMCs contain monocytes and other cells which may complicate interpretation of these results. To investigate whether T cell to T cell transmission was sufficient for the faster onset of viral gene expression, we repeated the experiment with purified CD4+ T cells (Fig 5B Inset and S7 Fig). We confirmed that cell-to-cell transmission between autologous T cells resulted in a more rapid onset of viral gene expression relative to cell-free infection.
We next proceeded to compare our predicted number of infections per cell using the timing of the onset of viral gene expression to that obtained by a second method. We have previously developed an approach to predict the number of infections per cell in cell-to-cell spread based on the reduced sensitivity to antiretroviral drugs relative to cell-free infection [27]. We therefore performed the PBMC infection in the presence of the integrase inhibitor raltegravir (RAL). As in the timing experiments, we used a 2-hour infection window. Coculture infection decreased sensitivity to RAL (Fig 6), consistent with our previous work and that of others showing that cell-to-cell spread decreases sensitivity to HIV inhibitors. For PBMC infection, IC50 of cell-free infection was 1.9nM and the maximum concentration of RAL used (60nM) decreased infection 12.2-fold. In contrast, IC50 of infection was 10nM and infection was reduced 2.6-fold at the same RAL concentration when transmission was by coculture. Reduced RAL sensitivity of coculture infection was also confirmed with transmission between purified autologous CD4+ T cells using a 2-hour infection window (Fig 6 Inset). In this case, 60nM RAL reduced cell-free infection by 33.3-fold. In contrast, coculture infection was reduced 3.6-fold. The best-fit MOI per cell needed to account for the reduced sensitivity of PBMC coculture infection to RAL was 4.8, which was similar to the number of virions predicted using infection timing under the same infection conditions.
We have observed faster onset of viral gene expression in coculture infection containing cell-to-cell spread of HIV relative to cell-free HIV infection. The earlier onset of viral gene expression in coculture was lost when target cells were separated from donor cells by a transwell membrane. A faster virus cycle in cell-to-cell spread relative to the non-directed, cell-free mode of infection has been previously observed directly [2, 13, 41] and inferred through modelling of infection dynamics [39, 40]. Here we used time-lapse microscopy of HIV infection to directly quantify and investigate the mechanism behind the faster onset of viral gene expression. We minimized possible differences between cell-to-cell spread and cell-free infection in the extracellular transit time from donor to target cell by limiting the time window of transmission to 2 hours. We have also minimized any contribution of virus sequence to different viral gene expression dynamics by using viruses with identical sequences derived from a molecular clone. Hence, variability in gene expression is a result of the interaction of the virus with the host cell. After exclusion of donor-target cell fusions, we found a minimum time for early viral protein expression in both infection modes, corresponding to a period of intracellular delay indicative of true infection [53–55]. We found that we could recapitulate the faster onset of viral gene expression by increasing the MOI of cell-free virus, and that there was no evidence for cooperativity or interference between co-infecting viruses. There was also no evidence for trans-acceleration of HIV gene expression onset from the surrounding infected cells.
Previous studies on cell-to-cell spread have concentrated on understanding the mechanisms by which cell-to-cell transmission occurs, and such mechanisms may lead to a faster onset of the expression of viral genes in the infected target cell in addition to making the infection more efficient. For example, it has been reported that the infected donor cell rapidly polarizes to the site of contact with the target cell [20] and that the subsequent transmission to the target cell occurs quickly [4, 18, 59, 60], though viral membrane fusion has been reported to be slower in cell-to-cell spread relative to cell-free infection [16]. Hence, a faster onset of HIV gene expression in cell-to-cell spread may be strictly mechanistic, due to more rapid entry of the virus. In this case, it would be expected that increasing cell-free MOI would not lead to faster onset, as increasing the MOI does not change the attachment and entry route. Since our data shows that cell-free MOI does control the onset of HIV gene expression, mechanistic factors such as more rapid entry in cell-to-cell spread are unlikely to play a major role.
Given multiple infections of the same cell in cell-to-cell spread, we would expect three possibilities of how co-infecting viruses could interact at the level of viral gene expression [61]. The first would be synergistic/cooperative interactions, where expression of one virus amplifies the expression of a co-infecting virus. The second would be no interaction, and the third would be that co-infecting viruses may compete for cellular resources and hence expression of one virus would interfere with the expression of co-infecting viruses. For example, comparing 10 co-infecting viruses to 10 infections identical in every way except occurring in 10 different cells, cooperativity would lead to the cell with 10 co-infecting viruses to show more rapid onset of viral gene expression relative to any one of the 10 single infections. No interaction between viruses would lead to the onset in the cell with 10 co-infecting viruses to be as fast as the fastest cell among the 10 singly infected cells, what we term a first-past-the-post mechanism. Interference or antagonism would lead to the cell with 10 co-infecting viruses to show a slower onset of viral gene expression than the fastest cell among the 10 singly infected cells, and possibly slower than the other singly infected cells.
Interactions between co-infecting viral genomes in HIV and other viruses have been extensively documented. For example, co-infecting viruses share post-integration components by a process known as complementation [62–66]. Hence, we would have predicted that there is at least some cooperativity in viral gene expression between co-infecting viruses as a result of the Tat positive feedback loop, as intracellular Tat concentration should increase with the number of expressed proviruses [42–44]. This mechanism of cooperativity would be expected to manifest as faster onset of viral gene expression since the delay to build up Tat levels by basal transcription should be reduced [44]. However, no detectable differences in the timing of the onset of HIV gene expression upon co-infection of YFP-HIV with unlabeled virus and no detectable differences when primary CD4+ cells were co-infected with two viruses argues against the presence of cooperativity at the onset of gene expression. Likewise, no interference was observed. Instead, we found that the mechanism most consistent with the faster onset of viral gene expression was that multiple infections per cell in coculture infection resulted in a pool of viruses which express viral genes at different times post-infection. The virus with the fastest onset of gene expression from this pool sets the start time for the generation of viral components by the infected cell.
We note that the lack of cooperativity as detected at the onset of HIV gene expression does not mean that co-infecting viruses do not interact, and interactions may occur later in the virus cycle. For example, interference may be expected to occur close to the time of peak virus production, where co-infecting viruses could compete for limited cellular resources to assemble virions [26, 67]. Such effects would influence the number of virions produced, but not the onset of gene expression as measured here.
We compared the predicted number of infections per cell based on the timing of the viral cycle to that predicted by the decreased sensitivity of coculture infection to an antiretroviral drug. The MOI per cell in PBMC coculture infection was predicted by timing to be 5.0 infectious viruses. This was similar to the predicted MOI based on the degree of insensitivity of PBMC coculture infection to the antiretroviral RAL (MOI = 4.8). Interestingly, the drug insensitivity of cell-to-cell spread to RAL was maintained despite keeping infection to one virus cycle using ATV. This indicates that the faster virus cycle of cell-to-cell spread is not necessary for drug insensitivity. However, a faster virus cycle may contribute to replication in the face of drug by amplifying an expanding infection.
Assuming that faster viral gene expression leads to more rapid viral dynamics, a more rapid onset of the viral cycle may confer a fitness advantage of rapid initial expansion, or transmission where the turnover rate of infected cells is high [2, 68]. Reasons for high turnover may include targeting of infected cells by cytotoxic T lymphocytes [69, 70], or a limited infection window due to bystander killing of target cells [71–74], all operating in environments such as lymph nodes where cell-to-cell infection is likely to occur [29, 31, 71, 73]. In exponential expansion at the R0 observed during primary HIV infection (~8, [75]), decreasing the infection cycle time by one quarter can lead to a 2 order of magnitude increase in the number of infected cells over several weeks. A large reservoir would be a barrier to a prolonged period of treatment interruption without rebound or to a permanent cure [76–81]. Thus, a faster viral cycle may seed a larger HIV reservoir, which would be more difficult to eliminate. However, if the most rapid virus cycle rate gives the highest fitness advantage, then cooperativity in gene expression would have been expected to evolve. Yet it does not seem to occur, perhaps indicating drawbacks to cooperativity such as more rapid cytotoxicity, or decreased ability of the virus to enter a quiescent state [82–85].
Blood was obtained from adult healthy volunteers after written informed consent (University of KwaZulu-Natal Institutional Review Board approval BE022/13).
The following reagents were obtained through the AIDS Research and Reference Reagent Program, National Institute of Allergy and Infectious Diseases, National Institutes of Health: the antiretroviral drugs ATV and RAL; Rev-CEM cells from Y. Wu and J. Marsh [48]; MT-4 cells from D. Richman [58]; HIV expression plasmid pNL4-3 from M. Martin [86] and pNL-AD8 from E. Freed [87]. The NL4-3YFP molecular clone was a gift from D. Levy [57]. Cell-free viruses were produced by transfection of HEK293 cells (ATCC) with molecular clones using TransIT-LT1 (Mirus) or Fugene HD (Roche) transfection reagents. Supernatant containing released virus was harvested after two days of incubation and filtered through a 0.45μm filter (Corning). The number of virus genomes in viral stocks was determined using the RealTime HIV-1 viral load test (Abbott Diagnostics). To produce the E7 clone, RevCEM cells were subcloned at single cell density and screened for the fraction of GFP expressing cells upon HIV infection using microscopy. To produce the G2 clone, E7 cells were stably infected with the mCherry gene under the control of the EF-1α promoter on a pHAGE2 based lentiviral vector (gift from A. Balazs), subcloned, and screened for clones with >99% mCherry positive cells. Similarly, the MT4-mCherry cell line was created by infecting MT4-cells with the pHAGE2 lentiviral vector expressing mCherry. PBMCs were isolated by density gradient centrifugation using Histopaque 1077 (Sigma-Aldrich). CD4+ cells were positively selected using CD4 Microbeads loaded onto MACS separation columns according to manufacturer’s instructions (Miltenyi Biotec). Culture and experiments were performed in complete RPMI 1640 medium supplemented with L-Glutamine, sodium pyruvate, HEPES, non-essential amino acids (Lonza), and 10% heat-inactivated FBS (Hyclone). Primary cells were additionally supplemented with IL-2 at 5ng/ml (PeproTech). PBMCs and CD4+ T cells were activated at 2*106 per ml density for one (donor cells) or three days (target cells) with PHA at 10μg/ml (Sigma-Aldrich).
For infection of RevCEM clones, 5x105 cells/ml E7 reporters were infected with 2x108 NL4-3 viral copies/ml (20ng p24 equivalent [88]) and used as infected donor cells. Infected and uninfected donors were incubated for two days, then stained with CellTrace Far Red (CTFR, Thermo Fisher Scientific) at 1μM and washed according to manufacturer’s instructions. G2 reporters at 5x105 cells/ml were either cocultured with 1:20 infected donor cells, or 1:20 uninfected donor cells and 109 NL4-3 viral copies/ml cell free virus. For RevCEM coculture experiments with cells infected with CCR5 tropic HIV, activated CD4+ cells at a concentration of 106 cells/ml were infected with 2x108 NL-AD8 viral copies per ml. Infected and uninfected CD4+ cells were incubated for two days. After two days, CD4+ cells were stained with CTFR as above. G2 cells were then infected with 109 copies/ml cell-free NL4-3, and cocultured with either infected or uninfected CD4+ cells, equal in number to the number of NL4-3 infected E7 cells added to the coculture positive control. For MT4 infections, cells were infected at a density of 5x105 cells/ml with 1.2x108 (MOI = 0.1) to 5x109 (MOI = 4) viral copies per ml of NL4-3YFP. For cooperativity experiments, MT4 cells were infected with 4x108 NL4-3YFP alone (MOI = 0.3) or co-infected with 4x108 copies of NL4-3YFP (MOI = 0.3) and 5x109 copies NL4-3 (MOI = 8). For PBMC infections, one day activated cells at a concentration of 106 cells/ml were used as donors and infected with 2x108 NL4-3 viral copies per ml. Donor cells were incubated for two days, and were separated from target cells by labelling them with CTFR or with carboxyfluorescein succinimidyl ester at 1μM (CFSE, Thermo Fisher Scientific) vital stain. CTFR or CFSE positive cells were excluded from the analysis, being either donors or donor-target fusions. Three day activated PBMC target cells at 106 cells/ml were then infected with either 1:10 infected donor cells, or with 1:10 uninfected donor cells and 5x108 copies of cell-free NL4-3. All cell-free and coculture infections of target cells were washed twice in medium after a two hour incubation with cell-free virus or infected donors, then resuspended in fresh growth medium with ATV. In the RAL sensitivity experiments, RAL was pre-incubated with target cells 4 hours before infection. Experiments comparing drug sensitivity and viral expression onset of co-culture and cell-free infections in primary CD4+ T cells were performed as with PBMCs. For experiments examining cooperativity in CD4+ T cells, infection with NL4-3YFP and NL4-3CFP was performed by adding 5x108 cell-free virions of each strain per 106 cells. CD4+ T cells were washed twice 2 hours post-infection and ATV was added as for PBMCs.
PBMCs and CD4+ cells infected with NL4-3wt or NL-AD8 were strained with anti-p24 FITC-conjugated or PE-conjugated antibody (KC57, Beckman Coulter) using the Cytofix/Cytoperm and the Perm/Wash buffers (BD Biosciences) according to manufacturer’s instructions. Cells were acquired with a FACSAriaIII or FACSCaliber machine (BD Biosciences) using 488 and 640nm laser lines. A minimum of 105 cells per sample were acquired. Results were analyzed with FlowJo 10.0.8 software. For CFP/YFP co-infection experiments, cells were acquired with a FACSAriaIII using the 405nm laser line for CFP, and 488nm laser line for YFP.
Cell density was reduced to 7x104 cells/ml and cells were attached to ploy-l-lysine (Sigma-Aldrich) coated 6-well optical plates (MatTek). Cell-free and coculture infections were imaged in tandem using a Metamorph-controlled Nikon TiE motorized microscope with a 20x, 0.75 NA phase objective in a biosafety level 3 facility. Excitation sources were 488 (GFP, YFP), 561 (mCherry), or 640 nm (CTFR) laser lines and emission was detected through a Semrock Brightline quad band 440–40 /521-21/607-34/700-45 nm filter. Images were captured using an 888 EMCCD camera (Andor). Temperature (37°C), humidity and CO2 (5%) were controlled using an environmental chamber (OKO Labs). Fields of view were captured every 30 minutes and a minimum of 1000 target cells were acquired per condition. Threshold for detection of the onset of HIV gene expression was set so that no positive cells were detected in the uninfected control. Cells with above threshold expression were scored as positive.
Cells were either infected by coculture in the lower compartment of a 6-well transwell plate with 0.4 μm pores (Costar) or separated across the membrane. To maintain a similar fraction of infected cells, 10-fold more donors were used when infection was across the membrane relative to coculture. Cell-free infection was performed in the lower compartment or across the membrane. After six-hour incubation, infection was washed, ATV added, and cells transferred to optical plates for imaging, keeping the donors in their initial compartments but not in the focal plane.
Movies were analyzed using custom code developed with the Matlab R2014a Image Analysis Toolbox. Images in the mCherry channel were thresholded to obtain images, and the imfindcircle function used to detect round objects within the cell radius range. Cell centers were found. GFP and CTFR signals underwent the same binary thresholding. The number of mCherry positive 16 pixel2 squares around the cell centers, negative for fluorescence in the CTFR channel and positive for fluorescence in the GFP channel, was used as the number of infected target cells. YFP signal in MT4 mCherry cells was analyzed in the same way except no CTFR stain was used, as infection was by cell-free virus.
For time-lapse experiments, data was normalized to compare infection between experimental conditions that had a similar, but not exactly equal number of infected cells. Normalization was by the average of the fraction of infected target cells during the last three hours to accurately capture the maximum infection level at the end of the viral cycle. Normalization by the maximum number of infected target cells was found to be noisy since it was sensitive to outlier values in the data.
Fitting of time-lapse data was done using a custom Python script using the Powell minimization algorithm from scipy (S1 Script). For drug sensitivity modelling, cell-free infection in the presence of increasing RAL concentrations was parametrized using the relation
d=1−11+(IC50D)h,
(3)
where d denotes the decrease in the experimentally determined fraction of infected cells relative to no drug, D is the drug concentration, and IC50, and h are the open parameters for the fit [89]. The number of infectious viruses per target cell (m) delivered in coculture infection was determined by fitting
Tx= IdrugI=(1−e−md)/(1−e−m),
(4)
where Tx is the experimentally determined number of coculture infected cells in the presence of different RAL concentrations normalized by the number of infected cells in the absence of RAL [27], and d is determined for each drug concentration by Eq 3. Script is provided (S2 Script).
|
10.1371/journal.pcbi.1003926 | ECOD: An Evolutionary Classification of Protein Domains | Understanding the evolution of a protein, including both close and distant relationships, often reveals insight into its structure and function. Fast and easy access to such up-to-date information facilitates research. We have developed a hierarchical evolutionary classification of all proteins with experimentally determined spatial structures, and presented it as an interactive and updatable online database. ECOD (Evolutionary Classification of protein Domains) is distinct from other structural classifications in that it groups domains primarily by evolutionary relationships (homology), rather than topology (or “fold”). This distinction highlights cases of homology between domains of differing topology to aid in understanding of protein structure evolution. ECOD uniquely emphasizes distantly related homologs that are difficult to detect, and thus catalogs the largest number of evolutionary links among structural domain classifications. Placing distant homologs together underscores the ancestral similarities of these proteins and draws attention to the most important regions of sequence and structure, as well as conserved functional sites. ECOD also recognizes closer sequence-based relationships between protein domains. Currently, approximately 100,000 protein structures are classified in ECOD into 9,000 sequence families clustered into close to 2,000 evolutionary groups. The classification is assisted by an automated pipeline that quickly and consistently classifies weekly releases of PDB structures and allows for continual updates. This synchronization with PDB uniquely distinguishes ECOD among all protein classifications. Finally, we present several case studies of homologous proteins not recorded in other classifications, illustrating the potential of how ECOD can be used to further biological and evolutionary studies.
| Protein structural domain databases offer a vital resource for structural bioinformatics. These databases provide functional inference for homologous structures, supply templates for structural prediction experiments, and differentiate between homologs and analogs. The rate of structure determination and deposition has increased dramatically over recent years, overwhelming the ability of current classifications to incorporate all new structures. We have developed a fast and reliable methodology for updating domain databases automatically, and created a revised hierarchy for domain classification that emphasizes evolutionary relationships. By classifying all known structures in our database with continuing automatic updates, we provide an up-to-date alternative to current resources. We illustrate several concepts that guided our classification scheme with examples of homology between domains in ECOD that are not observed in other resources.
| The billions of proteins in extant species constitute a bewilderingly diverse protein world. To understand this world, systematic classifications are needed to reduce its complexity and to bring order to its relationships. As proteins are the products of evolution, their phylogeny provides a natural foundation for a meaningful hierarchical classification. As in the classification of species, a phylogenetic classification of proteins identifies evolutionary relationships between proteins and groups homologs (proteins that are descendants of a common ancestor) together. Because homologs generally share similar three-dimensional (3D) structures and functional properties, such a classification provides a valuable platform for studying the laws of protein evolution by comparative analysis as well as for predicting structure and function by homology-based inference.
Many protein classifications are currently available. Comprehensive sequence-based classifications such as Pfam [1] and CDD [2] are among the most popular protein annotation tools. When sequence-only methods fail to reveal more distant evolutionary links, 3D structures allow us to see further back in time, as protein structure is generally better preserved than sequence in evolution [3]. Currently, the two leading structure classifications are SCOP (Structural Classification of Proteins) [4] and CATH (Class, Architecture, Topology, Homology) [5], both of which are widely used in analyzing protein sequence, structure, function, and evolution and in developing various bioinformatics tools. CATH (http://www.cathdb.info) is largely automatic with added manual curation and emphasizes more on geometry, while SCOP is mainly manual and focuses on function and evolution.
In the SCOP [4] (http://scop.mrc-lmb.cam.ac.uk/scop/index.html) hierarchical classification, closely related domains are grouped into families; families with structural and/or functional similarities supporting common ancestry are grouped into superfamilies; superfamilies with similar 3D architectures and topologies are grouped into folds; and folds with similar secondary structure compositions are grouped into classes. Cataloging remote homologies identified by a combination of visual inspection, sequence and structure similarity search, and expert knowledge, the SCOP superfamily is the broadest level indicating homology and offers invaluable insights in protein evolution. However, SCOP tends to be conservative in assessing evolutionary relationships, and many homologous links reported in literature are not currently reflected [6], [7], [8], [9], [10], [11]. Also, the recent dramatic increase of available structures in the PDB [12] (http://www.pdb.org) hinders careful manual curation in SCOP. Recently, a new version of SCOP (SCOP2) [13] was introduced that eschews hierarchical classification in place of a network of relationships (homologous and structural), although this database has not been made current with PDB. To partially alleviate this problem, ASTRAL now offers SCOPe, a sequence-based extension of the original SCOP hierarchy [14]. Nevertheless, not a single protein classification database has kept current with the PDB database. We maintain that the most recently determined structures, especially those evolutionarily distant from classified proteins, attract the most interest and hence are the most important to classify quickly and accurately. However, automatic updates, such as those in ASTRAL, are only able to deal with easily classifiable proteins.
Here we introduce the ECOD (Evolutionary Classification Of protein Domains) database. Our goal is threefold: (1) to construct a comprehensive domain classification based on evolutionary connections, (2) to extend the realm of connections to include remote homology, and (3) to maintain concurrent updates with the PDB. Because experimental data is very sparse compared to sequence data, establishing an evolutionary-based classification scheme of structures allows for biological insight into related proteins that otherwise lack functional information. In such a scheme, close homologs admittedly represent the most relevant source of functional inference. However for most proteins, only distant homologs have been studied in detail. Fortunately, many examples have shown that analysis of proteins in the context of their distant homologs provides functional clues that advance biological research [15], [16], [17], [18]. In addition, remote homology offers deeper insights in protein evolution. In order to extend distant evolutionary relationships beyond the SCOP superfamily level in ECOD, we apply state of the art homology-inference algorithms both developed in our group [19], [20] as well as by others [21], [22], manually analyze and verify the suggested homologous links, and incorporate findings from literature. For weekly updates, we rely on a computational pipeline that automatically and confidently classifies the majority of newly released structures and flags incompletely classified and unclassifiable structures, as well as a web interface that presents those difficult to deal with structures and pre-computed data in a convenient way for rapid manual inspection and classification.
ECOD is a publicly available database (http://prodata.swmed.edu/ecod/). By focusing on remote homology and weekly updates, ECOD strives to provide a more simplified and up-to-date view of the protein world than is currently available in existing classifications. As such, ECOD is unique in combining the following features: 1) the aforementioned weekly updates, following new releases from the PDB; 2) a hierarchy that specifically incorporates sequence-based relationships in a family level of close homology; 3) a classification that reflects more distant evolutionary connections; 4) a hierarchy that lacks a SCOP-like fold level, as the definition of “fold” is often subjective [23]; 5) domain partitions for all former members of the SCOPmulti-domain protein class; and 6) combination of membrane proteins with their soluble homologs where an evolutionary relationship can be hypothesized. Theoretically, ECOD catalogs rich and up-to-date information about protein structure for the studies on protein origins and evolution; and practically, it helps homology-based structure and function prediction and protein annotation by providing a pre-compiled search database.
We first developed a pilot version of ECOD based on SCOP 1.75 [4]. To detect remote homologies beyond the SCOP superfamily level, 40% identity domain representatives in the first 7 classes in SCOP 1.75 were retrieved from ASTRAL [24] and compared in an all-versus-all fashion. Four scores were computed for each pair: HHsearch probability [21], DALI Z-score [22], HorA combined score [20], and HorA SVM score [19]. Domain pairs with high scores were manually inspected and analyzed. The decision on whether any given pair is homologous was based on considerations of the aforementioned scores, literature, functional similarity (such as common cofactor-binding residues), shared unusual structural features [25], domain organization, oligomerization states, and disulfide bond positions. Since the SCOP superfamily level is reliable and conservative, we typically only merged SCOP superfamilies into homologous (H-) groups. In addition to merging SCOP superfamilies, we split SCOP entries with multiple domains or with duplications, and corrected rare inconsistencies in the SCOP classification. Cytoscape [26] clustering was used to aid manual analysis by displaying domains and high-scoring links. After 40% representatives were classified, other SCOP 1.75 domains were automatically mapped into the ECOD hierarchy using MUSCLE alignments [27]. Many hierarchical groups in the ECOD pilot version retained the names of their original SCOP counterparts.
Those structures not classified in SCOP 1.75 were partitioned and assigned to ECOD using a combination of sequence and structural homology detection methods. We used an iterative pipeline of three sequence homology detection methods of increasing sensitivity and decreasing specificity to partition input proteins into domains (Fig. 1). First, the input protein sequence is queried against a library of known ECOD full-length chains (containing both single-domain and multi-domain architectures) using BLAST [28],[29]. Where significant sequence similarity (E-value<2e-3) is detected to a known domain architecture with high coverage (<10 residues uncovered), the entire series of domains in the input chain was partitioned in one pass. Second, the protein sequence is queried using BLAST against a library of domain sequences. Here single-domain proteins and components of multi-domain proteins were assigned individually by sequence similarity (E-value<2e-3) and hit coverage (>80%). Finally, for detection of more distant homology, a query sequence profile was generated using HHblits [21]. This profile was used to query a database of ECOD representative domain profiles using HHsearch. Domains from the input chains could be classified by any combination of the three sequence-based methods (chain BLAST, domain BLAST, or domain HHsearch). Following partition, a boundary optimization procedure based on the structural domain parser, PDP, was run to eliminate small interstitial gaps between assigned domains and at termini [30].
Input protein chains with a set of detected domains with full residue coverage from the sequence pipeline were considered to be complete. Domains from these chains were then assigned to the ECOD hierarchy broadly using the classification of their hit domain. Following this assignment a combination of HMMER/Pfam and HHsearch-based clustering was used to finely tune family assignments [1], [31]. Domains were clustered into F-groups by Pfam where confident HMMER3-based assignments could be found. Where domains had no confident Pfam assignment, all-versus-all HHsearch-based complete linkage clustering was used to generate an F-group [32] where all domains shared 90% HHsearch probability. We specifically designate provisional representatives in F-groups where no member shares close homology with a representative ECOD domain for manual examination. Input protein chains that could not be fully assigned by the sequence pipeline were passed to the structural pipeline.
If a protein chain could not be assigned by the sequence pipeline, it was queried against a library of representative ECOD domain structures using DaliLite [33]. Domains were assigned where significant structural similarity existed to a known ECOD domain and where the aligned region passed a simple BLOSUM-based alignment score [34]. As in the sequence pipeline, the boundaries of structurally assigned domains were optimized, and those chains that could be completely assigned (100% residue coverage) were added to the classification. Where a chain could not be completely assigned, it was passed to the manual curators for boundary refinement or assignment. As we neared completion of the PDB, the need for structural search decreased as the number of remaining structures was small enough to manually curate.
Difficult structures that could not be completely and confidently classified by the pipeline required manual curation. We first inspected the mapping suggested by the pipeline. Oftentimes, the suggested mapping was correct for most or part of the query structure, and we typically accepted this mapping but modified the domain boundaries. For other queries where the suggested mapping was wrong or absent, we used HorA server [20] to search for remote homologs. In evaluating HorA results, we applied the same considerations used in developing the ECOD pilot version to determine homology between a query and a hit. When a homologous hit with similar topology could be found, the query was classified into the same T-group as the hit; when a homologous hit with different topology could be found, the query was classified in a new T-group but the same H-group as the hit; when only a possibly homologous hit with similar overall structure could be found, the query was classified in a new H-group but the same X-group as the hit; when no possible homologs can be identified, the query is classified in a new X-group by itself (see Results and Discussion for a description of the ECOD hierarchy). To facilitate manual analysis, we developed a web interface that presented relevant information in a clear format as well as recorded and incorporated feedback and annotations from manual curators.
ECOD is a hierarchical classification of domains based on their evolutionary relationships. Focusing on remote homology, ECOD organizes domains into very broad homologous groups. At the same time, ECOD families address closer evolutionary relationships, detectable at a sequence level. Most importantly, ECOD is comprehensive and up-to-date, including all entries in the PDB and updating weekly, thus uniquely providing researchers with the most current classification of protein domains at both distant and close homology levels.
ECOD is a hierarchical classification with five main levels (Fig. 2, from top to bottom): architecture (A), possible homology (X), homology (H), topology (T), and family (F). The architecture level (A) groups domains with similar secondary structure compositions and geometric shapes. The possible homology level (X) groups domains where some evidence exists to demonstrate homology (but where further evidence is needed). The homology level (H) groups together domains with common ancestry as suggested by high sequence-structure scores, functional similarity, shared unusual features [25], and literature. The topology level (T) groups domains with similar topological connections. The family level (F) groups domains with significant sequence similarity (primarily according to Pfam, secondarily by HHsearch-based clustering).
ECOD has 20 architectures that were developed both by consulting SCOP fold descriptions and inspecting numerous structures. We note that clear-cut boundaries between architectures do not always exist and that domain assignment to an architecture is sometimes subjective. This level is introduced largely for convenience of users and does not directly correspond to evolutionary grouping. A-level lies in between SCOP class and fold and groups proteins by simple visual features such as bundles, barrels, meanders, and sandwiches. Coiled-coils, peptides, fragments, largely disordered structures, and low resolution structures were put in special architectures with no X-, H-, T-, or F-levels, as confident evolutionary classification of these structures is challenging at the moment. Nucleic acids, in addition to proteins, are kept within a special architecture and are not currently classified. Within architectures, X-groups are ordered by structural similarity between them.
The ECOD X-level groups domains that may be homologous as is frequently suggested by similarity of their spatial structures. A domain's overall structure is traditionally referred to as its ‘fold’. Fold similarity usually refers to general resemblance in both architecture and topology and can result from either common ancestry (homology) or physical/chemical restrictions (analogy) [35],[36],[37]. Both SCOP and CATH have a fold level in the hierarchy: “SCOP fold” and “CATH topology”. However, the definition of fold can be subjective [23], and fold is a geometrical concept without explicit evolutionary meaning. Therefore, ECOD generally avoids the fold concept. However, domains that share strong overall architectural and topological similarity and are possibly homologous, but which lack further evidence to exclude analogy, are attributed to the same X-group but different H-groups. The conceptual difference between ECOD X-group and SCOP fold can be shown, for example, in the classification of domains with a ferredoxin-like topology. In SCOP, the ‘Ferredoxin-like’ fold is a large assembly of various superfamilies that share the (βαβ)×2 topology. Among all these superfamilies, 4Fe-4S ferredoxins seem unique for their small size and cysteine-rich nature (cysteines are used to coordinate the Fe-S clusters). Thus we suspect 4Fe-4S ferredoxins have an independent evolutionary origin and keep 4Fe-4S ferredoxins and other superfamilies in separate X-groups. On the other hand, although domains in the SCOP fold ‘Ribosomal proteins S24e, L23 and L15e’ do not have the ferredoxin-like (βαβ)×2 topology, their structures can easily be transformed into that topology by a circular permutation. Their structural similarity and functional similarity with the ‘RNA-binding domain, RBD’ superfamily in SCOP ‘Ferredoxin-like’ fold may imply homology. Therefore, ECOD classifies ‘Ribosomal proteins S24e, L23 and L15e’ and ‘RNA-binding domain, RBD’ as two H-groups in the same X-group as possible homologs. When further evidence coming either from additional sequences or 3D structures accumulates, classification decisions are adjusted to agree best with all available data.
We examined the distribution of domains mapped to SCOP folds and CATH topologies among ECOD X-groups. Of 1,799 ECOD X-groups, 598 include domains from only one SCOP fold and 564 include domains from only one CATH topology, reflecting agreement between classifications for these groups. 89 ECOD X-groups contain domains from multiple SCOP folds and 315 X-groups include domains from multiple CATH topologies. For example, the SCOP folds c.1-TIM beta/alpha-barrel and c.6-7-stranded beta/alpha barrel both contain domains mapped to the ECOD TIM beta/alpha barrel X-group. ECOD unifies such groups due to their shared structural similarity (7- versus 8- stranded) and similar locations of functional sites, but with insufficient evidence of homology to belong to the same H-group. 935 ECOD X-groups are not mapped to any SCOP fold, whereas 1,014 ECOD X-groups are not mapped to any CATH topology. The majority of these unmapped X-groups are simply due to proteins that are not classified by SCOP or CATH (722 and 872 X-groups, respectively); the remainder are shared proteins that are partitioned differently. Taken together, these results suggest that ECOD tends to merge both SCOP folds and CATH topologies into X-groups.
An ECOD H-group can contain more distant homologous links than the equivalent SCOP superfamily or CATH homologous superfamily. Although the majority of ECOD H-groups contain only a single SCOP superfamily (88%) or CATH homologous superfamily (81%), some H-groups contain many more (Fig. 3). For example, the Immunoglobulin-related and the Rossmann-related H-groups contain the most SCOP superfamiles (47 and 28, respectively) and CATH homologous superfamilies (81 and 40, respectively). Superfamilies were merged based on multiple high-scoring homologous links between domains. These merges reflect the homology between domain members of these previously split groups.
In total, 53 ECOD H-groups contain domains from two or more SCOP folds, and these H-groups contain domains from 151 unique SCOP folds, indicating that fold change in evolution of protein structures is not a very uncommon phenomenon. Similarly, 169 ECOD H-groups contain domains from two or more CATH topologies, and these H-groups contain domains from 357 unique CATH topologies. Additionally, 36 H-groups contain domains mapped to more than one CATH class, indicating homologous domains that nonetheless contain fairly different topologies.
To readily incorporate the observation that homologs can adopt different folds, ECOD has a topology (T-) level below the homology (H-) level. As a result, homologs with different topologies that SCOP necessarily separates into different folds (and thus different superfamilies) are unified in the same H-group but different T-groups in ECOD. For example, β-propellers are comprised of differing numbers of repeated β-meanders, all of which are evolutionarily related. The five different beta-propeller folds outlined in SCOP are organized in ECOD into a single H-group, with child T-groups for domains with differing number of blades [38]. Also, the domain contents of 11 SCOP folds are organized into multiple T-groups under the Rift-related H-group in the cradle-loop barrel X-group [39]. If we find sufficient evidence for homology between these proteins this consideration results in merging not only SCOP superfamilies, but also SCOP folds.
Within T-groups, ECOD organizes domains into families based on sequence similarity. We employ Pfam as the standard for family definition. ECOD domains were attributed to Pfam families by HMMER3 [31]. Therefore, the majority of ECOD F-groups are simply Pfam families. However, not all protein domains with known structure can be attributed to the current version of Pfam by sequence similarity. Those domains are grouped into families by HHsearch as outlined in Materials and Methods. As a result, ECOD contains 8,947 F-groups, 7,156 of which can be mapped to Pfam families, and 1,622 composed of homologous domains not mapped to any Pfam family.
Summary statistics for the ECOD database as of July 31stth, 2013 (version 22b) are presented in Table 1. The majority of the 317,021 domains in ECOD were assigned automatically to a smaller set of 15,969 manually curated domain representatives. Domains in ECOD were derived from five sources: 1) domains originally in SCOP ASTRAL40, inherited and reclassified manually in ECOD (11,462), 2) domains originally in SCOP, but not in the ASTRAL40 set, mapped by MUSCLE alignment with their ASTRAL representative (98,702), 3) novel domains not contained in SCOP, usually from chains deposited to the PDB in the intervening period between the release of SCOP v1.75 and ECOD, manually curated and added to the representative set (4,373), 4) domains automatically added to ECOD by detection of homology by pairwise sequence or structure search (153,381), and 5) domains added to ECOD by MUSCLE alignment of non-representative sequences to closely related ECOD representatives (48,817). The vast majority of domains classified in ECOD have been added by automatic methods. ECOD provides for domains which are assembled from multiple PDB chains, either due to photolytic cleavage (i.e. order-dependent assembly) or obligate multimers (i.e. order-independent assemblies). For order-independent assemblies, we distinguish between those domains where the assembly is primarily relevant for display, or appears to be biologically necessary. These are fairly rare in the database; only 132 representative order-independent assemblies have been defined. At the time of writing, 100% of PDB depositions could be accounted for in the ECOD classification (including those members of the special architectures).
We also compare ECOD to the most recent releases of SCOP and CATH. ECOD, SCOP, and CATH differ in domain partition strategy, classification hierarchy, and simply in the number of structures considered. At the time of writing, ECOD classifies 93,663 PDB depositions containing 239,303 protein chains, SCOP 1.75 contains 38,221 PDBs and 85,141 chains, and CATH v3.5 contains 51,334 PDBs and 118,792 chains. Of those chains classified in ECOD that are not in SCOP (and not in a special architecture), 137,794 were automatically classified and 2,484 were classified manually. Of those chains classified in ECOD, but not in CATH (and not in a special architecture), 106,474 were automatically classified and 2,521 were classified manually. The growth of the PDB over time is compared to the number of structures classified in ECOD, CATH, and SCOP (Fig. 4(a)). The difference between the number of structures in the PDB and those in the main architectures of ECOD can be primarily accounted for by the number of structures contained in ECOD special architectures (i.e. coiled-coil, peptide, non-peptide polymers, and low-resolution structures that could not be classified by sequence). The growth of the hierarchical levels from 2000–2013 indicates that although evolutionary distinct groups (i.e. X- and H- groups) are being discovered at a steady pace, the predominant source of new domains in ECOD is from sequence families (F-groups) being associated with existing homologous groups (Fig. 4(b)).
Since the July 2013 version, whose statistics are presented here, the subsequent 25 weekly releases by the PDB have been automatically classified (Fig. 5). Each week, protein chains are clustered at 95% redundancy, representatives for those non-redundant chains are classified; those remaining chains are classified when the initial automatic and manual classification pass are completed. For each weekly update, the majority (∼89%) of non-redundant (<95%) chains can be partitioned and assigned automatically (134.1±40.4). Those chains that cannot be resolved automatically are manually curated. On average, 11.7±4.9 chains per week were classified as manual representatives in ECOD, whereas 5.1±3.2 were chains not containing domains (i.e. peptides, coiled-coils, or fragments) that were resolved by assignment to special categories or other methods that did not modify the hierarchy. Overall, the majority of protein chains in weekly PDB releases can be classified automatically into ECOD.
We analyzed the distribution of domains in hierarchical levels in ECOD. The most populated homologous groups (H-groups) are placed in context with their architecture in ECOD (Fig. 6(a)) and are also ranked by population (Fig. 6(b)). The Ig-related and Rossmann-related H-groups, in addition to containing the most merged SCOP and CATH homologous groups, are the most populated H-groups in ECOD. The merging of many previously distinct helix-turn-helix (HTH) SCOP superfamilies in ECOD boosts the population of this H-group considerably compared to its original SCOP population. The inset (Fig. 6(b)) shows those most populated H-groups by number of F-groups. Where many sequence families have been merged by distant homology, such as the RIFT-related or Immunoglobulin-related domains, H-groups will contain many F-groups. In ECOD, as opposed to SCOP or CATH, there exist fewer distinct homologous groups with related topologies, as many of these groups have been linked by homology. For example, in ECOD, there is a single Rossmann-related H-group among the most populated (top 15) groups, whereas in the most populated SCOP superfamilies or CATH homologous superfamilies, there are two (NAD(P)-binding Rossmann fold domains and SAM methyltransferases) and four (3.40.50.720, 3.40.50.1820, 3.40.50.150, and 3.40.50.2300), respectively.
We compared our H-groups to SCOP superfamilies and folds by considering sequence and structure similarity of domain pairs within each level. ECOD manual representatives and ASTRAL40 domains were evaluated by HHsearch to reflect sequence similarity and TMalign to reflect structure similarity [21], [40]. SCOP superfamilies tend to contain more close homologs that can be detected by sequence homology search methods than ECOD H-groups (Fig. 7(c)). Domains classified in SCOP folds (excluding pairs from the same superfamily) emphasize structural similarity, as the distribution is mostly populated in the low sequence similarity region and the peak shifts right compared with others (Fig. 7(a,b)). On the other hand, as ECOD H-group readily incorporates homologous links from SCOP superfamilies and also many remotely homologous relationships that were previously overlooked, its peak sizes lie between SCOP fold and superfamily in high and low sequence similarity regions. Also it is worth noting that the peak of ECOD H-group does not have the right shoulder in the intermediate sequence similarity group but has a relatively evident left shoulder in the high sequence similarity group (Fig. 7(b,c)), which potentially supports the idea that ECOD classification is homology-centric.
We compared the domain partition observed in ECOD, SCOP, and CATH. Domain partition strategy can differ markedly between classifications, depending generally on whether the presence of compact structural units or overall sequence similarity is emphasized. The number of domains per chain observed in the domain classifications is presented in Figure 8(a). ECOD splits more protein chains (29%) into multiple domains than SCOP (23%), but splits slightly less than CATH (35%). The size distribution of domains in ECOD, SCOP, and CATH was compared (Fig. 8(b)). ECOD favors slightly shorter domains than SCOP, and favors slightly longer domains over CATH, but the size distributions are very similar. These results are consistent with the differences in domain definition strategy employed by different classifications. CATH emphasizes on structural integrity of the domain and its structural separation from other domains, SCOP focuses on the occurrence of an individual domain in different domain combinations, and ECOD attempts to find a compromise between these two strategies.
The difference in homologous links among equivalent domains was analyzed in ECOD, SCOP, and CATH. We define equivalent domains as those that share 80% residue coverage in all classifications. This subset of domains contains those domains whose partition is similar among classifications, but whose classification and homologous cluster size differ. We then analyze whether those domains that share a homologous link within one classification also share that link in other classifications. For the purposes of this analysis, only SCOP domains from canonical SCOP classes [a–d] are considered. Of the total domains in ECOD, 67,559 are defined equivalently (by 80% residue coverage) in SCOP and CATH. As many of these domains are identical or near identical in sequence, only domains with less than 95% sequence identity are used. There are 9,523 equivalent, non-redundant domains, shared among SCOP, CATH, and ECOD. Any pair of those equivalent domains belonging to the same H-group is considered to be homologous, 1,030,085 of these homologous domain pairs were observed in ECOD. Similar analysis was performed on SCOP superfamilies and CATH homologous superfamilies, where 711,894 and 680,726 homologous domain pairs were observed respectively. On average, 49.5% of domain pairs were shared between classifications, 36.6% of domain pairs were only observed in ECOD, 11.4% of domain pairs were observed only between ECOD and CATH (Fig. 9). Negligible numbers of domain pairs were observed in SCOP only, CATH only, or SCOP/CATH only. These results reflect a set in which most known homologous relationships among similarly partitioned domains are similar in ECOD as in SCOP and CATH. Additionally, ECOD catalogs many homologous relationships (among these similarly partitioned domains) that are not observed elsewhere.
SCOP recently diverged into two separate projects: SCOPe [14], which continues to update the original SCOP hierarchy using conservative automated methods, and SCOP2 [13], which is a dramatic reimagining of protein classification away from a hierarchal tree to a network model. We compared both of these more recent SCOP databases to ECOD. SCOP2 (February prototype version) eschews the traditional classification model; individual residues can be classified at multiple nodes in the network. We considered all SCOP2 domains, regardless of level, in comparison to ECOD. Of 995 PDBs and 1010 chains classified in SCOP2, equivalent domains were found in 725 PDBs and 732 chains. 70% of SCOP2 domains defined at the sequence family level were ECOD-equivalent. Conversely, only 56% of SCOP domains defined at the structural fold level (340/605) and sequence superfamily (272/482) level were equivalent to an ECOD domain. Only 61 of 257 domains defined at the hyperfamily (HF) level, are equivalent any domain in ECOD. Only 121 of 2,973 ECOD H-groups in this comparison were mapped to any domain in SCOP2. In general, the incomplete coverage of SCOP2 makes general statements about differences from ECOD premature.
SCOPe (v2.03-stable) uses a conservative automated method to add domains to the SCOP v1.75 hierarchy. Since both ECOD and SCOPe were derived from SCOP v1.75, we were particularly interested in classification of recent chains. ECOD v49 and SCOPe v2.03 (stable) contain 261,704 and 163,351 domains from shared protein chains, respectively. Of those SCOP-mapped ECOD domains, 94,292 were derived from SCOP v1.75 domains and 57,929 were independently classified. 27,142 ECOD domains derived from SCOPe shared chains do not map to any SCOPe domain, reflecting direct differences in domain partition strategy between SCOPe and ECOD. 1,493 SCOPe domains arise from structures classified only by SCOPe, but these structures are dominated by peptides and coiled-coils, regions that are not classified as domains by ECOD. 9,164 ECOD domains were derived from SCOP v1.75 domains, but are not mapped to SCOPe. These domains were generally the result of subdivision of a larger SCOP domain. There is a core set of domains that are shared by SCOPe and ECOD, both arising due to their shared origin and also due to independent classification of more recent domains. The differences in domain partition likely arise from differences in treatment of domain duplication and subdomains and are a potential target for further study.
We consider the growth-over-time analysis of ECOD in the context of the domain mapping between ECOD, SCOP and CATH (Fig. 10). Where an ECOD level (X-, H-, T-, or F-group) contains one or more domains with a mapping to a SCOP or CATH domain, we remove that level from consideration. We then re-plot the growth over time of ECOD using only those groups that contain no mapping to domains from other classifications. There is marked increase in novel ECOD classifications beginning in January 2005. The most recent deposition dates contained in SCOP 1.75 and CATH 3.5 are October 2008 and August 2011, respectively. However, the increase in novel classifications begins when the total PDBs and the PDBs classified in SCOP and CATH begin to diverge. The novel H-groups in ECOD (997) account for nearly 45% of total H-groups in ECOD. Those F-groups with no manual representative (where all domains were assigned automatically) are assigned a provisional manual representative. The majority of these automatically generated F-groups with no manual representative are derived from known Pfam families (Fig. 11). The increase in novel hierarchical levels in ECOD clearly demonstrates the value of an updated and comprehensive domain classification.
In addition to comparison of broad statistics of ECOD, we also present three examples of homologous relationships recorded in ECOD but not observed in other classifications. We consider any two homologous domains to have a “homologous link.” Firstly, we demonstrate the homologous link between SAM-dependent methyltransferases and NAD(P)-binding Rossmann-fold domains. These domains share topological connections, but a strand invasion causes them to bear distinct topologies, nonetheless, they share strong sequence similarity. Secondly, we show how members of the cysteine-rich domains of Frizzled share homology with other domain families that can primarily be detected by conserved patterns of cysteines. Finally, we describe a novel homologous link between Duf371 and the GutA-like PTSIIA component domain families within the topologically diverse cradle-loop barrel X-group. Each of these distinct examples demonstrates how the particular focus of distant homology in ECOD can reveal previously unknown relationships.
ECOD contains many homologous links that are not recorded in other classification databases. One example is the relationship between S-adenosyl-L-methionine-dependent methyltransferases (SAM MTases) and NAD(P)-binding Rossmann-fold domains (Rossmann domains). SAM MTases methylate a wide range of substrates using the methyl group donated by the cofactor SAM, which is comprised of an adenosine nucleoside and a methionine amino acid joined together. Rossmann domains are found in many oxidoreductases that transfer electrons between substrates and the cofactor NAD(P), which is comprised of a nicotinamide nucleotide and an adenine nucleotide joined together. Thus, SAM and NAD(P) share the adenosine part but differ in the other half, and the two enzyme superfamilies exploit the dissimilar parts of the cofactors to catalyze different reactions [41], [42], [43]. SAM MTases have a consensus structure of a 7-stranded β-sheet sandwiched between connecting α-helices (strand order 3214576 with strand 7 antiparallel to the other six strands, Fig. 12(a)) [44]. Rossmann domains have a consensus structure of a parallel 6-stranded β-sheet sandwiched between connecting α-helices (strand order 321456, Fig. 12 (b)) [45]. Thus, the SAM MTase structure can be viewed as Rossmann domain structure with a strand invasion: the additional strand 7 is inserted into the β-sheet between strands 5 and 6.
In SCOP, SAM MTases and Rossmann domains are classified in different folds (and therefore different superfamilies, SAM MTases: c.66.1; Rossmann domains: c.2.1), while in CATH, they are in the same topology group but different homology groups (SAM MTases: 3.40.50.150; Rossmann domains: 3.40.50.720). Although both SCOP and CATH indicate by their classification that SAM MTases and Rossmann domains are not homologous, literature suggests that they are actually related [46], [47]. As noted in reference [41], the overall structural similarity between SAM MTases and Rossmann domains is reflected in the observation that they are reciprocally the closest DALI hits to each other. In addition, SAM MTases and Rossmann domains bind their respective cofactors in a very similar fashion: the common adenosine part of the cofactors resides on top of a glycine-rich loop between the first strand and the first helix, and the adenosine ribose hydroxyls usually form hydrogen bonds with a conserved aspartate or glutamate residue at the end of the second strand (Fig. 12(a,b)) [41], [45], [46]. Indeed, the sequence-based homology detection algorithm HHsearch [21] and server HHpred [48] also provide statistical evidence that SAM MTases and Rossmann domains are related. In Cytoscape [26] display of SCOP domains and high-scoring links between them, numerous links with HHsearch probability above 90% exist between SAM MTases and Rossmann domains. In HHpred runs, for instance, when the Rossmann-domain in formaldehyde dehydrogenase (SCOP domain d1kola2, classified in c.2.1, Fig. 12(b)) is submitted as query to search against scop95_v1.75B database with secondary structure scoring turned off, the top hits within the same c.2.1 superfamily are followed by a region of mixed hits from both Rossmann domains superfamily (c.2.1) and SAM MTases superfamily (c.66.1). The highest-scoring hit from SAM MTases superfamily is hypothetical protein TM0748 (SCOP domain d1o54a_) with a 97.89% probability, E-value 9.4e-09, and identities 17% out of 110 aligned residues. Another SAM-MTase, ribosomal protein L11 methyltransferase (SCOP domain d2nxca1, Fig. 13(a) shows a same domain d2nxea1 with SAM bound), is detected with a 97.33% probability, E-value 3.4e-07, and identities 23% out of 102 aligned residues. Based on overall structural similarity, cofactor-binding resemblance, the number of confident homologous links observed between domains in each group, and statistically significant sequence similarity, ECOD classifies SAM MTases and Rossmann domains in the same homology (H-) group but different topology (T-) groups.
Frizzled receptors possess an extracellular cysteine-rich domain (FZ-CRD) for binding the Wnt ligands. FZ-CRD, as a mobile evolutionary module, has been found in other proteins such as the Smoothened receptor in Hedgehog signaling, secreted Frizzled-related proteins (SFRPs), and receptor tyrosine kinases MuSK and ROR. Sequence similarity searches and structural comparisons revealed distant similarities among FZ-CRD, Niemann-Pick type C1 protein (NPC1) that functions in cholesterol transportation, folate receptors and riboflavin-binding proteins (FRBPs) [17]. Recently, the core structures of two glypicans, proteoglycan molecules that regulate the signaling of a number of morphogens, were solved [49], [50]. Interestingly, comparative structural analyses suggested that glypicans also contain a cysteine-rich domain homologous to FZ-CRD and NPC1 [51]. Domains homologous to FZ-CRD and NPC1 have a wide distribution in eukaryotes, as they were also found in a number of other protein families currently without available structures, such as Hedgehog interacting proteins (HHIPs), RECK (reversion-inducing-cysteine-rich protein with Kazal motifs) proteins, the calcium channel component Mid1 in fungi, and the uncharacterized FAM155 proteins in metazoans [51].
The ECOD database unifies available structures of FZ-CRD, NPC1, folate receptor, and glypicans in one homologous group based on compelling sequence and structural similarities among them [17], [51]. These domains share similar disulfide bond patterns and adopt a similar overall structure fold with four core α-helices. Structural studies of three FZ-CRDs, in mouse Frizzled8 (Fig. 13(a)) [52], mouse SFRP3 [52], and rat MuSK [53] (Fig. 13(b)), revealed a common fold mainly consisting of four core α-helices (H1–H4 in Fig. 13). These FZ-CRD domains exhibit conservation of ten cysteines with a general pattern of ‘C*C*CX8CX6C*CX3CX6,7C*C*C’ (C: conserved cysteine; *: a variable number of residues, Xn: n residues, and Xm,n: m to n residues) (Fig. 13(g)). The disulfide connectivity patterns among the ten conserved cysteines are C1–C5 (between the first and fifth conserved cysteines), C2–C4, C3–C8, C6–C10, and C7–C9 (marked by black *, #, +, = , and & signs, respectively in Fig. 13(a,b and g)). The homologous cysteine-rich domain in glypicans possesses 12 conserved cysteines with a similar pattern of ‘C*C*CC*CX8CX2,3C*CX3CX6C*C*C’ (Fig. 13(c,g)) similar to that of the FZ-CRD. Such a pattern and disulfide bond connectivity (C1–C3, C2–C5, C4–C7, C5–C10, C8–C12, and C9–C11) (Fig. 13(g)) in glypicans are also seen in the structures of FRBPs including a folate receptor [54] (Fig. 13(d)) and a riboflavin-binding protein [55]. The structure of the cholesterol-binding domain of NPC1 [56] possesses eight of these 12 conserved cysteines, while lacking two disulfide bonds formed by C8–C12 and C9–C11 in glypicans, FRBPs (Fig. 13(c)). Together, the homologous cysteine-rich domains in Frizzled, NPC1, FRBP, and glypicans define a diverse superfamily of extracellular protein domains with an ancient eukaryotic origin and potential ligand-binding activities. Duplication and divergence of such a domain have resulted in a number of families with various functions in eukaryotic membrane transport and signaling. Despite overall similarity in fold and disulfide connectivity patterns, high structural divergence, reflected by low Dali Z-scores (Fig. 13(f)), was observed between some of these structures. The ECOD classification of this homologous group of proteins includes recently solved structures such as glypicans [49], [50] and the folate receptor [54]. In contrast, both the SCOP and CATH databases only have structures of FZ-CRDs from Frizzled receptors and SFRPs and do not include the structures of FZ-CRD of MuSK [53], NPC1 [56], glypicans, and folate receptor (although the sequence of MuSK is classified in the related CATH FunFam database).
ECOD establishes a previously unrecognized homologous link between a domain of unknown function (Duf371, PDB:3cbn) and the bacterial GutA-like PTS system glucitol/sorbitol-specific IIA component (PTSIIA, PDB:2f9h). While Duf371 is absent in SCOP, CATH classifies its fold (2.60.120.630) separately from that of PTSIIA (2.40.33.40). Duf371 forms an 8-stranded β-barrel from the intertwined β-strands of a tandem duplication (Fig. 14(a)). The duplicated structure elements can be superimposed (RMSD 1.3 Å), with a conserved His-containing motif from the N-terminal repeat overlapping a somewhat less conserved His-containing motif from the C-terminal repeat (Fig. 14(b)). Accordingly, PSI-BLAST [57] provides sequence evidence for this duplication, with both halves of the Duf371 query (PBD:3cbn, gi|169404770) confidently detecting the Methanocaldococcus fervens sequence Mefer0473 (3cbn[A:6-141] hits Mefer0473 with E-value 1e-30 in the first iteration, and 3cbn C-terminal range [A:77-142] hits with E-value 0.003 in second iteration). PTSIIA adopts a similar β-barrel topology as Duf371 and is noted in SCOP as consisting of two intertwined structural repeats (Fig. 14(c)). The overside connections between adjacent β-strands of the duplicated structure motifs in Duf371 and PTSIIA do not frequently appear in barrel architectures and distinguish the two folds.
A similar overside connection occupies the N-terminal half of pyruvate kinase (PK) β-barrel domain-like folds (ECOD/SCOP domain e1pklA1/d1pkla1). The PK barrel adopts a duplicated topology like PTSIIA and Duf371, although it lacks the C-terminal overside connection. The absence of this structural element in PK results in a 7-stranded β-barrel (Fig. 14(d)). The PK barrel half lacking the overside connection forms a ββxβ unit characteristic of the cradle-loop barrel metafold, which encompasses homologous folds of different topologies [39]. Based on the presence of a GD-box motif, the PK barrel was described as related to ancient RIFT-related folds (i.e. translation protein EF-Tu PDB: 1d2e) by a strand invasion of the N-terminal ββxβ unit that creates the overside connection [39]. Interestingly, the GD-box was also identified in both halves of PTSIIA (N-terminal GD and C-terminal GT) [58], but is not present in Duf371.
Structural similarity between Duf371 and PTSIIA is evidenced by their being reciprocal top Dali hits of each other (Dali Z-score 6), with the next best hits being to various RIFT-related homologs such as the PH barrel. The resulting structural alignment of the PTSIIA C-terminal sequence with both Duf371 sequence repeats is shown in Figure 14(e). A conserved C-terminal PTSIIA His residue (highlighted in black) marks the potential active site (Fig. 14(f)). Although the corresponding site in the Duf371 C-terminal repeat sequence is less conserved, an almost invariant threonine in the Duf371 N-terminal repeat aligns to the proposed PTSIIA functional site. Accordingly, the two folds may be related by a circular permutation of the structural repeats, maintaining a similar conserved active site position within the symmetry-related fold of Duf371 (Figure 14(g)). Considering the distinct topology of the duplicated structural motifs containing unusual overside connections, the unique way the two motifs tangle together to form an 8-stranded barrel, and the maintenance of similar active site positions, ECOD classifies PTSIIA and Duf371 as homologs in the same T-group within the RIFT-related H-group.
The ECOD database summarizes our views about partitioning of protein structures into domains and this evolutionary classification is a comprehensive resource for the research community. Data about a protein can be retrieved by PDB ID, keyword(s), or protein sequence search. Protein domains of interest are placed close to their close homologs, facilitating analysis of closely related protein structures. Information about more distant homologs is available by browsing representatives of this protein's homology group. ECOD database emphasizes distant evolutionary relationships that otherwise cannot be found. Finally, it is the only classification of protein domain structures that is kept current with the PDB, and every structure is classified with a week delay from its release by the PDB. This feature is significant because other classifications lag behind in updates and researchers are frequently interested in the newest protein structures. Future developments of ECOD will include incorporation of protein sequences without experimentally determined structures to cover as much of the protein world as possible.
|
10.1371/journal.pntd.0004681 | Predicting Dengue Fever Outbreaks in French Guiana Using Climate Indicators | Dengue fever epidemic dynamics are driven by complex interactions between hosts, vectors and viruses. Associations between climate and dengue have been studied around the world, but the results have shown that the impact of the climate can vary widely from one study site to another. In French Guiana, climate-based models are not available to assist in developing an early warning system. This study aims to evaluate the potential of using oceanic and atmospheric conditions to help predict dengue fever outbreaks in French Guiana.
Lagged correlations and composite analyses were performed to identify the climatic conditions that characterized a typical epidemic year and to define the best indices for predicting dengue fever outbreaks during the period 1991–2013. A logistic regression was then performed to build a forecast model. We demonstrate that a model based on summer Equatorial Pacific Ocean sea surface temperatures and Azores High sea-level pressure had predictive value and was able to predict 80% of the outbreaks while incorrectly predicting only 15% of the non-epidemic years. Predictions for 2014–2015 were consistent with the observed non-epidemic conditions, and an outbreak in early 2016 was predicted.
These findings indicate that outbreak resurgence can be modeled using a simple combination of climate indicators. This might be useful for anticipating public health actions to mitigate the effects of major outbreaks, particularly in areas where resources are limited and medical infrastructures are generally insufficient.
| Climatic determinants are amongst the most frequently cited in studies aimed at understanding and explaining the dynamics of vector-borne infections, and dengue in particular. French Guiana, a French overseas territory in which the vector Aedes aegypti is well established, experiences an epidemic cycle of dengue with large and prolonged epidemics occurring approximately every 3 years. Dengue is one of the most prioritized infectious diseases, and it requires an intense mobilization of local public health authorities, health services, and health professional and vector control services. A specific surveillance, preparedness and response plan has been developed based upon these needs. Gaining an accurate understanding of the drivers of dengue transmission is required to develop a model to predict the risk of an epidemic and to plan activities aimed at controlling it. Here, we assessed the effects of climatic factors on dengue spread to develop a predictive model of the epidemics in French Guiana on a country-wide scale. The goal of the model is to anticipate and plan both preventive and control activities. Given climate conditions, the model predicts that a dengue epidemic is likely to occur in early 2016. These conditions, which are favorable for Aedes mosquito proliferation, could also enhance the diffusion of other arboviruses, such as the Zika virus, in northeastern South America.
| Dengue fever (DF) is one of the most important mosquito-borne diseases in the world [1,2]. Recent estimates indicate that there are 390 million dengue infections per year, of which 96 million manifest as disease [3]. Infection is caused by the dengue virus (DENV), which has four closely related serotypes (DENV1 to DENV4) [4] that are transmitted to humans by infected Aedes sp. mosquitos. Infection produces a spectrum of illnesses that range from indiscernible or mildly nonspecific febrile syndrome to severe disease forms, including dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS). Currently, there are no specific dengue therapeutics, and prevention strategies are limited to vector control measures [5]. The recent development of the first dengue vaccine represents a major advance in our ability to control the disease [6–8].
In Latin American and Caribbean countries, the reintroduction and dissemination of Aedes aegypti occurred in the 1970s [9]. Since then, regular outbreaks have occurred on a 3- to 5-year cycle, and an increase has been observed in the frequency of severe forms of dengue [10]. In French Guiana, a French overseas territory of 250,000 inhabitants that is located in South America along the Atlantic Ocean, the epidemiology of dengue evolved from endemo-epidemic to hyper-endemic conditions over the two last decades [11]. Since the first DHF cases were reported in 1992 [12], transmission in French Guiana has followed a seasonal pattern that is punctuated every few years by major outbreaks that have been linked to the circulation of one or two predominant serotypes [11,13].
With the increasing frequency of such epidemics and the associated public health and socioeconomic issues [14], the surveillance, prevention and control of dengue have become social, political and public health challenges that require specific preparedness activities, particularly in areas where resources are limited. Although dengue ecology is known to be influenced by a complex multi-scale interplay of intrinsic factors that include human host demographics, vectors, and viruses and extrinsic factors that include environmental, meteorological and climate conditions, the factors that drive DF epidemics are not yet clearly understood [15–22].
Interactions between climate and DF outbreaks have been studied worldwide [23–34]. The findings of these studies suggest that the effects of climate parameters on the incidence of DF can vary widely from one study site to another [35,36] and that they depend largely on local context and epidemiological patterns. In South America, studies designed to determine the impact of climate on DF epidemics have suggested a role for El Niño events as triggers for epidemics [35]. El Niño conditions are likely to influence DF dynamics indirectly by modulating temperature, humidity and rainfall. In French Guiana, the sole study that focused on the DF-climate relationship identified a synchronous positive association between the occurrence of El Niño events, warmer temperatures, less abundant rainfall and dengue epidemics [37]. These results were obtained using basic analytical methods, and the study investigated El Niño conditions on a coarse annual scale. These results must be explored further, particularly if they are to be useful for prediction purposes. Moreover, the quality of epidemiological data (i.e., estimated suspected cases) that were available for the period covered by the study (1963–1993) was highly questionable.
Thus, even if associations between El Niño conditions, meteorology and DF epidemics are suspected, there is currently no climate-based model to assist in developing an early warning system in French Guiana. Based on the a priori hypothesis put forward in Gagnon et al. [37], the current study explores the potential of integrating sea surface temperature (SST) conditions to serve as a proxy for epidemic risk several months before the onset of a DF outbreak. In addition, we push our analysis further by also investigating the use of large-scale atmospheric circulation and regional climate patterns as more optimal indicators for predicting outbreaks. Using a long-term epidemiological surveillance dataset, this study explores the possibility of using a predictive model to assist public health authorities in implementing timely, appropriate and efficient prevention and mitigation strategies.
French Guiana is an overseas region of France that is located in northern South America between Brazil and Surinam. The climate is equatorial, hot and wet. The monthly mean temperatures (near 27°C) and relative humidity, which rarely falls below 80%, are nearly constant year-round. Spatial variations across the territory, particularly in the coastal area (regrouping 90% of the population), are low. Among meteorological parameters, only rainfall presents significant seasonal variations influenced by the migration of the intertropical convergence zone (ITCZ). The mean annual cumulative rainfall is approximately three meters, and there are four alternating seasons: a long rainy season from the beginning of April to mid-July, a long dry season from mid-July to mid-November, a short rainy season from mid-November to mid-February, and a shorter dry season from mid-February to the beginning of April. Large inter-annual variations in the total cumulated rainfall have been observed, and they are partly governed by large-scale atmospheric and oceanic patterns. A well-documented issue is the impact of El Niño conditions. During El Niño (La Niña) years, a rainfall deficit (surplus) occurs in French Guiana [38–42].
Epidemiologic data on DF were obtained from two different sources, depending on the collection period. For the data from 1991–2006, a surveillance system was used, and data were based on a weekly census of biologically confirmed cases (BCCs) that were recorded by the Arbovirus National Reference Centre, which is based at the Pasteur Institute of French Guiana. In 2006, a multi-source surveillance system was implemented by the Regional Epidemiology Unit of the Institut de Veille Sanitaire that included all seven biological laboratories (public hospital and private laboratories) that are located in the coastal area. Concurrently, in 2006, a new dengue diagnostic test based on NS1 antigen detection was made available to all laboratories, and it contributed substantially to improving surveillance. Cases were biologically confirmed by isolating the virus and detecting viral RNA using reverse-transcription PCR (RT-PCR), NS1 antigen detection methods or serological tests that are based on an immunoglobulin M (IgM)-capture enzyme-linked immunosorbent assay (MAC-ELISA) [13]. This surveillance system was authorized by the French Data Protection Agency (CNIL, N°1213498). The DF incidence rates in French Guiana, which are defined as the yearly number of cases/100 000 inhabitants, were calculated for the 1991–2013 period. A standardization procedure was performed separately for the 1991 –April 2006 and May 2006–2013 periods. For this procedure, we used a z-score scaling method to take into account the improvement in the epidemiological surveillance system that was observed in April 2006. This approach led to a trend toward increasing numbers of cases and enabled us to work with a single dataset. The standardization was calculated using the following equation:
z=x−x´σ
where x, x´ and σ were the observed value, mean and standard deviation of the incidence, respectively. The epidemic years were identified by applying the tercile method to the normalized and standardized sum of the monthly cases that occurred during the high incidence period. The first tercile was defined as the “low” incidence group, the second was defined as the “intermediate” incidence group, and the third was defined as the “high” incidence group.
We used sets of meteorological parameters and large-scale atmospheric and global SST data for 1990–2013 for this study. Meteorological records included rainfall, temperature and relative humidity and were obtained from Météo-France. We calculated monthly country means from these daily station data throughout the study period. A set of atmospheric and SST predictors was constructed from the ERA-Interim reanalysis data that were obtained from the European Centre for Medium-Range Weather Forecasts [43]. The ERA-Interim system assimilates observations and outputs using a forecast model. The climate fields were available at a 0.75°x0.75° spatial resolution and 60 vertical levels.
First, time-lagged Spearman’s correlations were used to explore associations among the occurrence of El Niño events, warmer temperatures, less abundant rainfall and dengue epidemics as previously suggested by Gagnon et al. [37]. Different El Niño-related SST and sea level pressure (SLP) indices were tested, including Niño areas 1 to 4, the Southern Oscillation Index (SOI) and the multivariate ENSO index (MEI). Yearly DF incidences were correlated with the monthly climate data for each month in the preceding year.
Second, the relationships between DF outbreaks and large-scale atmospheric and oceanic parameters were assessed using a composite analysis [44] and following an exploratory approach. The composite method was used to identify the conditions that characterized a typical epidemic year and to assess the optimal indices to use to analyze DF outbreak predictions. Two samples (the composites) were built that contained the climate data for both epidemic and non-epidemic years. The differences between the means of the two samples were calculated at each grid point between 50°N-50°S and 150°W-0°E. The significance of the differences between the epidemic and non-epidemic years was assessed using Student’s t-tests.
Considering that major outbreaks affect a very large part of French Guiana, we built a climate-based forecast model using the climate factors identified as having an influence on DF at a country level. A logistic binomial (epidemic or non-epidemic year) regression model was used. If p is the probability of an outbreak, then (p1−p) is the odds of observing an outbreak. Thus, the following logistic regression model was used:
log(p1−p)=β0+∑ i=1kβixi
where log represents the natural logarithm, k represents the number of selected climate predictors, βi represents the coefficient of the ith predictor and xi represents the ith predictor. This model can be restated as follows:
p=exp(β0+∑ i=1kβixi)1+exp(β0+∑ i=1kβixi)
Logistic binomial regressions were fitted using univariate and multivariate methods by applying all of the possible predictor combinations. The model that maximized the AUC (area under the curve) from the receiver operating characteristic (ROC) analysis [45] and minimized the AIC (Akaike information criterion) [46] was selected. The final model performances were evaluated by calculating ROC scores and cross-validating the data [47]. The ROC is a method of testing the skill of categorical forecasts using the hit rate (HR) and false alarm rate (FAR). The HR indicates the proportion of epidemic years that were categorically forecast (sensitivity). It ranges from 0 to 1 (1 being desirable) and is calculated as follows:
HR=HitsHits+Misses
The FAR is the proportion of non-epidemic years that were forecast as epidemic years (1-specificity). It ranges from 0 to 1 (0 being desirable) and is defined as follows:
FAR=False alarmsHits+False alarms
Second, a cross-validation on chunks of multiple years was performed to measure the model stability. Leave-one-out cross-validation (LOOCV) (i.e. 23-fold) was used. The model was refitted according to the number of observations, and the observations were then temporarily removed one by one. The resulting LOOCV δ was the cross-validation estimate of prediction error.
The year-to-year variability in DF incidence rates in French Guiana was described over a 23-year period from 1991–2013 (Fig 1A). The monthly mean cycle of DF standardized anomalies showed that there was strong seasonality (Fig 1B). The mean onset of the high incidence period was in January (positive anomalies) during the short rainy season. DF case peaks generally occurred in March, and the anomalies then decreased until May (negative in June). The high incidence period was therefore defined as January–May. Eight major outbreaks (third tercile) were identified: 1992, 1997, 1998, 2005, 2006, 2009, 2010 and 2013 (Fig 1C).
Spearman’s lagged correlations indicated the presence of associations between DF and monthly pre-epidemic climate factors (Fig 2). Among El Niño indicators, the Niño 3 area index showed the highest correspondence with DF. A significant negative correlation was observed between DF and rainfall in October (r = -0.49, p-value = 0.02) and November (r = -0.52, p-value = 0.01), which are one and three months before the mean onset of the epidemics, respectively. El Niño event-related indices and temperatures were not significantly associated with DF (p-value > 0.05). However, an interesting, persistent, positive and nearly significant correlation was observed between the Niño 3 area index and DF during the summer months, and this relationship deserves further investigation.
SST composite maps were calculated for the 12 months from January to December. The results indicated that epidemic years were characterized by increased Pacific Ocean SSTs during the pre-epidemic months of July and August (Fig 3, only July–December is shown here). This warming was particularly strong (approximately 1.5°C) at the equator at approximately 120°W, and the maximal spatial extent was observed in July.
The analysis of differences in atmospheric circulation between epidemic and non-epidemic years at the end of the dry season in October–November (when there were significant negative correlations between DF and rainfall; Fig 2) showed that epidemic years were characterized by northward positioning and a strengthening of the Azores High in November (Fig 4). The mean difference between epidemic and non-epidemic years was approximately 5 hPa and was maximal at 40°N, 30°W.
Based on previous results, the following climate indices were included in logistic binomial univariate and multivariate models (Table 1): (1) October–November, mean rainfall in French Guiana (FG-ON-RAIN); (2) July–August, mean Equatorial Pacific Ocean (2° N-20°S, 135°W-90°W) SST (EPO-JA-SST); and (3) November, Azores High (45°N-35°N, 40°W-20°W) SLP (AH-N-SLP). Because previous SST and SLP indices were found to be associated with rainfall in French Guiana, their common association in the same model was discarded.
The multivariate model that included the two predictors EPO-JA-SST and AH-N-SLP yielded the best results for the AIC (27) and AUC (0.88), which suggested that it had good predictive value. Warming in the mean Equatorial Pacific Ocean SST in July–August and the strengthening of the Azores High in November greatly increased the probability that an outbreak would occur in the following year in French Guiana (Fig 5). Accordingly, 80% of the epidemic conditions were correctly predicted (HR = 0.80). Outbreaks in the years 2001 and 2005 were incorrectly predicted to be non-epidemic, and two years were predicted as false alarms (1994 and 1999). Finally, the LOOCV δ of 0.18 indicates that the model was robust and that only 18% of the years were misclassified when the LOOCV procedure was used. Yearly cross-validated probabilities are shown in S1 Fig. The scatter plot between the observed DF incidence rate standardized anomalies and the predicted outbreak probabilities (Fig 6) revealed a nearly linear relationship (Pearson’s correlation: r = 0.76; P-value < 0.01).
Forecasts for 2014 and 2015 (not included in the training dataset) indicated that the model predictions were consistent with the non-epidemic conditions that were observed in French Guiana (the DF IR/100 000 inhabitants was 350 in 2014 and 106 for January to September, 2015) (Table 2). In 2016, as a result of the warm SST conditions over the Equatorial Pacific Ocean that occurred in August and July (25.26°C) and the high pressures over the Azores High in November (1021.36 hPa), the model predicted that French Guiana would likely experience an outbreak (probability of 0.92).
We investigated the relationship between climate and DF outbreaks in French Guiana to assess the possibility of including climate factors as predictors of epidemiological risk. Our findings highlighted a strong association between large-scale climate patterns and epidemic conditions in French Guiana. A simple and efficient statistical model was established to predict epidemic years. This model uses the summer Equatorial Pacific Ocean SST conditions six months prior to the mean onset month of the epidemic (July-1 –August-1) and the SLP of the Azores High three months prior (November-1), and it appropriately forecasted eight of the ten outbreaks that occurred in the 1991–2013 period. Outbreaks occurred after [i] warming in the Equatorial Pacific Ocean and [ii] northward displacement of the Azores High, which causes a rainfall deficit at the end of the dry season. This work refines the results of Gagnon et al. [37], who used the mean rainfall anomalies from May to April and concluded epidemics are associated with less abundant rainfall.
Warming events in the Equatorial Pacific Ocean are known to modify the mean climate over South America. During a warming phase, the northern region of the continent experiences drought conditions because the eastern subsidence of the Walker circulation is reinforced, which weakens convection and precipitation [48]. These results were consistent with the lagged correlation analysis, which indicated that a rainfall deficit would occur during the October and November preceding an epidemic year, and the a priori hypothesis derived from the work of Gagnon et al. [37], which indicated that there is an association between DF epidemic years in French Guiana and El Niño events. However, our analysis showed that anomalous SST patterns did not precisely correlate with El Niño events. Indeed, the SST anomalies of El Niño events were more intense in winter months. Importantly, certain epidemic years corresponded to strong (1997–1998) and moderate (1992, 2009–2010) El Niño events, although this association was not systematic. For example, an El Niño event did not occur from 2005–2006, but French Guiana did subsequently experience a dramatic epidemic.
In northern South America, rainfall is associated with water vapor that is transported from the north Atlantic via the northeasterly trade winds and thus from the SLP gradients. At the end of the dry season (October–November), the convergence of trade winds transports moisture over French Guiana, which fuels convection and precipitation. Positive SLP anomalies over the Azores High favor a northward position of the ITCZ over the Atlantic. Precipitation consequently decreases over the north part of South America in November.
Not all El Niño events are the same, and their effects on weather/climate may therefore differ. In 1994 and 1999, the model predicted an epidemic year would result from the high SLP and SST indices. However, the observed dry season rainfall patterns (rainfall surplus) were inconsistent with the usual occurrences, and this may explain the lack of an epidemic situation. In addition, 2001 and 2005 were epidemic years that were not predicted by the model. In these two years, dry season rainfall showed no positive or negative specific anomalies. For 2005, these observations were consistent (moderate SLP and SST indices versus moderate dry season rainfall). However, in 2001, given the low SST and SLP indices, a wetter dry season was expected than was observed. Thus, if a large part of the dry season rainfall variability in French Guiana is driven by the SST and SLP of the targeted areas, it is not always the case. These data highlight the complexity of predicting epidemiological patterns for non-climate-contrasted years and suggests that rainfall variability is not driven only by the two large scale indicators that were identified in the present study. Nevertheless, global predictions for the 1991–2013 period were better when the SST and SLP indices were used than when the dry season rainfall index was used (Table 1). Interestingly, the years that were incorrectly classified by the model preceded the implementation of the new enhanced surveillance system in 2006. Even when epidemic years that were identified by the tercile method were confirmed by historical reports, the representativeness of the surveillance system prior to 2006 could not differentiate an increased incidence that was caused by the presence of multiple isolated clusters from a major generalized epidemic. Historical reports may have overestimated the number of epidemiological circumstances, particularly in 2005, when one additional laboratory was included in the serological diagnosis process. Furthermore, three of the four misclassified years showed intermediate incidence rates (2nd tercile) that corroborated our results. Specifically, the two identified false alarms (1994 and 1999) showed relatively high incidence rates for non-epidemic years. In addition, one of the two unpredicted epidemic years (2001) was associated with lower incidence rates than the other epidemic years.
Our findings indicate that an important rainfall deficit at the end of the dry season enhances the risk of epidemic in the following year, and these types of conditions are likely to impact the vector population. Two non-exclusive main hypotheses related to mosquito densities can be stated. First, the eggs of Ae. aegypti, which is the only urban vector for DF in French Guiana, are known to be able to resist desiccation and to thereby survive dry episodes [49]. During a particularly dry season, the majority of the breeding sites dry up, but when the first rains of the wet season occur, their breeding sites are once again hydrated, and their eggs hatch synchronously, resulting in a rough proliferation of adult mosquitoes that is favorable to the emergence of an epidemic via to their introduction to infectious patients. The second hypothesis is related to human behaviors. Although precipitation is known to contribute to the multiplication of breeding sites, drought can also indirectly expand the vector’s range. Indeed, during pronounced dry seasons, some people may adapt their lifestyles by maintaining additional water-collection containers. Thus, because of increases in breeding sites around and within households, Ae. aegypti can maintain significant background densities during the dry season. As a consequence, the virus can remain in the area during the dry season, leading to a higher potential of an epidemic when the wet season returns. Further entomological field investigations should be performed to test these hypotheses. The evolution of the surveillance system that was used for data collection beginning in 2006, following the introduction of new laboratories and new methods of diagnosis, increased the difficulty of performing meaningful comparisons of the scope of epidemics, particularly those that occurred before 2006. It is also important to take the circulating serotypes into consideration to enhance the assessment of the model’s predictions. Circulating serotypes that have affected only a small portion of the population before the predicted year could play a role by increasing the transmission risk, given the size of the susceptible population. Conversely, serotypes that have recently caused epidemics could limit the transmission risk despite propitious climatic conditions. In this study, we explored the reason that the model wrongly predicted certain epidemiological situations looking at the predominant serotypes. Two of the four years that were wrongly predicted could be explained by the serotypes circulating during the previous year. In 1999, for which the model erroneously predicted an epidemic, both circulating serotypes (DENV1 and DENV4) had already caused epidemics in 1997 and 1998. In addition, the 2001 epidemic, which was not predicted by the model, was caused by DENV3, which had not caused an epidemic in the ten previous years. Other well-known factors that might play a key role in transmission, including the immune status of the host population or the presence of outbreaks in neighboring countries, were not included in the present analysis. Finally, considering the potential competitive viral suppression in vectors that can be caused by co-infections, the emergence of new viruses that can also be transmitted by Aedes mosquitoes in French Guiana, such as the Zika virus, could limit the transmission of dengue fever. It will be interesting to see how the emergence of the Zika virus in 2016 may interacts with dengue fever transmission in a propitious climatic context.
Future studies should attempt to validate hypotheses regarding the impact of the identified climate factors and associated meteorological patterns (i.e., the rainfall deficit at the end of the dry season) on direct measurements of vector behavior and breeding sites. Among other possible future developments, we plan to take into account the sub-country incidence of dengue to model the propagation of epidemics within the country.
Among the wide panel of factors that can influence DF outbreaks, these results suggest that large-scale climate factors play an important role. We found that the climatic indices that were assessed in this study were important for DF monitoring and for predicting outbreaks in French Guiana over a period of 2–3 months. This delay may give public health authorities the ability to anticipate outbreaks and implement social communication and vector control measures, and to adapt healthcare capacity and increase preparedness in a timely manner. Importantly, this model could be easily and regularly updated using newly collected data that was retrieved from the ongoing dengue surveillance system. Because the identified climate indicators are simple and easy to access, they could be used to estimate the probability of future epidemics occurring according to climate change simulations and help to evaluate the effectiveness of potential intervention strategies.
|
10.1371/journal.pgen.1007816 | BLM prevents instability of structure-forming DNA sequences at common fragile sites | Genome instability often arises at common fragile sites (CFSs) leading to cancer-associated chromosomal rearrangements. However, the underlying mechanisms of how CFS protection is achieved is not well understood. We demonstrate that BLM plays an important role in the maintenance of genome stability of structure-forming AT-rich sequences derived from CFSs (CFS-AT). BLM deficiency leads to increased DSB formation and hyper mitotic recombination at CFS-AT and induces instability of the plasmids containing CFS-AT. We further showed that BLM is required for suppression of CFS breakage upon oncogene expression. Both helicase activity and ATR-mediated phosphorylation of BLM are important for preventing genetic instability at CFS-AT sequences. Furthermore, the role of BLM in protecting CFS-AT is not epistatic to that of FANCM, a translocase that is involved in preserving CFS stability. Loss of BLM helicase activity leads to drastic decrease of cell viability in FANCM deficient cells. We propose that BLM and FANCM utilize different mechanisms to remove DNA secondary structures forming at CFS-AT on replication forks, thereby preventing DSB formation and maintaining CFS stability.
| Common fragile sites (CFSs) are large chromosomal regions which are more prone to breakage than other places in the genome. They are a part of normal chromosome structure and are present in all human beings, but are also hotspots for chromosomal rearrangement during oncogenesis. Understanding how CFSs are protected to prevent genome instability is thus extremely important for revealing the mechanism underlying cancer development. We found that Bloom syndrome protein BLM is involved in resolving DNA secondary structures that arise at AT-rich sequences in CFSs, suggesting a critical function of BLM in protecting CFSs. We also found that this BLM function is distinct from the role of Fanconi anemia protein FANCM in protecting CFSs, and loss of both BLM and FANCM activities leads to cell death. These studies reveal important mechanisms of the maintenance of CFS stability in mammalian cells.
| Genome instability is a hallmark of cancer cells [1]. Certain chromosomal loci, such as CFSs are hotspots for genome instability and are prone to chromosomal rearrangement [2]. CFSs are part of normal chromosomal regions that are present in all individuals, but are more susceptible to breakage than other genome loci under replication stress [3]. CFSs are preferential sites for sister chromatid exchanges and viral DNA integrations [2, 4, 5]. They are associated with chromosomal breakpoints observed in cancer cells which often involve deletions of tumor suppressor genes and amplifications of oncogenes [6–11].
Multiple mechanisms have been proposed to elucidate replication stress-induced CFS breakage (often termed as CFS expression) [12]. CFSs tend to be replicated late, and often lack sufficient replication origins [13, 14]. Due to origin paucity at CFSs, all available origins are utilized under normal conditions and no more dormant origins can be activated upon replication stress, which leads to unfinished DNA replication in mitosis and chromosome breakage [14]. Meanwhile, CFSs often contain very large gene, which could induce collisions of replication and transcription machineries, contributing to CFS expression [15]. In addition to these mechanisms, CFSs contain AT-rich sequences which are predicted to form strong DNA secondary structures [16]. These AT-rich sequences can block DNA replication in vitro [16, 17]. Importantly, it has been shown that replication stalling indeed occurs in vivo at the AT-rich sequences derived from FRA16D when they are present in yeast, and also at the AT-rich sites in FRA16C and FRA16B in mammalian cells [18–20]. We observed that CFS-derived AT-rich sequences induce DNA double strand break (DSB) formation and homologous recombination (HR)-mediated mitotic recombination [21]. Therefore, these CFS-derived AT-rich sequences (CFS-AT) are genetically unstable, and thus are one of the major contributors to CFS instability. However, not much is known about how the genome stability at CFS-AT is maintained to prevent CFS breakage.
FANCM is an ATP-dependent branch-point translocase, which is mutated in Fanconi anemia (FA) patients [22]. We previously found that FANCM has a unique activity independent of the FA core complex and FANCD2/FANCI to preserve stability of AT-rich sequences derived from CFSs [23]. Upon replication stress, CFS-derived AT-rich sequences form secondary structures when single-stranded DNA (ssDNA) is exposed at replication forks, and FANCM uses its translocase activity to promote fork reversal to remove such secondary structures, thereby preventing DSB formation. Homologous recombination (HR) is one major pathway to repair DSBs formed upon replication fork collapse [24]. In accordance with this, proteins that are involved in HR, such as BRCA1 and CtIP are important for maintaining stability of AT-rich sequences and suppressing CFS expression [21].
Bloom’s syndrome (BS) is a genetic disorder that is associated with a wide range of abnormalities including growth retardation, immunodeficiency, genome instability and predisposition to cancer [25, 26]. The BS associated gene product, BLM helicase, is a member of RecQ family, which acts as a 3’ to 5’ DNA helicase [27, 28]. BLM unwinds a variety of DNA substrates including Holliday junction, forked duplex, D-loop and G-quadruplex (G4) DNA [27, 29, 30] and is involved in multiple pathways contributing to the maintenance of genome stability. BLM is associated with topoisomerase IIIα, RMI1 and RMI2 to form a BTR complex that removes the Holliday junction by a mechanism called dissolution, thereby preventing crossover and sister chromatin exchange [31, 32]. BLM functions together with DNA2 in a non-overlapping pathway of Exo1 to mediate long-range end resection to generate 3’ single-stranded DNA tail from a DSB [33, 34]. BLM is also required for ensuring complete sister chromatid decatenation and thus is important for the suppression of anaphase bridges and faithful chromosome segregation [35].
In this study, we identified a new role of BLM in preventing DSB formation at CFS-AT sequences in a manner dependent on BLM helicase activity and ATR-mediated phosphorylation of BLM. This BLM function is non-overlapping with the role of Fanconi anemia (FA) protein FANCM in the maintenance of CFS stability identified in our previous study [23]. We propose that BLM resolves DNA secondary structure of CFS-AT sequences using a mechanism that is different from that of FANCM, and functions together with FANCM to preserve the integrity of CFSs.
As described previously, insertion of Flex1, the AT-rich sequence derived from FRA16D, into the EGFP-based HR reporter (HR-Flex) significantly increases mitotic recombination when comparing with the HR-Luc reporter which contains a luciferase sequence insertion of similar size [21] (Fig 1A). This HR-mediated mitotic recombination is caused by DSB formation at Flex1 due to Flex1 instability on replication forks, and in support of this, replication stress induced by hydroxyurea (HU) and aphidicolin (APH) enhances mitotic recombination at Flex1 [21]. Interestingly, suppression of BLM expression by shRNAs leads to increased mitotic recombination in U2OS (HR-Flex) cells (Fig 1B) [21]. Loss of BLM also increases HU-induced mitotic recombination (Fig 1C). These studies suggest that BLM plays a protection role in preventing instability of CFS-derived AT-rich sequences upon replication stress.
In addition to Flex1, AT-rich sequences derived from other CFSs also induce mitotic recombination [23]. When BLM is depleted by shRNAs, mitotic recombination induced by 16C/AT1 and 16C/AT3, two AT-rich sequences derived from FRA16C, is further increased (Fig 1D). This suggests that BLM-mediated protection is not limited to a specific AT-rich sequence but is a general mechanism to avoid instability at AT-rich sequences derived from CFSs.
To examine whether BLM is recruited to CFS-derived AT rich sequences, we performed chromatin immunoprecipitation (ChIP) of Flag-BLM to the Flex1 surrounding regions in the HR-Flex reporter which is stably integrated in the genome using two sets of primers P1 and P2 (Fig 2A). Flag-BLM is efficiently recruited to Flex1 in comparison to the GAPDH locus. APH treatment leads to further enrichment of Flag-BLM to Flex1 (Fig 2B). These data suggest that BLM is localized to Flex1 to suppress Flex1-induced mitotic recombination.
To examine whether BLM prevents DSB formation at CFS-derived AT-rich sequences upon replication stress, we performed ChIP analysis of γH2AX at Flex1 present on pCEP4-Flex1 plasmids with or without BLM depletion. γH2AX is significantly increased at Flex1 after exposure to HU when BLM is inactivated by shRNAs (Fig 2C and S1A Fig). We also performed ChIP analysis of γH2AX at endogenous FRA3B locus close to an AT-rich sequence and found that γH2AX is also significantly enriched there when BLM is depleted by shRNAs after APH treatment (S2 Fig). These data support the model that loss of BLM function induces DSB formation at AT-rich sequences in CFSs, thereby causing increased mitotic recombination at these sequences.
When we placed Flex1 in the Epstein-Barr virus (EBV) replication origin-containing plasmids (pCEP4-Flex1) that are propagated as episomes in mammalian cells, plasmid instability is increased comparing to the plasmids containing the luciferase control sequence (pCEP4-Luc) [21]. Plasmid instability of pCEP4-Flex1 is further increased when we knocked-down BLM by shRNAs (Fig 2D). We recovered pCEP4-Flex1 plasmids after propagating them in U2OS cells with or without expressing BLM-shRNAs for 10 days and sequenced Flex1. We found that deletions present at Flex1 on the pCEP4-Flex1 plasmids recovered from BLM-shRNAs expressing cells are significantly increased compared to that from control cells. While most deletions are present in the AT-dinucleotide region at Flex1, some deletions are extended outside of the AT-dinucleotide region (Fig 2E). These data support the idea that BLM plays an important role in protection of CFS-derived AT-rich sequences by preventing DSB formation. Instability of Flex1-containing plasmids observed in BLM depleted cells is likely caused by DSB formation and deletions present on the recovered pCEP4-Flex1 plasmids are likely the products of end joining at DSBs generated in Flex1.
BLM unwinds DNA secondary structures, such as G4 DNA [29]. We hypothesize that BLM uses its helicase activity to remove DNA secondary structures formed at AT-rich sequences, thus inhibiting DSB formation and mitotic recombination (See Discussion). We expressed Flag-tagged BLM wild-type allele (WT) and helicase mutant allele K695A/D795A (KD) [36] carrying silent mutations at the shRNA targeting sites in U2OS (HR-Flex) cells. Mitotic recombination was determined after delivery of shRNAs to inhibit the expression of endogenous BLM. BLM-WT, but not BLM-KD mutant suppresses Flex1-induced mitotic recombination (Fig 3A). Loss of BLM helicase activity does not influence its recruitment to Flex1 as revealed by ChIP analysis (Fig 3B). These data suggest that the BLM helicase activity is not required for its recruitment to Flex1, but is important for the suppression of Flex1-induced mitotic recombination.
To examine whether DSB formation is increased at Flex1 when BLM helicase activity is inhibited, we performed ChIP of γH2AX at Flex1 in U2OS (pCEP4-Flex1) cells expressing BLM-WT or BLM-KD helicase mutant. Indeed, loss of BLM helicase activity increases DSB formation at Flex1 (Fig 3C and S1B Fig). Loss of BLM helicase activity also increases pCEP4-Flex1 plasmid instability (Fig 3D). We propose that the BLM helicase activity is needed for unwinding DNA secondary structures formed at Flex1, which is important for preventing DSB formation at Flex1 and inhibiting Flex1-induced mitotic recombination (See Discussion).
BLM is phosphorylated by ATR at Thr99 and Thr122 sites upon replication stress [37]. To show whether ATR-mediated phosphorylation of BLM is important for suppressing Flex1-induced mitotic recombination, we expressed BLM-WT and BLM-T99A/T122A mutant in U2OS (HR-Flex) cells with endogenous BLM depleted by shRNAs. Spontaneous recombination at Flex1 is significantly higher in the BLM-T99A/T122A mutant cell line compared to that in the BLM-WT cell line (Fig 4A), suggesting that ATR-mediated phosphorylation of BLM is important for maintaining Flex1 stability. ChIP analysis showed that the recruitment of BLM-T99A/T122A mutant to Flex1 site is as efficient as that of BLM-WT (Fig 4B), indicating that ATR-mediated phosphorylation of BLM does not influence BLM recruitment to Flex1. However, ChIP of γH2AX showed that DSB formation is increased at Flex1 in the BLM-T99A/T122A mutant cell line compared to the BLM-WT cell line (Fig 4C and S1C Fig). pCEP4-Flex1 plasmid instability is also increased in the BLM-T99A/T122A mutant cell line (Fig 3D). These data suggest that ATR-mediated phosphorylation of BLM is important for protecting AT-rich sequences at CFSs to avoid DSB formation.
Flex1-induced mitotic recombination is increased upon HU treatment [21], which is further increased upon loss of BLM activity (Fig 1C). This suggests an important role of BLM in suppressing Flex1 instability upon replication stress. Since oncogene expression often leads to replication stress [38, 39], we examined whether BLM is important for maintaining Flex1 stability upon oncogene expression. Indeed, inactivation of BLM significantly increases mitotic recombination upon overexpression of H-Ras-V12 (Ras) (Fig 5A).
It has been shown that oncogene expression induces CFS expression [40]. We showed that γH2AX is significantly increased at Flex1 when Ras is expressed with a further increase when BLM is depleted by shRNAs (Fig 5B). By performing FISH analysis on mitotic chromosome, we showed that inactivation of BLM results in increased FRA16D expression after Ras expression (Fig 5C). Collectively, our study suggests that BLM is important for protecting AT-rich sequences at CFSs and prevents CFS expression upon replication stress induced by oncogene expression.
We showed previously that FANCM is important for maintaining Flex1 stability and its translocase activity is required for such activity [23]. Since BLM binds to FANCM through RMI1 [41, 42], we asked whether BLM protects Flex1 by modulating FANCM activity. We examined mitotic recombination at Flex1 when BLM and FANCM are inactivated individually or simultaneously (Fig 6A). Inactivation of BLM alone significantly increases Flex1-induced mitotic recombination, but the extent is less than that after FANCM inactivation. Simultaneous inactivation of BLM and FANCM causes hyper mitotic recombination at the level higher than inactivation of either BLM or FANCM alone. These data suggest that BLM and FANCM are not epistatic to each other for protecting Flex1 stability and that the mechanism utilized by BLM to protect Flex1 is not through the protection pathway mediated by FANCM. In consistent with a non-overlapping function of BLM and FANCM in protecting Flex1, the FANCM-MM2 mutant defective in BLM binding [S3 Fig, [41]] does not show a defect in suppression of Flex1-induced mitotic recombination compared to FANCM WT (Fig 6B). This suggests that the BLM function in protecting Flex1 is not through interacting with FANCM.
We also monitored DSB formation at Flex1 when BLM or FANCM is inactivated. ChIP analysis of γH2AX at Flex1 showed that inactivation of both BLM and FANCM causes more DSB formation than when either of them is inactivated after HU treatment (Fig 6C and S1D Fig). This study suggests that BLM plays a distinct role from FANCM and functions together with FANCM to prevent DSB formation at Flex1.
It was described that FANCM and BLM are recruited to telomeres in cells which specifically use the alternative lengthening of telomeres (ALT) mechanism to maintain telomere integrity, and depletion of BLM and FANCM leads to synthetic lethality in ALT cells [43]. These studies suggest a specific role of BLM and FANCM in protecting ALT telomeres and ALT cell viability. However, when we inactivated BLM in FANCM knockout (KO) telomerase positive and non-ALT HCT116 cells, cell viability was also drastically decreased (Fig 7A). These data suggest that the synthetic lethality interaction of BLM and FANCM is not only due to their cooperative functions at ALT telomeres and is likely related to a more general role in coping with replication stress. Our findings that BLM and FANCM play distinct roles in maintaining stability of AT-rich sequences at CFSs may also contribute to the synthetic lethality interactions of these two proteins.
We tested further whether the helicase activity of BLM is important for cell viability of FANCM deficient cells. We expressed BLM-WT or BLM-KD helicase mutant in FANCM WT or FANCM KO HCT116 cells, followed by inactivation of endogenous BLM by shRNAs. Loss of BLM helicase activity drastically reduces cell viability of FANCM KO HCT116 cells while having minimal effect on FANCM WT cells (Fig 7B), suggesting that the BLM helicase activity is essential for cells to survive when FANCM is deficient in non-ALT cells.
CFSs are hotspots for chromosomal rearrangement and are often mapped to breakpoints in cancer cells [2, 44]. We identified a new role of BLM in protection of CFSs and showed that the helicase activity of BLM and ATR-mediated phosphorylation of BLM are important for maintaining stability of AT-rich sequences at CFSs. We also showed that the role of BLM in preventing DSB formation and mitotic recombination at CFS-derived AT-rich sequences is non-overlapping with that of FANCM. The synthetic lethality interaction of BLM and FANCM is observed in both ALT and non-ALT cells and loss of BLM helicase activity in FANCM deficient cells leads to cell death. These studies reveal distinct and cooperative functions of BLM helicase and FANCM translocase to maintain genome stability at CFSs. The role of BLM in the protection of CFSs would likely contribute to its tumor suppression function.
CFS instability is caused by multiple mechanisms and the presence of AT-rich sequences which are prone to forming secondary structures is one of the causes for CFS expression [45]. Upon replication stress, forks are often stalled at AT-rich sequences at CFSs, leading to fork collapse and DSB formation [18, 19, 21]. We showed that BLM is important for preventing DSB formation and mitotic recombination at AT-rich sequences derived from CFSs.
In the HR-Flex reporter, if DSBs are generated at Flex1 and are repaired by HR, EGFP open reading frame would be restored to produce green cells [21]. Our results showed that inactivation of BLM leads to increased HR-mediated mitotic recombination at Flex1. BLM functions together with DNA2 in end resection during HR [33, 34] and impaired function of BLM is expected to reduce HR efficiency. However, besides BLM/DNA2, Exo1 is also involved in long-range end resection and thus BLM is not absolutely required for HR [33]. Thus, elevated mitotic recombination detected in our reporter in BLM deficient cells indicates that the effect caused by increased DSB formation at Flex1 exceeds the effect of HR reduction due to end resection defect caused by loss of BLM. The role of BLM in protecting CFS-AT is likely underestimated by using the HR-Flex reporter. γH2AX ChIP analysis indicated that DSB formation is increased at Flex1 when BLM is depleted. Therefore, these studies suggest that BLM plays an important role in protecting AT-rich sequences derived from CFSs and suppressing DSB formation at these structure-forming DNA sequences.
BLM is a 3’ to 5’ helicase [27]. Biochemically, BLM is incapable to unwind duplex DNA from blunt-ended terminus or from an internal nick [29]. However, BLM unwinds the bubble and X-structure substrates much more efficiently than 3’-tailed duplex. BLM also efficiently unwinds G4 DNA structures. BLM is thus considered as a DNA structure-specific helicase. CFS-derived AT-rich sequences are also predicted to form strong secondary structures by the Mfold program [S4 Fig, [18, 46]]. We propose that BLM plays an important role in unwinding DNA secondary structures at CFS-derived AT-rich sequences. Upon replication stress, DNA secondary structures would form at CFS-derived AT-rich sequences and stall DNA replication, leading to fork collapse and DSB formation (Fig 7C). Removing DNA secondary structures by BLM unwinding activity would thus prevent fork stalling and DSB formation. The requirement of BLM helicase activity for preventing DSB formation and mitotic recombination at Flex1 supports this model.
We showed that FANCM is important for maintaining stability of CFS-AT sequences in a manner dependent on its translocase activity [23]. FANCM interacts with BLM through RMI1 and this interaction is important for repairing interstrand crosslinks (ICLs) [41]. We showed that the FANCM-MM2 mutant defective for BLM interaction inhibits Flex1-induced mitotic recombination to a similar level as FANCM-WT, suggesting that the role of BLM in suppression of Flex1 instability is not due to its interaction with FANCM. We further showed that mitotic recombination and DSB formation at Flex1 are both increased when BLM and FANCM are simultaneously inactivated compared to BLM or FANCM single deficient cells, suggesting that the role of BLM in preventing instability at Flex1 is distinct and non-overlapping with that of FANCM. These studies support the model that BLM uses mechanisms different from FANCM to protect CFS-AT sequences.
FANCM contains ATP-dependent branch point translocase activity and promotes replication fork regression [47, 48]. We propose that FANCM uses its fork remodeling activity to promote fork regression to remove DNA secondary structures at Flex1 during replication and upon replication stress [23], whereas BLM may directly bind to loops or other secondary structures formed at CFS-AT vicinity to unwind these structures (Fig 7C). Along this line, it has been shown by single molecule FRET analysis that BLM binds to single-stranded DNA at the side of G4 structures to unfold G4 [49]. Unwinding activity mediated by BLM and fork regression activity promoted by FANCM may complement to each other to effectively remove DNA secondary structures formed at AT-rich sequences. On the other hand, BLM can also catalyze replication fork regression [50]. An alternative mechanism could be that FANCM initiates fork reversal to remove DNA secondary structures at forks, but more extensive fork regression may need BLM helicase activity coupled with decatenation activity of Topo IIIα that is bound with BLM [51, 52].
Loss of FANCM or BLM activity both increases mitotic recombination at Flex1, suggesting that HR is used to repair DSBs accumulated at CFS-derived AT-rich sequences when either FANCM or BLM branch is deficient. In support of this, we showed that CtIP and BRCA1 are required for mediating increased mitotic recombination not only when FANCM is defective [23], but also when BLM function is impaired (S5 Fig). Therefore, HR backs up both FANCM and BLM pathways for protecting structure-prone AT-rich sequences. However, when both FANCM and BLM are defective, substantial DSBs are accumulated and HR may not be sufficient to repair all DSBs, thereby leading to cell death.
It has been described that FANCM and BLM are synthetic lethal in ALT cells, which is caused by impaired telomere replication due to FANCM deficiency, combined with impaired HR function when BLM is defective [43]. We showed that simultaneous inactivation of BLM and FANCM in non-ALT cells also causes cell death, suggesting that the synthetic lethality interaction of FANCM and BLM is not limited to ALT cells and the concerted activities of FANCM and BLM critical for maintaining cell viability maybe more than their involvement in the protection of ALT telomeres.
Our study demonstrates that both FANCM and BLM are important for maintaining stability of structure-forming AT-rich sequences derived from CFSs. We propose that the non-overlapping functions of BLM and FANCM to resolve DNA secondary sequences at replication forks could be one of the important causes for their synthetic lethality interactions (Fig 7C). DNA secondary structures can arise not only at AT-rich sequences in CFSs but also at other structure-prone DNA sequences such as G4s. Computational study predicts more than 700,000 G4 structures existing in the human genome [53]. If the cooperative functions of BLM and FANCM are required for resolving not only CFS-AT but also other DNA secondary structures that are abundantly present in the human genome, it is not surprising that BLM and FANCM have a synthetic lethality interaction. The role of FANCM and BLM in ALT telomere protection can also be attributed to their functions in maintaining stability of structure-forming DNA sequences present at telomeres. Mammalian telomeres are enriched in guanines and are prone to the formation of G4s [54, 55]. It is proposed that the function of FANCM to prevent replication stalling at telomeres is due to its activity to remove G4 secondary structures during telomere replication [43]. Since BLM unwinds G4 in vitro [29, 49], it may also be involved in maintaining G4 stability at telomeres in vivo. Meanwhile, we also acknowledge that the role of BLM in HR may additionally contribute to the synthetic lethality interaction of BLM and FANCM. Fork collapse at DNA secondary structures would require HR-mediated repair and replication restart that involve BLM for end resection.
We showed that ATR-mediated phosphorylation of BLM is important for maintaining Flex1 stability. It is possible that ATR-dependent phosphorylation of BLM stimulates BLM helicase activity, thereby contributing to Flex1 protection. Alternatively, phosphorylation of BLM by ATR may modulate the way of BLM binding to Flex1, and thus influence Flex1 unwinding by BLM. In this aspect, ChIP analysis showed that the recruitment of the ATR phosphorylation mutant BLM-T99A/T122A to Flex1 is largely unchanged. However, it is still possible that ATR-mediated phosphorylation of BLM may influence the affinity of BLM for direct Flex1 binding after its recruitment or modulate the conformation of BLM in complex with Flex1 resulting in change of unwinding activity.
It was shown that the BLM helicase activity and ATR-mediated phosphorylation of BLM are both required for replication restart upon replication stress, but the underlying mechanisms are not clear [37]. We found that both helicase activity and ATR-mediated phosphorylation are important for maintaining CFS-AT stability. We speculate that the intrinsic activity of BLM to remove DNA secondary structures is also used at other genomic loci such as G4s which are very abundant in the genome. Since maintaining fork stability is important for replication restart, we propose that one mechanism involving BLM helicase activity and ATR-mediated phosphorylation in replication restart is linked to its function in resolving DNA secondary structures at replication forks and preventing fork collapse.
BLM is a multifunctional protein which couples recombinational repair with replication fork protection. Our study reveals its new function in conjugation with FANCM to protect stability of AT-rich sequences in CFSs to maintain CFS stability. This role along with its function in end resection at early steps of HR and dissolution to prevent crossover at late steps of HR [56], may underlie the mechanisms of how BLM contributes to the maintenance of genome stability and prevention of cancer.
Human U2OS, 293T and HCT116 cells were cultured at 37°C in a humidified atmosphere of 5% CO2 in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin. FANCM knockout HCT116 cells were previously described [57].
BLM wild-type and indicated mutants were generated by PCR and subcloned into Babe-puro vectors containing 1x N-terminal Flag-tag or NBLV0051 (Novo Bio) vector containing a 3x N-terminal Flag-tag. FANCM wild-type and FANCM-MM2 mutants were also subcloned into NBLV0051 (Novo Bio) vector containing a 3x N-terminal Flag-tag. Point mutations, small hairpin RNA (shRNA) target site-resistant mutations were generated using the QuikChange Site-directed Mutagenesis Kit (Stratagene). The EGFP-based HR-Flex reporter and HR-Luc reporter containing the 0.34 kb Flex1 and luciferase-derived fragment (Luc), respectively, and pCEP4-Flex1 and pCEP4-Luc plasmids were described previously [21]. EGFP-based HR reporter: HR-16C/AT1 and HR-16C/AT3, containing AT-rich sequences derived from FRA16C reporters were described previously [23].
Silencing of endogenous BLM and FANCM was achieved by retroviral or lentiviral infection using pMKO or pLKO vectors to express corresponding shRNAs [58, 59]. shRNA sequences for BLM (GAGCACAUCUGUAAAUUAAUU) and FANCM (GAACAAGAUUCCUCAUUACUU) were designed by Dharmacon.
Western blot analysis was performed as described [60]. Cells were lysed with NETN buffer (20 mM Tris, pH 8.0, 1 mM EDTA, 150 mM NaCl, and 0.5% Nonidet P-40) for 30 min. Cell lysates were boiled in 2xSDS loading buffer and subjected to SDS-PAGE.
Commercial antibodies used include anti-BLM (Bethyl Laboratories, A300), anti-H2AX-S139p (Cell Signaling, #2577), anti-Ras (Santa Cruz Biotechnology, sc-520), anti-FLAG (Sigma, F1804), anti-Ku70 (Santa Cruz Biotechnology, sc-17789), anti-β-actin (Sigma, A5441). Antibodies against FANCM was kindly provided by Dr. Weidong Wang [57].
Spontaneous mitotic recombination assay was described before [21]. Briefly, cells carrying HR-Flex or HR-Luc reporter were pre-sorted by FACS (fluorescence-activated cell sorting) to clear background and cultured for indicated days. The mitotic recombination frequency (EGFP-positive events) then was determined by FACS analysis using a BD Accuri C6 flow cytometer and accompanying data analysis software (CFlow, Becton-Dickinson).
Plasmid stability assay was performed as described [21]. Briefly, Flex1 or Luc containing pCEP4 plasmids carrying Epstein-Barr virus (EBV) replication origins and nuclear antigen (encoded by the EBNA-1 gene) to permit extrachromosomal replication in human cells and a hygromycin marker [21], were transfected into the cells expressing wide-type or BLM mutants by lipofectamine 2000 (Thermo Fisher Scientific), followed by hygromycin selection. The endogenous BLM was knocked down by shRNA, and cells were cultured in the absence of hygromycin for 10 days. Then hygromycin was re-introduced into the medium to determine the percentage of cells retaining the plasmids. Sequencing of Flex1 on pCEP4-Flex1 was performed after propagating pCEP4-Flex1 10 days in U2OS cells with or without expressing BLM-shRNAs.
Cell proliferation was determined by counting cells every 24 hours using hemocytometer and normalized to the first day. Cell viability was shown by colony formation. HCT116 and its derivative cell lines were plated at 5000 cells per 10 cm plate and were grown in complete media for two weeks. Cell colonies were fixed with cold methanol and stained with 1% crystal violet.
ChIP assays were performed as described previously [21]. Briefly, cells were subjected to either no treatment or treatments with APH, followed by cross-linking with 1% formaldehyde for 10 min at room temperature. The cross-linking reaction was terminated by adding glycine to a final concentration of 125 mM and incubating for 5 min. After washing twice with cold PBS, cells were resuspended in lysis buffer (1% SDS, 10 mM EDTA, 50 mM Tris.HCl, pH 8.1) supplemented with protease inhibitor cocktail (“PIC”, cOmplete, Roche) and the samples were subject to sonication to break chromatin into fragments with an average length of 0.2–1.0 kb. The supernatants were collected by centrifugation and were pre-cleared with Protein A/G Sepharose beads (Amersham Biosciences). Immunoprecipitation (IP) was performed using H2AX-S139p antibody (Cell Signaling #2577) or Anti-Flag antibody (Sigma F1804) followed by washing with TSE I (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris.HCl, pH 8.1, 150 mM NaCl), TSE II (0.1% SDS, 1% Triton X-100, 2 mM EDTA, 20 mM Tris.HCl, pH 8.1, 500 mM NaCl), buffer III (0.25 M LiCl, 1% NP-40, 1% deoxycholate, 1 mM EDTA, 10 mM Tris.HCl, pH 8.1), and TE. The protein-DNA complex was then eluted from beads by elution buffer (1% SDS, 0.1 M NaHCO3), and cross-linking was reversed by adding in 4 μl of 5 M NaCl and incubating at 65°C for 6 hr, followed by proteinase K digestion for 2 hr at 42°C. DNA was extracted by QIAquick kit (QIAGEN) according to the manufacture instructions.
For ChIP at Flex1 in the HR reporter stably integrated in the genome or at AT-rich sequences in the endogenous FRA3B locus, recovered DNA was analyzed by quantitative PCR (qPCR) and the readout was normalized to the vector control which is set as 1. GAPDH locus was used as a control to show the specificity of protein binding to Flex1. The primers used for ChIP at Flex1 in HR reporter: P1F 5’CTCCAATTCGCCCTATAGTGAGTCGTATTA, P1R 5’TTACTTGTACAGCTCGTCCATGC, P2F 5’GGCAGTACATCAATGGGCGTG, and P2R 5'CCTTTAGTGAGGGTTAATTGCGCG; at AT-rich sequences in FRA3B: 3B-P1F 5’TTAGCCTACTTCAGGGTTTCT and 3B-P1R 5’TGGAGAGGTTACTACTGGCA; and at GAPDH: GAPDH-F 5’CCCTCTGGTGGTGGCCCCTT and GAPDH-R 5’GGCGCCCAGACA CCCAATCC.
For γH2AX ChIP at Flex1 or Luc surrounding regions on the pCEP4-Flex1 or pCEP4-Luc plasmids, recovered DNA was amplified by regular PCR or qPCR with primers 5’TCAGGGGGAGGTGTGGGAGG and 5’GCAGTCCACAGACTGCAAAG. Regular PCR was programmed as preheat at 95°C for 3 min followed by 35 cycles of 95°C 15 sec, 50°C 15 sec, 70°C 30sec. Regular PCR products were resolved by 1.5% agarose DNA gel [21]. Alternatively, qPCR was performed for γH2AX ChIP.
Fragile site analysis using fluorescence in situ hybridization (FISH) was performed as described [21]. Briefly, cells were treated with 0.4 μM APH for 18 hr to induce CFS expression, followed by incubation of 0.1 μg/ml colcemid at 37°C for 45 min as described in the standard method for chromosome preparations. Collected cells were resuspended in 75 mM KCl hypotonic solution pre-warmed to 37°C and incubated at 37°C for 30 min, followed by several changes of fixative solution (3:1 methanol/acetic acid). Cells were dropped onto slides and incubated for 2 hr at 60°C prior to FISH analysis. FISH experiments were performed according to standard protocols [61]. Green 5-Fluorescein dUTP-labeled probes 264L1 (FRA16D) from RPCI-11 human BAC library (Empire Genomics) were used as probes for FISH analyses. Chromosomes were counterstained with DAPI.
All the statistical data are the results of at least three independent experiments and presented as mean±SD. Numerical data for all figures are included in S1 Data.
|
10.1371/journal.pcbi.1004883 | Modularity Induced Gating and Delays in Neuronal Networks | Neural networks, despite their highly interconnected nature, exhibit distinctly localized and gated activation. Modularity, a distinctive feature of neural networks, has been recently proposed as an important parameter determining the manner by which networks support activity propagation. Here we use an engineered biological model, consisting of engineered rat cortical neurons, to study the role of modular topology in gating the activity between cell populations. We show that pairs of connected modules support conditional propagation (transmitting stronger bursts with higher probability), long delays and propagation asymmetry. Moreover, large modular networks manifest diverse patterns of both local and global activation. Blocking inhibition decreased activity diversity and replaced it with highly consistent transmission patterns. By independently controlling modularity and disinhibition, experimentally and in a model, we pose that modular topology is an important parameter affecting activation localization and is instrumental for population-level gating by disinhibition.
| The capacity to transmit information between connected parts of a neuronal network is fundamental to its function. The organization of network connections (the topology of the network) is therefore expected to play an important role in determining network transmission. Since modular topology characterizes many brain circuits on multiple scales, investigating the role of modularity in activity gating is clearly desirable. By engineering such modular networks in vitro, we were able to perform such an investigation. Under these experimental conditions, we can independently control the degree of modularity, as well as inhibition in the network. We show that a combination of these two properties is highly beneficial from a communication perspective. Namely, it equips connected modules and large modular networks with the capacity to gate and temporally coordinate activity between the different parts of the network.
| Activity gating and control over propagation are fundamental capacities of neural circuits. It is widely accepted that population-level gating is strongly affected by changing the balance between excitation and inhibition in connected sub-populations of neurons [1–6]. However, while the role of excitation-inhibition has been widely investigated, the contribution of circuit topology to activity gating has received much less attention. Modular topology is of particular interest, as it is a fundamental feature of biological neuronal circuits [7–11]. Modular circuits are composed of highly connected groups of neurons (modules) which are loosely connected to other groups. Such a network organization is found at many spatial scales, ranging from anatomically defined brain regions to groups of neurons [7,8,12,13].
Experimental investigation into the contribution of modular topology to gating phenomena is faced with major challenges. While large brain areas have well-documented connectivity and activity maps [12,14,15], accessing complete brain circuits at smaller scales is still limited. Foremost, the connectivity maps are highly untraceable in the three-dimensional architecture of the tissue. In addition, since cell assemblies are often dispersed in space, their simultaneous identification and recording are still beyond the reach of contemporary technologies [16]. Finally, as neuronal circuits are not prone to design, systematic studies are impossible. Consequently, studies aimed to relate activity propagation and gating to network architecture (such as modular networks) are mainly restricted to theoretical investigations [17].
Indeed, theoretical studies indicate that modular organization greatly impacts network functionality [18]. Modular circuits provide control over activity propagation [9,10,19,20], time-scale separation [21], dynamical complexity and the computational capacity of the network [10,22–25].
In this study we experimentally address, for the first time, the relation between circuit modularity and activity gating. To overcome the inherent limitations associated with studying intact tissues, we utilized a cell patterning technique to induce self-organization of modular networks in culture. We found that pairs of connected modules support conditional propagation that is dependent on the activity intensity in the sending module. In large networks of many connected modules, conditional propagation enhances the diversity of activation patterns, and is manifested as events initiated at different modules which then propagate to different distances. Interestingly, blocking network inhibition decreased the activity diversity and replaced the conditional propagation with highly reliable transmission. These features are absent in the activity repertoire of uniform networks. Thus, we show how a combination of modular circuit architecture and disinhibition supports gating.
For the sake of clarity, the results are organized according to systems with growing complexity. We begin by addressing the properties of connected cluster pairs. We then address chains of more than two clusters. Next, we look at the activation repertoires of larger systems (networks of connected clusters). Finally, we attempt to explain our results with a computational model addressing the accumulative effects of disinhibition and modular topology.
To investigate how activity propagates through modular neuronal networks, we used a unique biological model system of engineered clusters in vitro. Control over circuit architecture was achieved using heterogeneous surfaces with different degrees of adhesiveness (see Materials and Methods). Specifically, islands of highly adhesive surfaces were realized on a non-adhesive background. Due to their innate propensity to cluster, neurons self-organized into modular circuits within several days in culture (Fig 1), in accordance with previous work [26–28]. Each of the clusters comprised of several tens to hundreds of neurons (S1A Fig), connected through a bundle of fasciculated neurites (Fig 1). The number of cells per cluster was estimated from the cluster area using N = 0.0079S-1.9, where N is the number of cells and S is the cluster area in μm2. This relation was calculated in a previous publication in which clusters were grown under the same experimental conditions [29]. In contrast to previous methods [30–33], our procedure allows the formation of small-scale sub-networks of different sizes, ranging from cluster chains (Fig 1B) to two-dimensional networks of many connected clusters (Fig 1C). The position of each cluster was aligned with a micro electrode, allowing local electrical recording from each cluster (Fig 1B and 1C-right).
Using the embedded electrodes (Fig 2A), the simultaneous activity of all clusters was recorded. The vast majority of measured clusters were found to be spontaneously active after several days in culture. A typical voltage trace recorded from one electrode is shown in Fig 2B. Each electrode recorded the superimposed activity of many neurons within each cluster. It was previously shown that voltage traces represent spike summation and correspond to the increase and decrease in population activity [29,34]. Accordingly, to represent the activity intensity of such clusters, we averaged the rectified voltage traces over short time windows and color coded them (Fig 2C) (see Materials and Methods for details). Spontaneous activity of individual clusters was characterized by typical features previously observed in developing networks [29,35,36]. We observed the activation of network bursts (NBs) which are short epochs of network intense firing, separated by longer periods of sporadic activity (Fig 2D). To verify that these events represent the collective activity of many neurons that synchronize within the NB time window, we performed recordings from single clusters using dense electrode arrays (30 μm spacing) with a smaller electrode surface area (314 μm2 in contrast to 2827 μm2 in our regular electrodes) which pick up activity from more local populations of neurons. The variability in the spiking profiles across different electrodes exemplified the synchronized nature of spiking within NBs of single clusters (S2 Fig). While the activity within clusters appeared to be synchronized, between connected clusters the synchrony was transient.
We begin by examining the activity of connected cluster pairs (Fig 3A). NBs were found to be either confined to a single cluster or spread over nearby connecting clusters (Fig 3B). To investigate whether the activity propagation between connected clusters was related to the activity intensity in the clusters, we examined the propagation between pairs of adjacent clusters. For each pair, we defined one cluster as the "sending cluster" and detected all NBs occurring in this cluster (see details in Materials and Methods). From this NB pool, we selected only NBs which were activated in the sending cluster before the neighboring cluster, defined here as the "receiving cluster". This selection process consisted of rejecting NBs according to the delay calculated from the peak position of the cross-correlation function of the smoothed (convoluted with a Gaussian kernel, σ = 10ms) activity intensities of the sending and receiving clusters. Positive offset delay was attributed to propagation from a receiving to a sending cluster. Fig 3A–3C illustrates activity propagation in a representative cluster pair. 300 consecutive NB traces, recorded from cluster 1 (the sending cluster) in Fig 3A, are shown in Fig 3C-left. These NBs were reordered according to increased NB intensity (sum over the NB activity intensity trace) in the sending cluster. The receiving cluster responses to these NBs are shown in Fig 3C-right. The total intensity of these responses (averaged over all NBs) were only slightly lower than in the sending cluster (Z-score of the difference in AI was 0.083), but significant (PV = 0.013, Mann-Whitney-Wilcoxon test). Low intensity NBs did not propagate to the receiving cluster, while strong NBs did. It should be noted, however, that a small fraction of strong NBs failed to propagate. We termed this selective activation of NBs in the receiving cluster following an NB in the sending cluster as conditional propagation.
To quantify this behavior over many cluster pairs, we calculated for each NB in the selected pool, and for each cluster pair, the probability of having different normalized NB intensities (intensities divided by the intensity standard deviation) in the receiving cluster as a function of normalized NB intensities in the sending cluster. Data of the cluster pair in Fig 3A are presented in Fig 3D-left. This representation further illustrates that low intensity NBs did not yield strong responses in the receiving cluster, while NBs stronger than a certain value successfully propagated to the receiving cluster. It is important to note that in the example shown here, a distinct threshold between non-propagating and propagating NBs is apparent (Fig 3D-left). To quantify this threshold-like behavior, we calculated the bi-modality measure [37] on the distribution of normalized NB intensities projected on the identity line (S4A Fig) for the cluster pair in Fig 3D-left (S4B Fig), and for all cluster pairs (S4C Fig). Most clusters showed values larger than (0.555) corresponding to a tendency to bi-modality over uni-modality (S4C Fig). We verified these results by performing visual inspection of the distributions and determined that nearly half (48%) of the pairs showed a clear intensity threshold in the propagation probability. However, taking into account also pairs that did not show strong bi-modality, the general rule was that strong activations in the sending cluster yielded strong responses in the receiving cluster and vice versa. This effect was quantified by calculating the correlation between the normalized NB intensity of the receiving and sending clusters for all NBs in 163 cluster pairs, from 26 cultures (Fig 3D-right). Two types of cluster networks were considered in the analysis: The first type is one dimensional chain of clusters and the second involves a cluster chain with one of the clusters connected to a larger network (of many connected clusters). Within these networks, only pairs connected exclusively through a neurite bundle (and not through any other pathway) were analyzed. The data in Fig 3D-right exhibits a clear preference towards positive values, further suggesting that information about firing intensity is utilized by neuronal networks to control propagation between connected sub-populations. No significant correlation was observed between NB intensity correlation strength and the normalized difference in cluster cell numbers (C = 0.084, PV = 0.28).
We next sought to investigate how modularity affects propagation delays. Delays in our engineered networks were evaluated by extracting the location of the peak in the cross-correlation function of every pair and every NB (as previously described). Only NBs propagating from the sending cluster to the receiving cluster were considered. The delay distribution across all NBs is shown in Fig 3E-left. The average delay was several tens of milliseconds (red line in Fig 3E-left) and the maximal value reached was 250 ms. With a distance between clusters of 500 μm, a delay of 100 ms corresponds to a propagation speed of 5 μm/ms. Long delays were also observed in other cluster pairs, as shown by the distribution of average delays (Fig 3E-right). We did not find a significant correlation (C = 0.060, PV = 0.43) between the average delay and the normalized absolute difference in cell numbers (
|N1−N2|2(N1+N2)
, Ni being the number of cells in each cluster). Interestingly, a weak positive correlation was found between average delay and the number of cells in the receiving cluster (C = 0.202, PV = 0.007)(S1C Fig), but not in the sending cluster (C = 0.098, PV = 0.19), suggesting that the delay is associated with the network recruitment time in the receiving cluster. Correspondingly, the average recruitment time for all NBs in a cluster, measured as the time between NB onset and NB peak (see NB detection in Materials and Methods), had similar time scales to the observed average delays (S3 Fig).
Fig 4A shows an example of 100 consecutive NBs recorded from a three cluster chain (only NBs in which all three clusters were active are shown). Visual inspection clarified that most of the time cluster 2 fired before cluster 3, indicating that propagation in cluster chains is asymmetric. To examine the different propagation patterns and their abundance in this network, we clustered them by calculating the similarity matrix between NBs using a dendrogram [38] (see Materials and Methods) (Fig 4B). For clarity, only NBs in which at least two clusters were active are presented. Interestingly, clear NB groups with high similarity are observed (marked by solid line rectangles in Fig 4B). These different groups correspond to different propagation patterns as indicated by the average NB profile within each group (Fig 4C). Propagation in this modular network, although bi-directional, is rarely symmetric.
The non-uniformity of the similarity matrix within each rectangle represents NB pattern variation within each group, as seen by examining the coefficient of variation of off-diagonal terms within a group (0.21, 0.04, 0.21, 0.03 and 0.07 in groups 1 to 5 respectively). The higher coefficient of variation in groups involving the activation of cluster 1 suggests that specific links in the network are more variable than others. For example, groups 2A and 2B (see Fig 4C) are characterized by the same propagation directions, but with a different average delay (between clusters 1 and 2). The activation probability of different propagation patterns vary considerably (P = 0.137, 0.673, 0.103, 0.005, 0.082 for groups 1 to 5 respectively). Such variability can be represented by the pattern entropy (E = −∑iPilog2Pi = 1.006 for this example) as discussed in detail in the next section.
Clustering into propagation groups existed in all network examined (69 chains form 26 cultures). However, this grouping was highly variable between networks both in terms of separability between groups and the number of groups. Consequently, to reliably quantify asymmetry, we needed a measure that reduces the complexity of patterns to simple asymmetric relations between pairs. To do so, we used the cross-correlation function between connected clusters over long recordings of eight hours (Fig 4D). We subtracted the integral over the positive side of the cross-correlation function from its negative side and divided it by the total sum. We analyzed only positive correlations between clusters, thus negative cross-correlation values were not included in the integral. The resulting long-term asymmetry is a measure between -1 and 1 with the sign determining the propagation direction and the value corresponding to the asymmetry magnitude. For example, in Fig 4A–4C, propagation pattern 2 dominated the network's activity (Fig 4B). This pattern represents events initiated in cluster #2 and propagated to the neighboring clusters. Correspondingly, the asymmetry measure for cluster 1 and 2 is -0.21 and for cluster 2 and 3 is 0.31. Pooling over all cluster pairs (n = 89) we find that the average absolute asymmetry was 0.26±0.21 (mean ± standard deviation). Such asymmetry is in accordance with previous reports indicating that activity asymmetry exists between connected sub-populations, even if they are very similar to each other [32]. We next examined if asymmetry is affected by the relative differences in cell number between connected clusters. Although the differences between cell numbers were not large (S1B Fig), we found a small positive correlation between the normalized cell count and the asymmetry value (C = 0.331, PV = 0.001) (S1D Fig).
An interesting question that arises from these results is whether and how activity asymmetry is affected by asymmetry in the cluster pair, and by the manner in which this cluster pair is connected to the rest of the network. To address this question, we applied the asymmetry measure to pairs of clusters in which only one of the clusters was connected to a larger clustered network. Such pairs had a clear structural asymmetry. A schematic drawing of such pairs is shown in the inset of Fig 4E, where c1 and c2 are the two clusters and x denotes a group of connected clusters. We differentiated between two connectivity patterns according to the number of links (connections to other clusters), n, cluster c2 had. The asymmetry statistics for n = 1 and n>1 are shown in white bars and black bars respectively in Fig 4E. In most networks, the long-term activity asymmetry was negative (Fig 4E), corresponding to activity propagation from the cluster group towards the chain's end (x towards c1 in the schematic drawing in Fig 4E). This result suggests that the structural asymmetry of the embedding network contributed to the functional asymmetry of the cluster pair. Thus, if a cluster in a pair is connected to other clusters, it will more likely drive activity in this pair. Interestingly, the average asymmetry of pairs connected through more than one link was more negative than that of pairs connected through one link (-0.26 and -0.10 respectively, PV = 0.07, t-test), suggesting that the drive is stronger when the structural asymmetry is stronger. We note that asymmetry was calculated on long time series to represent a gross averaged estimation in a valid manner. When inspecting the delay of the cross-correlation function (or the asymmetry) on a single cluster pair, we find that asymmetry is modulated in the network (Fig 4F). This is observed in the ratio between propagation to one and to the opposite direction calculated on one hour windows (black line in Fig 4F). Despite this dynamic change, the average values of asymmetry were mostly negative (Fig 4E).
We further examined whether we can gate activity propagation by global disinhibition. Specifically, whether we can dramatically increase the probability that NB propagates between clusters by applying inhibitory synaptic blockers (Fig 5). Under control conditions (normal growth media), cluster chains exhibited a large spectrum of activation profiles from confined bursts (in single clusters) to network-wide activation (Fig 5A). Upon disinhibition, using a GABAA (γ-aminobutyric acid) channel antagonist (Bicuculline, 30μM), the conditional propagation was replaced by network-wide synchrony (Fig 5B). Careful examination of the propagation patterns within the NBs under disinhibition revealed that not only did the network synchronize to operate as a single unit (Fig 5B), but also the synchrony was characterized by highly ordered propagation patterns (Fig 5D). To monitor these patterns over consecutive NBs, we represented each NB as a vector (as shown in Fig 5C and 5D-right) and plotted it, for consecutive NBs, under control conditions (Fig 5E) and under disinhibition (Fig 5F). In Fig 5E and 5F, blue dots correspond to cluster activation during an NB, and the arrows correspond to the activity propagation direction (extracted from the peak lag of the cross correlation function). Under disinhibition, the network’s activity collapsed to a stereotypic pattern in which the entire network was fully activated and cluster number 5 functioned as an activation focus. Under disinhibition, 95% of the NBs (Ntotal = 4097) were initiated by cluster 5, in contrast to only 4% (Ntotal = 7206) under control conditions. Such a repeated propagation pattern is in contrast with the much wider repertoire observed under control conditions (Fig 5E).
The variability in NB patterns was further quantified over long-term recordings by defining the NB pattern entropy. Each NB was reduced to a binary series of zeros and ones corresponding to the activation of different clusters (blue dots in Fig 5E and 5F). The occurrence probability for every binary pattern, Pi, was calculated and presented as a pie chart for control and disinhibition conditions (Fig 5G). The entropy, E, was calculated from these probabilities as E = −∑iPilog2Pi. Under control conditions, a wide NB pattern distribution and relatively high entropy value (E = 2.32) were observed. However, after disinhibition, the number of patterns dramatically decreased, corresponding to a lower entropy (E = 0.94). For entropy calculations, we used only long enough chains (three or more active clusters) that can support a high enough variability in NB patterns. Since every chain was of a different length and consequently had different potential entropy values, the entropy was normalized to the maximal possible entropy in the chain, Emax = log2(2N − 1) (the term -1 was added to subtract the case in which no cluster was activated). Five out of six chains exhibited a decrease in normalized entropy following disinhibition (Fig 5H). This transition from a wide to a narrow pattern distribution following disinhibition suggests that under control conditions, networks maintain a certain relation between inhibition and excitation. This relation, in combination with the modular architecture, allowed each cluster to be activated autonomously while still being connected to other clusters, and thus also having the potential to activate them. When disinhibited, the network lost its diversity and collapsed into stereotypic global activation. Disinhibition (implemented here using Bicuculline) served as a gating mechanism, altering signal propagation.
As illustrated above, modularity introduces new features to the activity repertoire of uniform networks. To explore whether these features are preserved in networks of many connected clusters, we examined large, two-dimensional networks of connected clusters (Fig 1C) and contrasted their activity with that of large uniform networks (Fig 1A). The most conspicuous difference appeared to be the degree of synchrony. While uniform networks mostly showed network-wide activation that spanned a large fraction of the cell population (Fig 6A-bottom; S5 Fig), clustered networks exhibited NBs of different sizes that propagated to different distances (Fig 6A-top; S5 Fig). This diversity is the direct result of the conditional propagation inherent in the modular bridge between clusters. In large modular networks, the cumulative number of such bridges between any two clusters increased with the distance between them. We next examined whether co-activation of clusters depended on distance. This effect was quantified by measuring the average Pearson correlation between the activity of cluster pairs for long time periods (>8 hours). A decrease in correlation with distance was evident (Fig 6C-top), highlighting the locality of the activation in large modular networks. Neighboring clusters had a higher probability to fire in synchrony. The tendency for local activation was found to coexist with epochs of global activation (Fig 6A and 6B-top) exemplifying the network's potential for activation diversity.
Large clustered networks were also characterized by long delays. Delays between cluster pairs were calculated from the peak of the smoothed cross-correlation function and averaged over all NBs (see Fig 3). These delays accumulated during burst propagation, and in many cases the last cluster to be activated during an NB began its firing long after the first cluster already ceased bursting (Fig 6B-top). Delays ranged from tens to hundreds of milliseconds (corresponding to an average propagation speed of 6.5±0.2 μm/ms, mean±SE), and increased with the number of bridges between clusters (Fig 6C-middle). Time delays partially affected the decrease in correlation with distance (Fig 6C-top). Conditional propagation was also quantified by calculating the transfer probability. Namely, given the occurrence of an NB in one cluster, the probability that an NB occurred within one second in the other cluster. This probability was also found to decrease with distance (Fig 6C-bottom).
Finally, we note that conditional propagation, long delays and high diversity in the network degree of activation and synchrony are much less pronounced in the activity repertoire of large uniform networks. Uniform networks are characterized by large scale network events (S5 Fig). Once a network event is initiated, it quickly propagates (with an average propagation speed of 22.1±1.2 μm/ms, mean±SE) and recruits most of the network (Fig 6A, 6B-bottom; S5 Fig). Uniform topology appears to lead to uniform activation which does not vary much with distance (Fig 6C). Previous studies showed that uniform networks support a mode of partial network activation called aborted bursts, during which only a subset of the population is active [39]. Owing to these aborted bursts (which are also observed in our data) the average transfer probability in uniform networks is well below unity (Fig 6C-bottom). However, in uniform networks such aborted bursts are not confined to a local area in the network and are always activated in the same sub-population of neurons [39]. Consequently, neither the transfer probability nor the delays in uniform networks depend dramatically on distance (Fig 6C-middle).
As shown above, disinhibition drastically increased propagation between the modular units and induced global network synchronization with defined propagation patterns. In large clustered networks, clusters are connected through multiple pathways which may lead to different disinhibition effects. To test the propagation patterns in large clustered networks before and after disinhibition, we examined the cross-correlations function between clusters over single NBs. Since these are two-dimensional networks (unlike the one-dimensional cluster chains discussed above), we extended our propagation analysis by calculating the propagation vector for every cluster. We first identified NB windows in the network (see Materials and Methods), and calculated the cross-correlations of smoothed (convolution with a Gaussian, σ = 50ms) activity traces between every cluster and all of its neighbors, corresponding to the fact that long range connections between the clusters were rare due to the grid-like organization of the clusters. Clusters with very weak activity during the NBs (active for less than 10 ms) were not analyzed and only NBs with at least five active clusters were considered. The cross-correlation between every cluster pair was represented by a vector with a magnitude corresponding to the location of the peak of the cross-correlation function, and a direction determined by the physical direction between the clusters (Fig 7A—gray arrows). The propagation vector was calculated by averaging the cross-correlation vectors over all active clusters during each NB (Fig 7A—blue arrow). The angle of this vector with the positive x-axis was marked by ϴ (Fig 7A). The propagation vector for each cluster during 20 consecutive NBs is plotted in the physical space of the network in Fig 7B (magnitude was set to be the same for all clusters). In agreement with the high diversity of activation patterns in clustered networks, different NBs propagated to different directions. This was also evident by examining a larger pool of NBs (Fig 7D-left). Here the angle of the propagation vector, ϴ, was presented in color code for every cluster during consecutive NBs.
As in the case of the cluster chains, following disinhibition (application of 30 μM Bicuculline), the high NB pattern diversity was replaced by stereotypic patterns. This shift (or gating) in activation profile is represented by the narrow distribution of propagation directions for different clusters over consecutive NBs (Fig 7C and 7D-right). Furthermore, the propagation patterns revealed the emergence of a clear activity initiation focus, similar to the case of the cluster chains discussed above (Fig 5E and 5F). Under disinhibition, 42% of NBs (Ntotal = 366) were initiated in the lower left cluster (marked by a white dot in Fig 7C), in contrast to only 0.4% (Ntotal = 2345) under control conditions. The propagation variability of the network was quantified by calculating the standard deviation of ϴ for all NBs in the recording, followed by averaging over all clusters. Under control conditions the propagation variability was 1.367 radians, and after disinhibition it decreased to 0.775 radians. The mean propagation variability for different networks is presented in Fig 7E-left. In all analyzed networks (six out of six) a similar reduction in angle distribution was observed. Reduced variability was also observed in the number of participating clusters during NBs. While in control conditions, the NB sizes varied considerably from activation of single clusters to activation of the whole network (Fig 7F-right); after disinhibition, most of the NBs were synchronized over the entire network (Fig 7G-right). The variability in cluster activation was quantified by calculating the entropy of cluster activation patterns, as previously described for cluster chains. All analyzed networks (six out of six) showed a decrease in activation entropy following disinhibition (7E-right).
In Fig 7F and 7G, we explored the number of participating clusters in NBs with different intensities. We plotted the number of active clusters as a function of NB intensity for control (Fig 7F-left) and disinhibition (Fig 7G-left) conditions. In these plots, the intensity of each NB was normalized to the number of active clusters during this NB. Both in control and disinhibition conditions, the total normalized network intensity increased with the number of participating clusters. Thus, the extent of global activation (the number of clusters recruited during the NB) affected the local degree of activation (NB intensity in single clusters). However, under disinhibition most NBs recruited the entire network, while in control conditions, different NBs recruited a different number of clusters (Fig 7F and 7G-left). This is further illustrated by plotting the distribution of the number of recruited clusters over all NBs in control (Fig 7F-right) and disinhibition (Fig 7G-right) conditions.
The results described above demonstrate that modular topology support gating by disinhibition. To better understand this effect, we developed a computational model based on two coupled clusters in which different network parameters could be systematically modified. As cortical cultured networks exhibit complex organization of dynamical patterns, we adopted a previously published model that reproduced the main features of these patterns [32,40]. Neurons (N = 50 in each cluster) were modeled as Morris-Lecar elements with modified Tsodyks-Markram synapses and synaptic noise (see Materials and Methods). One out of every five neurons is an inhibitory neuron. For isolated clusters, the connectivity probability within clusters (intra-connectivity) was 0.25 and 0.2 for clusters 1 and 2 respectively, and the connectivity between clusters (inter-cluster connectivity) was initially set to zero. Intra-cluster connections were then replaced with inter-cluster connections with a probability λ which is defined as the modularity of the system. At the limit of λ = 0.5 the two clusters converge to a large uniform network. The model parameters (see Materials and Methods) were chosen to fit experimental data.
When the inter-connectivity was set to zero (isolated clusters), expectedly, each cluster exhibited short epochs of network bursts that were separated by sporadic single neuron activation, similar to activity patterns of isolated clusters in culture [29]. For slightly larger inter-cluster connectivity (λ = 0.02), some NBs successfully recruited the connected cluster, while others failed to elicit an NB in the connected cluster (Fig 8A). To quantify this property, we calculated the transfer probability for different modularity values. To do so, we detected NBs in the two clusters. An NB was considered as transmitted if following the activation of cluster 1, an NB peak was detected in cluster 2 within a time frame of 200 ms (see Materials and Methods). Since both clusters were spontaneously active, the transfer probability was non-zero even if the clusters were disconnected (λ = 0). To compensate for this, the number of “transferred” NBs at λ = 0 was subtracted from the measured number of transferred NBs and the total number of fired NBs before calculating the transfer probability.
We further examined how the transfer probability depends on the strength of inhibitory synapses [41]. For low inhibition levels, the transfer probability (Fig 8B) was highly dependent on the inter-cluster connectivity and changed between 0 and 1 (full transmission), indicating that modularity directly controls transmission probability. However, this modulation occurred over a narrow λ range and was insensitive to disinhibition (eliminating inhibitory synapses analogous to globally applying Bicuculline to the in vitro network) (Fig 8B). Increasing the decay time constant for inhibitory synapses (τd), or the inhibitory synaptic strength (A) twofold, increased both the transition range and the sensitivity to inhibition block (Fig 8C). This suggests that various mechanisms that increase inhibition within the time scale of an NB, such as selective increase in synaptic strength, or neuromodulation of synaptic decay dynamics, can support gating by disinhibition.
We next tried to identify other putative mechanisms which can increase disinhibitory effects in modular networks by focusing on the properties of the bridge connecting the two clusters [4]. We set the inter-cluster connection probability between pre-synaptic excitatory and post-synaptic inhibitory neurons to be as high as the excitatory to excitatory probability. This dramatically increased the range over which transmission was modulated (Fig 8D). Furthermore, following disinhibition, the transfer probability dramatically increased (Fig 8D—red curve), opening the gate between the two clusters. Thus, under these settings, our model captured the conditional propagation between sub-populations in a modular network and the gating of this conditional propagation by disinhibition.
An alternative to increasing inhibition strength is targeting inhibition, for example, by selectively directing inhibition to excitatory neurons. The existence of such targeting was verified both in the cortex and other brain regions [42,43]. Specifically, we examined whether it is possible to control transmission by selectively directing the output of inhibitory neurons, which receive inter-cluster input, to excitatory neurons (see detailed connectivity schemes in Materials and Methods). Indeed, doing so increased the control over transmission to almost the same extent as when increasing the feed-forward inhibition (Fig 8E). Interestingly, the two mechanisms described above were largely insensitive to the elimination of inhibitory connection through the bridge (Fig 8D and 8E–gray line), congruous with the notion of local inhibition.
Similar to the in vitro modular networks described above, the propagation of activity between connected sub-populations in the model was characterized by long delays (calculated as the time lag between the locations of NB peaks in both clusters, see Materials and Methods). Delays of several tens of milliseconds decreased with the connectivity between clusters (Fig 8F—blue curve). These delays were independent of the strength of inhibitory drive or whether it was removed altogether, as in the case under inhibitory block, suggesting that they are a consequence of the modular organization (see Discussion).
Finally, we examined whether information about the firing rate is transmitted through the bridge between the two sub-populations, as in our in vitro experiments. We analyzed the correlations between the total firing rates in the sending cluster as a function of the firing rate in the receiving cluster (similar to Fig 3D). We found that the correlation increases as a function of modularity (Fig 8G) and begins to saturate for λ = 0.1.
In this study, we systematically examined, experimentally and theoretically, the effect of network modularity on activity transmission between neuronal assemblies. A clear hallmark of the modular networks we studied is their capacity to support long delays in the order of 100 milliseconds (corresponding to a 5 μm/ms velocity between connected clusters) (Figs 3E and 6C-middle). Since axonal propagation speeds in culture are fast (>200 μm/ms, [44]), the observed delays are likely to be associated with the time it takes the receiving cluster to generate a network response due to multiple synaptic delays (recruitment time) (S3 Fig; for a more elaborate discussion see [39]). These delays are shorter on average in large networks of connected clusters, presumably due to the increased number of pathways between any two clusters, but are still much longer than in uniform networks.
Why is the capacity to support long delays useful? Foremost, long delays give rise to time scale separation between the activities of different modules and are a means to dissociate the intra-module from the inter-module processing [21]. Interestingly, our networks (Fig 3E-right) showed high variability in the average delays between different networks. This variability may stem from network architecture variability, and implies that modularity has the potential to support variable delays. Indeed, our model shows that delays can be controlled by modifying the coupling between sub-populations (Fig 8F). In a previous report (performed under the same experimental conditions as here) we showed that increasing the coupling in modular networks resulted in shorter delays [45]. We also observed variability in delays within specific networks over time (Fig 3E-left), implying that delays may be dynamically regulated to control transmission, for example by short- term plasticity [46].
An additional property of modular circuits is their activation asymmetry. Asymmetry may be important for controlling transmission directionality. In cluster chains, asymmetry was manifested by propagation patterns which were more probable than others (Fig 4A–4C). It was previously reported that coupled networks of similar sizes exhibit inherent asymmetry, and that this asymmetry is associated with the structural asymmetry of the connecting bridge [32]. Accordingly, a small subset of neurons at the bridge controls the propagation between networks. Our results support these previous findings, but suggest that functional asymmetry is also affected by the manner by which the coupled network is embedded within a larger network. When a network of coupled clusters was connected to a larger network, activity mostly propagated from the cluster connecting the large network to its neighbor (Fig 4E). Furthermore, higher asymmetry in the embedding network resulted in higher asymmetry between coupled clusters (Fig 4E). Interestingly, inhibition played a major role in determining asymmetry. Blocking the inhibition drastically affected the propagation direction between connected clusters (Fig 5E vs. 5F), suggesting that transmission directionality can be modulated to a large extent by reducing inhibition. However, further experiments, in which the cluster composition (e.g. the number of excitatory vs. inhibitory neurons) is monitored, are required to understand the morphological basis of asymmetry and transmission.
Neural networks have to maintain a fine balance between segregated activation (where activity is restricted to a specific sub-population) and integrated activation (where activity spreads to connected sub-populations) [9,47]. In the brain, functional segregation is associated with the structural modularity of the circuit [10], and the balance between functional segregation and integration explains the high functional complexity in the network [19,22,48,49]. We have shown that a similar fundamental property exists in small modular circuits. Namely, networks support a wide variety of activations from activity epochs which are confined to one sub-population to large-scale activation of the entire network. We observed such features both in one-dimensional cluster chains (Fig 5A and 5E), and in two-dimensional clustered networks (Figs 6A and 6C-top and 7B and 7F), but to a much lesser extent in uniform networks (Fig 6A, C-bottom). We found that both the number and intensity of activated clusters (Figs 5E, 5G, 7E and 7F), as well as the propagation direction (Figs 5B, 5D, 7B and 7E), were highly variable between consecutive NBs. In addition, due to long delays, this spatial activation variability resulted in temporal variability of NB durations. Interestingly, both temporal and spatial variability were previously reported for small-scale circuits in cortical slices. Organotypic slices show avalanche-like patterns typified by a wide distribution (heavy tail) of event sizes and durations [50]. The hypothesis that this diversity reflects the transient activation of different cell assemblies is supported by our results.
We suggest that the activation diversity in our networks is the outcome of the conditional propagation between sparsely coupled clusters. In small modular networks, activation of one sub-population does not necessarily lead to the activation of the other (Fig 3B and 3C). Our model supports this idea and illustrates how conditional propagation can be controlled by changing the degree or architecture of coupling between sub-populations (Fig 8C–8E). We note that we focus here only on a specific dimension of activation diversity. Uniform networks exhibit rich dynamical behavior along many spatial and temporal degrees of freedom [38,39,51,52]. However, a fundamental property of their activity is network synchrony on a time scale of ~100 ms [39]. Each of our clusters exhibits similar activity profiles to uniform networks [29], but the weak connections between clusters allows to spatially and temporally decouple their activity on this time scale.
We further suggest that the fact that conditional propagation spontaneously emerges in modular networks (in contrast to uniform networks) is associated with the networks’ self-regulation. We previously showed that isolated clusters of different sizes, and different connectivity, sustain moderate activity levels [29]. This corresponds to well-documented reports of structural and functional self-regulation of excitability in neurons [41,53–55]. Thus, neurons increase or reduce their propensity to fire when activity levels are low or high respectively. However, since in modular networks the fraction of connections between sub-populations is lower than within sub-populations, moderate activity in a cluster may still be below the self-regulated activation limit in the connected cluster (assuming that all sub-populations employ similar self-regulation mechanisms), resulting in the threshold-like behavior we observed (Fig 3C and 3D). Thus, one possible mechanism for transient increase in transmission is by increasing firing rates in the sending cluster, which may be one of the components contributing to transmission under disinhibition. Conversely, the post-synaptic currents to neurons in the receiving cluster at the time of the burst are another factor determining transmission. In principle, such currents can be modulated by neuromodulators or by a third cluster impinging on the receiving cluster. The latter is expected to be more prominent as the number of connections in the network increase. Indeed, while in one-dimensional chains, activity slightly decayed as it propagated between connected pairs, contributing to propagation failure, in two-dimensional clustered networks, a considerable fraction of events still managed to propagate to a large fraction of the network (S5A Fig). This may suggest that transmission in these two-dimensional networks is enhanced by the fact that each cluster is connected to several clusters, thus increasing the probability that their simultaneous activation recruits the receiving cluster. Further studies of single clusters receiving convergent input from two or more controlled clusters are required to explore activity integration in such networks.
By design, our modular networks were spatially regular (Fig 1C). Such a design minimized the variability in connections between clusters and variability in cluster sizes (S1B Fig), since self-organization is constrained by the regular pattern. This allowed us to focus on effects of modularity while partially avoiding the influences of other topological properties. For example, it was shown that if clusters freely organize without spatial constraints, cultures drive themselves towards assortative topologies with a “rich-club” core [56]. Such an organization equips modular networks with additional functional features, such as higher resilience to network damage compared to uniform networks. We previously reported [45] that modular networks with strong connectivity between modules do not give rise to long delays and conditional propagation. In addition, uniform networks in cultures are rarely uniform [51,52], and the uniformity probably depends on the level of granularity at which connectivity is examined. For this reason, we focus here on the extreme case of high intra-cluster connectivity with weak inter-cluster connectivity. Only at this level do we see a marked transition in the network’s activity profiles.
We showed that disinhibiting networks allows us to control transmission through modular circuits (Figs 5 and 7). Disinhibiting the network replaced the conditional propagation with reliable transmission (Figs 5A–5E, 7F and 7G). Not only was the diversity in the number of activated clusters removed (Figs 5E–5H and 7E–7G), but also the diversity in propagation directions (Figs 5E and 5F, 7D and 7E). These results were reproduced by our model. Blocking inhibitory synapses increased transmission, effectively “opening the gate” between connected modules (Fig 8D and 8E). Interestingly, this effect was weaker when the inter-cluster connectivity scheme was similar to the intra-cluster scheme (Fig 8C), suggesting that either stronger feed-forward (Fig 8D) or targeted (Fig 8E) inhibition may be instrumental in controlling propagation.
In addition to increased transmission, blocking inhibition in modular networks led to the emergence of an activation focus, which was absent before disinhibition (Figs 5F and 7B). The global disinhibition we induced in our examinations is used to illustrate the capacity of disinhibition to gate the system between different propagation states. In vivo, the inhibition-excitation ratio can be controlled locally, for example by neuro-modulators [57]. Further studies in which the degree of inhibition is manipulated in selective clusters (for example using Channelrhodopsin and Halorhodopsin) will determine the degree to which such gating can be controlled. Our global disinhibition was used to establish a proof of principle and is more akin to pathological conditions, as in the case of epilepsy where inhibition control is suspected to fail in large neuronal populations [58]. Indeed, under such conditions, the emergence of a focus (Figs 5F and 7B), and the repeated activation waves (Fig 5B), are clear hallmarks [59]. Interestingly, in addition to inhibition deficiency [60], lack of sparse functional connectivity between brain modules was also associated with epilepsy [61].
We note that in previous theoretical studies, gating was investigated in the context of balanced networks showing asynchronous irregular patterns [3,5]. In contrast, our study targets a different regime of population activity patterns. Primarily, our neurons are not constantly driven as the aforementioned model neurons. In such networks, excitatory-inhibitory balance does not result in persistent irregular patterns, but in synchronized bursting behavior. Nevertheless, excitatory-inhibitory balance does exist in these networks and is vital to the networks’ functionality [41]. Consequently, our model system is relevant for investigating gating during increased activity transients as occurring during synchronized and/or bursting activity. Such transient increase in excitability may have a fundamental role in transferring information between different cell populations [17,62,63].
To conclude, it is widely accepted that structure and function are closely related in neuronal circuits. However, the contribution of circuit topology to circuit function often remains hidden due to the difficulty in isolating small circuits in intact tissue. By engineering modular circuits in vitro, we explored the functional consequences of modularity and demonstrated that modular topology and disinhibition are instrumental in gating activity, directly demonstrating how structure can shape function in small neuronal circuits.
The entire neo-cortex of (E18-19) Sprague Dawley rat embryos of either sex were removed, chemically digested and mechanically dissociated by trituration, as detailed in a previous publication [29]. Dissociated cells were suspended in a growth medium and plated onto patterned substrates at a density of 700 cells/mm2. To promote the long-term cell survivability, a “feeder” colony of cells was added to the culture chamber [64]. The surrounding feeder culture did not directly contact the patterned culture. The mitotic inhibitor, FuDr (80μM FuDr, Sigma, Cat. No. F0503 and 240μM Uridine, Sigma, Cat. No. U3303) was added after four days in culture. Cultures were maintained at 37°C with 5% CO2 and 95% humidity. The growth medium was partially replaced every three to four days. The procedure was done in accordance with the NIH standards for care, and use of laboratory animals and was approved by the Tel Aviv University Animal Care and Use Committee. Overall 69 cluster chains (from 26 cultures), 15 large clustered networks, and eight uniform networks were tested in this study.
Extra-cellular recordings were conducted using a low noise pre-amplifier board (MEA1060-BC amplifier, gain ×1,100 with a band-pass filter of 10 Hz to 3 kHz, by Multi Channel Systems, MCS, Reutlingen, Germany). Signals were sampled at 10 kHz and stored on a personal computer equipped with a 60 channel, 12-bits data acquisition board (MC_Card, MCS GmbH), and an MC_Rack data acquisition software (MCS GmbH). An additional 200 Hz high pass filter (2nd order Butterworth) was applied to the data stream by the software. Recordings were performed 12 to 28 days in vitro.
The patterning method was adapted from a previous publication with slight modifications [27]. Briefly, PDL (Sigma, Cat. No. p7889) islands on top of commercial MEAs (MCS GmbH) were prepared with a soft lithography process using polydimethylsiloxane (PDMS) stencils. An SU8-2075 (MicroChem Corp) mold with approximately 150 μm thickness was casted onto a patterned silicon wafer. The pattern consisted of a rectangular grid (6 x 10) of circles with diameters ranging between 80 and 200 μm with 500 μm spacing. The PDMS stencil was prepared by spin coating the wafer with PDMS. After detaching the PDMS substrate from the mold, the stencil was placed on commercial MEAs and aligned with the electrode locations. The PDL solution was applied to the PDMS stencil and the PDL was dried on a hot plate at 37°C for half an hour. The PDMS stencil was removed before cell plating. The probability of inter-cluster connections depended on island diameter: Larger islands resulted in networks with a higher degree of connectivity. The network's self-organization lasted up to ten days in culture, after which the patterns became stable.
To quantify network level activity, we calculated the activity intensity (AI) of each cluster:
AI={A,A≥00,A<0,A=∑i=1M|V(i)|M−NT
where V is the voltage waveform, M is the number of samples in each activity intensity bin, and NT is the activity intensity noise threshold. The noise threshold is added to remove the contribution of noise to the AI value and was calculated as follows: The unbiased kurtosis (measuring Gaussianity) of the voltage trace is calculated in time bins of 20 ms. The kurtosis of a univariate Gaussian distribution is 3. Active bins were characterized by super-Gaussian distributions; therefore bins with kurtosis values higher than 3.1 were rejected. The waveforms of the rest of the bins were used to estimate the average absolute value of the noise voltage, which is the noise activity intensity threshold, NT. Once NT is obtained, AI is calculated according to the equation for AI (above). We have previously shown that the activity intensity measure can serve as a good estimate for changes in firing rate of superimposed spikes [29].
To detect NBs in single channels we used a previously described method [29]. Briefly, we first calculated AI in bins of 2 ms. Next, we counted the number of active AI bins (having non-zero values) in moving windows of length W = 100 ms (steps of 10 ms). W = 100 was chosen because it is long enough to achieve a smooth NBs profile (but not longer than a typical NB). In the model W = 10 was long enough, since the large number of sampled neurons resulted in a smoother profile to begin with. Single sporadic spikes, as well as short threshold crossings, may contaminate NB profiles. As the rate of these events is well below 10Hz, we eliminated them by zeroing data points below a threshold value of T (T = 10). For the model, this value was chosen as T = 5 since the activity is not contaminated by noise. To ensure that the activity near NB edges was included, a second convolution with the same kernel, followed by thresholding with a value of 1, was performed. In the resulting time series, NBs are represented by a series of consecutive positive values. Finally, to ensure that short transient decreases did not result in a separation of NBs to two events, NBs occurring less than G milliseconds apart (end of previous to beginning of next) were merged into one NB (G was set to 100 ms for single clusters in the experimental data and 50 ms for the model in which response variability was lower). To increase the accuracy of the NB start time, end time and peak time detection, the AI function was extracted during the time windows of the previously calculated NB occurrences. The beginning and end of NBs were taken as the first non-zero value from left and right respectively. The peak of the NB is determined by smoothing the AI (using a convolution with a Gaussian, σ = 50 ms), and extracting the time of the maxima. For identifying NBs globally in the whole network, instead of in single channels, events were counted in all channels instead of only in one, and G was set to 1 s (in accordance with the large delays in clustered networks). In addition, channels with very weak activity during each NB (active for less than 10 ms) were not considered. To extract NBs in the computational model, the spike timings of all neurons were counted instead of activity events. The results of NB detection and parameter extraction were verified by manual inspection for all clusters.
Detection of correlations between bursts was performed using a hierarchical clustering algorithm [38]. Briefly, AI traces were extracted during a 1000 ms window surrounding the detected NB peaks and smoothed by convoluting with a Gaussian (σ = 10 ms). The time invariant correlation between the ith and jth NBs were calculated as follows:
Rij=maxt{⟨Cijn(t)⟩n}
, where
Cijn
is the normalized cross-covariance between the AI trace during the ith and jth NBs of the nth cluster, and t is the time index of the cross-covariance function. To identify groups of similar bursts, Rij is reordered using the dendrogram hierarchical clustering algorithm. The dendrogram is calculated on the Euclidian distance matrix, Dij, between the ith and jth rows in R: Dij2 = ∑k(Rik − Rjk)2.
Simulated networks consisted of two clusters, each with 50 Morris-Lecar neurons (see neuron model for details). Neurons were connected through modified Tsodyks-Markram synapses (see synapse model for details). Connectivity was defined by the matrix aij, where aij = 1/0 corresponds to an existing/non-existing connection between the pre-synaptic terminal of neuron i and the post-synaptic terminal of neuron j. One of every five neurons was randomly selected as inhibitory. Each network was simulated for 300 s using an Euler integrator with a 0.1 ms time step.
The modularity of the network was determined by the parameter λ as follows. The connectivity probability within cluster 1 (neurons 1 to 50) and cluster 2 (neurons 51 to 100) was initially set to 0.25 and 0.2 respectively, and the connectivity between clusters was set to 0. Next, we randomly replaced intra-cluster connections with inter-cluster connections with probability λ. Thus, for λ = 0 the network is composed of two isolated (disconnected) clusters, and for λ = 0.5 the intra-cluster connectivity is equal to the inter-cluster connectivity, and the initial separation to two clusters can no longer be observed.
We simulated three connectivity schemes, which differ by the number and distribution of inhibitory (I) and excitatory (E) synapses: proportional inhibition, strong feed-forward inhibition, and direct targeting inhibition. We kept the E/I neuron ratio constant, although in the experimental conditions some variability may occur. Such a choice is consistent with the self-regulation of synaptic transmission which compensates changes in the network structure to maintain E/I balance [41,65].
In general, the connectivity between two neurons can be one of the following: E→E, E→I, I→E and I→I. These schemes were differentiated by ⟨Nv→w,x→y⟩ which denotes the expected value of the number of synapses from cluster v to w, where x and y stand for the type of the pre and post-synaptic neuron type (E or I). For example, N2→1,E→I is the number of E → I synapses from cluster 2 to cluster 1.
For the above connectivity schemes, two conditions were used: inhibition block and local inhibition. For inhibition block (analogues to application of Bicuculline), all inhibitory synapses were disabled by setting A = 0 (see synapse model). For local inhibition, only inhibitory inter-cluster connections were disabled (A = 0).
Neurons were modeled as Morris Lecar elements [66]:
CmV˙=Iext−gCaMSS(V−VCa)−gKW(V−VK)−gL(V−VL)
W˙=ϕ(WSS−W)cosh(V−V32V4)
MSS(V)=0.5(1+tanhV−V1V2)
WSS(V)=0.5(1+tanhV−V3V4)
where V is the membrane potential, Iext is the externally applied current, W and M are the fraction of open K+ and Ca+2 channels respectively, and Cm,ϕ,V1,V2,V3,V4,gk,gCa,gL are constants, adopted with slight modifications from Rinzel and Ermentrout [66] (see Table 1 for a full list). These parameters were selected to simulate a class I neuron which can generate cellular-level [66] and network-level [40], bursting in accordance with the activity of isolated clusters [29]. We also examined networks of leaky integrate and fire neurons. Although we could qualitatively reproduce our results with these neurons, Morris-Lecar neurons gave a much better fit to the experimental observations (for discussion see [40]). Neurons received both synaptic and noise input: Iext = In + Isyn
The noise, In, fed into every neuron, was selected from a Gaussian distribution (mean,
μ=7.55[μAcm2]
and standard deviation,
σ=4[μAcm2]
) (independently identically distributed for each neuron) for every simulation step. This choice was driven by the assumption that noise originated from spontaneous synaptic release of neurotransmitter [67]. Such noise could be modeled as an Ornstein–Uhlenbeck process [68]. Considering that for single synapses, the time between spontaneous releases is a Poisson process [69] and that the number of synapses onto a neuron is large, the overall spontaneously evoked noise current can be approximated by a single Gaussian variable (in accordance with the central limit theorem).
Synapses were modeled as modified Tsodyks-Markram elements [70]:
x˙=z(−tan(1.2z−1.2))τrec−uxδ(t−tAP)
y˙=−yτd+uxδ(t−tAP)
z˙=yτd−z(−tan(1.2z−1.2))τrec
where x, y and z are the fractions of synaptic resources in the recovered, active, and inactive states of the synapse. τrec and τd are time constants representing the recovery and decay of active resources respectively (see Table 2 for a full list). tAP is the arrival time of the last action potential to the pre-synaptic terminal. u is the fraction of resources activated upon action potential arrival. In excitatory synapses, u was constant (u = U0). In inhibitory synapses, u was a dynamic variable enabling synaptic facilitation during bursts:
u˙=−uτfacil+U0(1−u)δ(t−tAP)
We modified the x and z terms in the original Tsodyks-Markram model by adding a tangent function to prevent tonic endless spiking. Such tonic spiking was observed when the network fired in high rates and originated from the linearity between the recovery rate and the amount of inactive resources. Our modification is in accordance with findings indicating that synapses do not increase their recovery rate following depletion, but rather decrease it after a certain level of depletion [71–73].
The synaptic input to a neuron, Isyn, was calculated by summing over synaptic currents from all connected neurons:
Isyn(t)=∑iAiyi(t)
where Ai is the synaptic strength. To reflect the non-uniformity of synaptic strengths, they were selected from a Gaussian distribution (μ = Anom, σ = Anom/2), where Anom is the nominal value of the synapse [70,74] (see Table 2 for a full list). To limit the distribution of strengths, only values between 0.8Anom ≤ A ≤ 1.2Anom were considered (values were redrawn from the distribution until a value within limits was drawn).
|
10.1371/journal.pgen.1006117 | Chromosomal Translocations in the Parasite Leishmania by a MRE11/RAD50-Independent Microhomology-Mediated End Joining Mechanism | The parasite Leishmania often relies on gene rearrangements to survive stressful environments. However, safeguarding a minimum level of genome integrity is important for cell survival. We hypothesized that maintenance of genomic integrity in Leishmania would imply a leading role of the MRE11 and RAD50 proteins considering their role in DNA repair, chromosomal organization and protection of chromosomes ends in other organisms. Attempts to generate RAD50 null mutants in a wild-type background failed and we provide evidence that this gene is essential. Remarkably, inactivation of RAD50 was possible in a MRE11 null mutant that we had previously generated, providing good evidence that RAD50 may be dispensable in the absence of MRE11. Inactivation of the MRE11 and RAD50 genes led to a decreased frequency of homologous recombination and analysis of the null mutants by whole genome sequencing revealed several chromosomal translocations. Sequencing of the junction between translocated chromosomes highlighted microhomology sequences at the level of breakpoint regions. Sequencing data also showed a decreased coverage at subtelomeric locations in many chromosomes in the MRE11-/-RAD50-/- parasites. This study demonstrates an MRE11-independent microhomology-mediated end-joining mechanism and a prominent role for MRE11 and RAD50 in the maintenance of genomic integrity. Moreover, we suggest the possible involvement of RAD50 in subtelomeric regions stability.
| The parasite Leishmania relies on gene rearrangements to survive stressful conditions. However, maintaining a minimum level of genomic integrity is crucial for cell survival. Studies in other organisms have provided evidence that the DNA repair proteins MRE11 and RAD50 are involved in chromosomes organization, protection of chromosomes ends and therefore in the maintenance of genomic integrity. In this manuscript, we present the conditional inactivation of the Leishmania infantum RAD50 gene that was only possible in MRE11 deficient cells and suggest the genetic background is crucial for RAD50 inactivation. We demonstrate the occurrence of chromosomal translocations in the MRE11 and RAD50 deficient cells and described a MRE11-independent microhomology-mediated end-joining mechanism at the level of translocation breakpoints. We also suggest a possible involvement of RAD50 in subtelomeric regions stability. Our results highlight that both MRE11 and RAD50 are important for the maintenance of genomic integrity in Leishmania.
| Genomic integrity maintenance is essential for cellular development and viability [1–3]. Failure to repair DNA will lead to genomic instability (reviewed in [4–7]). DNA structural changes can manifest as inversion, deletion, duplication, translocation, chromosome end-to-end fusion, aneuploidy [8–10] and some of these events such as gene amplification have been associated in Leishmania with response to drug and oxidative stress [11–14]. Increased numbers of DNA rearrangements have been reported in many inherited cancer susceptibility human syndromes [9]. Specific DNA repair genes are mutated in these genomic disorders such as ATM in the Ataxia telangiectasia syndrome, MRE11 in the Ataxia telangiectasia-like disorder, NBS1 in the Nijmegen breakage syndrome and BLM in the Bloom’s syndrome [15–18]. It has been suggested that errors occurring during DNA replication such as stalled or broken replication forks can lead, if left unrepaired, to DNA double strand breaks (DSBs) that are precursors of DNA rearrangements [10,19]. DSBs can also occur during replication or result from exposure to DNA-damaging agents such as ionizing radiation or chemotherapeutic drugs [1,20]. The two main strategies to cope with DSBs are non-homologous end joining (NHEJ) and homologous recombination (HR) [20]. However, only a few NHEJ factors are present in Leishmania (MRE11, Ku70/Ku80 and APTX) while Artemis, XRCC4 and the DNA ligase IV are absent, suggesting that this pathway is not functional in the parasite [21–23]. Another pathway that is normally suppressed when NHEJ is present is called microhomology-mediated end joining (MMEJ) or alternative end joining and has been reported in the related parasite Trypanosoma brucei [22–24]. In MMEJ, small regions of homology (2 to 20 nucleotides) are used for ligation after resection of each DNA ends in a Ku-independent manner [20]. In this process, DNA ends created from DSBs are recognized by PARP-1 and resected by the MRN (MRE11-RAD50-NBS1) complex followed by annealing and ligation of the two ends by XRCC1/DNA ligase III [25,26]. The HR pathway has been shown to be important for the recovery of stalled replication forks, genomic integrity and telomere maintenance [27–29]. In HR, DSBs are first recognized by the MRN complex and resected by EXO1 and MRE11. Therefore, MMEJ and HR share the same initial step of resection which involves the MRN complex in order to produce regions of homology. Nevertheless, the length of DNA resection as well as the length of the homologous sequence differ between the two processes [23,30]. The tripartite MRN complex has been shown to act as DSBs sensor and DSBs repair effector and is also associated with telomere maintenance [31–33], displaying a major role in the maintenance of genomic stability [34–37]. The complex is composed of MRE11 and RAD50, highly conserved between species, and NBS1 (also represented by XRS2 in yeast) is less conserved and only present in eukaryotes. We previously demonstrated that LiMRE11 displays the same DNA binding and exonuclease activity as human MRE11, but the protein is not essential in Leishmania infantum [38]. In addition, we showed the importance of MRE11 and its nuclease domain in extrachromosomal linear amplicons formation under drug pressure. The RAD50 protein is a DNA binding ATPase that displays sequence and structural homology to structural maintenance of chromosome (SMC) family members. An anti-parallel coiled-coil domain contains a central zinc hook (CXXC) motif and might contribute in holding together separate DNA ends [8,34]. The third member of the complex is NBS1 and possess a MRE11 binding domain. NBS1 is thought to stimulate the MRE11-RAD50 complex DNA binding and nuclease activities but biochemical activities of the NBS1 protein itself are not yet elucidated [8,34]. Disruption of the MRE11 and RAD50 genes have been shown to increase gene rearrangements rate up to 1000 fold [2,39]. Null mutations in any of the MRN proteins lead to embryonic lethality in mice and have been associated in yeast with DNA rearrangements and chromosome loss events as well as defect in both HR and NHEJ [39–44]. In this manuscript, we present the conditional inactivation of L. infantum RAD50 orthologue, a gene essential in the MRE11 proficient wild-type background but apparently dispensable in the MRE11-/- background. We also demonstrate chromosomal translocations in the MRE11 and RAD50 deficient cells. These translocations happened through a MRE11-independent MMEJ mechanism where sequence microhomology were found at the translocations breakpoints.
It is standard practice to generate null mutant parasites by replacing the entire ORF with resistance markers. The genes coding for the blasticidin-S deaminase (BLAST) and puromycin acetyltransferase (PURO) were cloned between the 5’- and 3’- L. infantum RAD50 flanking regions and the BLAST or PURO constructs were transfected independently by electroporation. Hybridization with a 5’UTR probe should lead to 3.2, 1.7 and 1.5 kb SacI-SacI bands in the wild-type (WT), BLAST RAD50-/+, PURO RAD50-/+ cells respectively (Fig 1A). We generated BLAST RAD50-/+ and PURO RAD50-/+ heterozygous lines (S1B Fig, lanes 2–3), but surprisingly, in the BLAST/PURO/WT RAD50-/-/+ line we observed a remaining intact RAD50 allele (S1B Fig, lane 4). Despite many attempts, the generation of a RAD50 null mutant failed. This generation of polyploidy at specific locus is frequently observed in Leishmania [45–47] and is thought to occur at locus reputed to be essential. To provide further support for the essentiality of RAD50, we first introduced a RAD50 rescue plasmid (Psp-NEO-RAD50WT, Fig 1A) in the BLAST RAD50-/+ cells. Upon the transfection of the PURO cassette we could generate a chromosomal BLAST/PURO RAD50-/- cell with no more intact RAD50 chromosomal copy (Fig 1B, lane 4) but with the presence of the extrachromosomal rescue RAD50 copy with its diagnostic 4.9 kb SacI-SacI band (Fig 1A) hybridizing with the RAD50 ORF probe (Fig 1C, lane 4). Removing the drug NEO pressure (the marker of the rescuing episome) for several passages would lead to either maintenance or loss of the plasmid depending on whether RAD50 is essential or not. Cells grown in absence of selection for the NEO marker maintained the episome (up to 55 passages) (Fig 1C, lane 5). This was not due to an unusual stability of the plasmid since introduction of the same NEO plasmid in WT cells and then growth in absence of selection pressure led to the loss of the rescuing episome after 35 passages (Fig 1C, lanes 2, 3).
To further investigate the essentiality of the RAD50 gene, we generated a mutated version of the Psp-NEO-RAD50 after introduction of the K42A mutation in the RAD50 ATPase domain (Psp-NEO-RAD50K42A) (S2A Fig). The wild-type and mutated recombinant proteins were purified and the K42A mutation indeed impaired the ATPase activity of RAD50 (S2D Fig). The Psp-NEO-RAD50K42A construct was transfected in the BLAST RAD50-/+ cells. In contrast to cells complemented with Psp-NEO-RAD50WT (Fig 1B, lane 4, S2B Fig, lane 4), cells complemented with Psp-NEO-RAD50K42A showed a remaining RAD50 allele after integration of the PURO marker (S2B Fig, lane 5). The presence of the Psp-NEO-RAD50K42A plasmid was confirmed by hybridization with a RAD50 probe (S2C Fig). This result indicates that the ATPase activity of the rescue RAD50 copy is necessary to allow inactivation of both RAD50 chromosomal alleles.
While inactivation of RAD50 was not possible in a WT background, it was easily achieved in a MRE11-/- mutant. Indeed an MRE11-/- mutant was already available [38] and its RAD50 locus was shown here to be intact, as a 3.2 kb fragment was present after SacI digestion of the genomic DNA (Fig 1B and 1C, lane 6). In the MRE11-/- background we could inactivate both RAD50 alleles with the BLAST and PURO markers without the need for a rescuing plasmid (Fig 1B and 1C, lane 7). To confirm the absence of both MRE11 and RAD50 genes in the MRE11-/-RAD50-/- strain, we performed PCR amplification using two sets of primers (Fig 1D). The use of primers sets aa’ and bb’ should only amplify a PCR fragment if the MRE11 and RAD50 genes are present respectively. As expected, no PCR amplification was detected for both MRE11 and RAD50 genes in the MRE11-/-RAD50-/- strain (Fig 1D, lane 7). We also carried out qRT-PCR for RAD50 mRNA levels in a number of lines and inactivation of one RAD50 allele by either PURO or BLAST reduced the mRNA by half compared to WT (S1C Fig, lanes 2,3). A similar fold decrease was observed in the PURO/BLAST RAD50-/- cells that have an extra copy of the gene (S1B Fig, lane 4), indicating that this new allele is actively expressed (S1C Fig, lane 4). The level of RAD50 mRNA in the MRE11-/- cells was similar to the WT strain but was undetectable in the MRE11-/-RAD50-/- cells (S1C Fig, lanes 5,6). Overall, a nice correlation between RAD50 copy number and mRNA expression was observed.
We also attempted inactivating the RAD50 gene in cells with only one MRE11 allele but mutated for its nuclease activity (HYG/PUR-MRE11H210Y [38]). We failed to generate a RAD50 null mutant in this HYG/PUR-MRE11H210Y cells since we observed the maintenance of a third RAD50 chromosomal allele after integration of both BLAST and NEO resistant markers in the RAD50 locus (S3B Fig, lane 3). RAD50 thus appears to be essential in MRE11H210Y nuclease dead cells.
The availability of a RAD50 null mutant (in the MRE11-/- background) has allowed to test for a number of phenotypes. The MRE11-/- and MRE11-/-RAD50-/- mutants had similar growth properties (S4A Fig), susceptibility to the DSBs inducing alkylating damaging agent methyl methanesulphonate (MMS) (S4B Fig); and displayed a reduced ability to carry out homologous recombination (HR) (S4E Fig). The RAD50-/- mutant with its episomal rescue had similar growth phenotype and recombination proficiency as the WT cells (S4 Fig). To ensure that the phenotypes observed in the MRE11-/- null mutant were not due to a reduction in RAD50 protein levels, we overexpressed RAD50 as part of an episomal construct (Psp-RAD50) in the MRE11-/- cells. The MRE11-/- and MRE11-/- Psp-RAD50 cells had similar growth properties and susceptibility to MMS (S4C and S4D Fig).
In response to drug pressure, Leishmania amplifies specific portion of its genome either as part of extrachromosomal circular or linear amplicons. Circles are dependent on RAD51 and RAD51-4 [14,48] while linear amplicons depend on MRE11 [38]. Selection of Leishmania WT cells for resistance to the antifolate methotrexate (MTX) often leads to the extrachromosomal amplification of the pteridine reductase gene PTR1 (usually as part of linear amplicons) or of the dihydrofolate reductase-thymidylate synthase gene DHFR-TS (usually as part of circular amplicons) [49]. An example of a PTR1 containing linear amplicon (at 450 kb) is provided in a WT cell that was selected fro MTX resistance (Fig 2A). The 770 kb PTR1 hybridizing band corresponds to the chromosomal alleles. We showed previously that in contrast to WT cells, the MRE11-/- mutant selected for MTX resistance did not have PTR1 amplified as part of linear amplicons (Fig 2B) [38]. We investigated the ability of the MRE11-/-RAD50-/- null mutants to perform extrachromosomal amplification by selecting clones for MTX resistance in a stepwise manner (up to 1600 nM, a 16-fold increase in resistance compared to parent cells). Leishmania chromosomes extracted from ten MTX resistant clones derived from MRE11-/-RAD50-/- parasites were separated by pulse field gel electrophoresis (PFGE) (S5 Fig) and hybridized with the PTR1 gene. Hybridization data revealed the 770 kb PTR1 containing chromosome but no hybridizing bands diagnostic for PTR1 linear amplicons (Fig 2C and S5 Fig). However, clones 1, 4, 7 and 8 had a PTR1 circular amplification, as deduced from the characteristic hybridization profiles of circles in PFGEs, including the hybridization in the slots (corresponding to open circles) and the hybridizing smears (corresponding to topoisomers of the circles) [50] (Fig 2C lanes 1, 4, 7, 8). Amplification of the DHFR-TS gene is rarely observed in L. infantum selected for MTX resistance and this was further confirmed, where only the 520 kb chromosomal copies hybridized to a DHFR probe (Fig 2D). However since several resistant mutants had no PTR1 amplification (Fig 2C) we hybridized the same resistant clones with a DHFR-TS probe. We observed the 520 kb band corresponding to the DHFR-TS containing chromosome but no sign for either circular or linear amplicons (Fig 2F). However, we detected in all clones a 795 kb band that surprisingly was also present in the parent MRE11-/-RAD50-/- cells before MTX exposure (Fig 2F, lane 0). A similar, but clearly not identical (950 kb vs 795 kb) chromosomal rearrangement was previously observed in the MRE11-/- strain after MTX pressure (Fig 2E, lane +) [38], suggesting that this locus is prone to chromosomal rearrangement. The chromosomal rearrangement involving the DHFR-TS chromosome in the MRE11-/-RAD50-/- strain was further studied by genome sequencing of the nuclease deficient strains.
The genomes of the WT, MRE11-/- and MRE11-/-RAD50-/- lines were subjected to Illumina next-generation paired-ends sequencing (NGS). Sequencing reads were first aligned to the genome of L. infantum JPCM5 using bwa-mem alignment [51]. The alignments were then screened for discordant read pairs and split reads alignments using the Lumpy-sv and the Delly software [52,53]. This provided a list of chromosomal translocations present in the MRE11 and MRE11/RAD50 null mutants. A total of five translocations were observed in the MRE11 and MRE11/RAD50 deficient cells (Table 1). The analysis of the genomic sequences allowed the detection of the translocation of part of the DHFR-TS chromosome observed in Fig 2F. It involved 433 kb of chromosome 12 and 362 kb of chromosome 06 (encoding DHFR-TS), giving a hybrid chromosome T 12–06 of 795 kb (Fig 3A). To further characterize experimentally this translocation, we hybridized the chromosomes of the WT, MRE11-/- and MRE11-/-RAD50-/- cells with probes spanning the translocation breakpoints (filled and open squares and circles in Fig 3A). Hybridization with the gene LinJ.12.0671 (■) revealed a band corresponding to chromosome 12 (568 kb) and an additional 795 kb band corresponding to the T 12–06 translocation in the MRE11-/-RAD50-/- strain (Fig 3B). The same 795 kb band hybridized to LinJ.06.0480 (●), a gene derived from chromosome 06 and part of the T 12–06 translocation (Fig 3B). The genes LinJ.12.0690 (□) and LinJ.06.0470 (○) should not be part of the hybrid chromosome (Fig 3A) and indeed when these genes were used as probes they only hybridized to bands corresponding to chromosomes 12 and 06 respectively (Fig 3B, Table 1). Two additional translocations also implicated chromosome 12 (T 12–17, T 12–18, Table 1) as described below. This is possible because several chromosomes of Leishmania are polyploids [13,54–56] and read counts indicate that chromosome 12 is tetraploid in our L. infantum WT strain (Fig 3C). One translocation led to a hybrid composed of 386 kb of chromosome 12 fused to 159 kb of chromosome 17 (Fig 4A). The size of chromosome 12 and the hybrid chromosome are too similar for their discrimination by PFGE but hybridization with LinJ.17.1180 (■) showed a band corresponding to chromosome 17 (667 kb) and the 545 kb band corresponding to T 12–17 (Fig 4B). The third translocation implicating chromosome 12 involved chromosome 18 (408 kb of chromosome 12 and 167 kb of chromosome 18 leading to an hybrid chromosome of 575 kb, (Fig 4C)). The size of chromosome 12 and the hybrid were again too similar for discrimination by PFGE but hybridization with LinJ.18.1520 (●) revealed the 575 kb band corresponding to T 12–18 (Fig 4D lanes 2, 3). The hybridization patterns in the two nuclease mutants are more complex with no band exactly migrating with the intact chromosome 18 band (Fig 4D). In the case of the MRE11-/-RAD50-/- mutant, this can be explained in part by an additional translocation of chromosome 18 with chromosome 20 (Fig 4E) where the new hybrid chromosome (T 18–20) had 674 kb of chromosome 18 and 69 kb of chromosome 20 for an estimated length of 743 kb (Fig 4D and 4F lane 3). In the MRE11-/- mutant we observed three bands hybridizing with the LinJ.18.1520 (●) probe (Fig 4D, lane 2). The band of 575 kb corresponds to the T 12–18 hybrid chromosome, the highest band at 778 kb corresponds to one of two versions of T 18–20 (with an internal duplication that was highlighted by reads depth analysis, see below). This band also hybridized with LinJ.20.1570 (▲) (Fig 4F, lane 2). The middle band hybridizing to LinJ.18.1520 (●) appears slightly smaller than the 720 kb WT chromosomal copy (Fig 4D, lanes 1 and 2) and may correspond to a truncated form of chromosome 18. The final translocation highlighted by NGS involved chromosome 08 and 17 (Table 1) and T 08–17 consists of 175 kb of chromosome 17 and 395 kb of chromosome 08 leading to an hybrid chromosome of 570 kb (Fig 5A). Hybridization with LinJ.08.0290 (■) revealed a 570 kb band corresponding to T 08–17 and a 465 kb band slightly smaller than the expected size of chromosome 08 (495 kb) (Fig 5B, lane 3). This smaller 465 kb band also hybridized with LinJ.08.0280 (□) (not part of T 08–17), and may, similarly to one copy of chromosome 18 discussed above, correspond to an internal deletion or truncation of the original chromosome but the bioinformatics analysis did not provide support for these potential scenarios. Hybridization with probes derived from chromosome 17 further supported the formation of the hybrid chromosome T08-17 (Fig 5B, lane 3).
The Lumpy-sv and Delly software also revealed a fusion between chromosome 27 and chromosome 02 that was already present in the WT cells, highlighting a difference between the L. infantum 263 WT strain compared to the reference L. infantum JPCM5 WT (S6 Fig). Most of chromosome 27 (1044 kb) is fused with the last two genes on chromosome 02 (4 kb) (S6A Fig). Sequence homology between the end of chromosomes 2 and 27 has already been described for L. major [57] with subtelomeric repeats and this rearrangement occurring in the WT may correspond to telomere exchange rather than translocation.
In the past, we have used normalized reads depth coverage over the 36 chromosomes to predict copy number variations [11,12]. Sequenced reads of the 36 chromosomes of the L. infantum 263 strain indicated that while the majority of chromosomes were mostly diploid, chromosomes 12, 13 and 31 were polyploid. There were no changes in ploidy in the nuclease mutants except when translocation occurred. Normalized log2-transformed read counts for chromosome 06 in WT cells, MRE11-/-, and MRE11-/-RAD50-/- revealed a shift at the T 12–06 breakpoint in the MRE11-/-RAD50-/- null mutant, leading to an increased number of reads for part of chromosome 6 starting with gene LinJ.06.0480 (Fig 3D). At one telomere end of chromosome 06 in WT and MRE11-/- strains we observed increased number of reads for a region of 60 kb (Fig 3D) that corresponds to a linear extrachromosomal amplicon that we have previously characterized in L. infantum 263 WT [14]. This linear amplicon was lost in the MRE11-/-RAD50-/- strain. Reads depth coverage indicated that chromosome 12 is tetraploid in L. infantum 263 WT and that MRE11-/- and MRE11-/-RAD50-/- parasites probably contained only one intact copy of chromosome 12 (Fig 3C) but all the hybrid chromosomes T 12–06, T 12–17 and T 12–18 contribute to a higher ploidy for most sequences of chromosome 12 (Fig 3C). The normalized reads depth of chromosome 12 also highlighted the T 12–06, T 12–17 and T 12–18 breakpoints (Fig 3C). Similarly, the breakpoints for T 12–18 and T 18–20 also fitted with a change in reads depth on chromosome 18 and chromosome 20 (S7 Fig). In the case of chromosome 18, reads depth showed a shift at the T 12–18 and T 18–20 breakpoints in the two mutants and overlapping regions between T 12–18 and T 18–20 (from LinJ.18.1330 to LinJ.18.1530) were present in three copies (S7A Fig). In the MRE11-/- strain, normalized read counts for chromosome 18 also highlighted internal duplication of 15 kb close to the T 18–20 breakpoint, increasing the size of the T 18–20 (S7A Fig). In the same translocation, part of chromosome 20 also showed a duplication of 20 kb (S7B Fig) in the MRE11-/- cells, increasing the size of the translocation T 18–20 from 743 kb to 778 kb in that strain (Fig 4E). Reads mapping to chromosome 20 also showed a decreased number of reads at one telomere end in the MRE11-/-RAD50-/- cells (S7B Fig), a phenomenon that we observed for several other chromosomes (see below). Finally, normalized reads depth coverage over chromosomes 08 and 17 highlighted the T 08–17, T 12–17 and T 08–17 breakpoints (Fig 5C and 5D).
We used PCR to validate the new junctions created by the fusion of portions of chromosomes in all translocations. Oligonucleotide primers located on each side of the translocation points were designed to target the new junctions. The ORFs identities (or position of intergenic regions) closest to the breakpoints can be found in Table 1. We were able to precisely define the junction of 4 of the 5 translocations (T 12–06, T 12–18, T 18–20 and T 08–17). In all cases the sequencing of the fusion points in the hybrid chromosomes revealed that the rearrangements occurred at the level of microhomologies between 7 and 17 bp (Fig 6A–6D). There were no common sequence features between the various repeats. We were unsuccessful to map precisely the translocation breakpoints for T 12–17. Indeed the breakpoint is located inside a region containing repeated DNA sequences along 60 kb and specific PCR amplification of the junction has not been possible. The genome of Leishmania is constituted of large polycistronic clusters of genes that are co-expressed [58,59]. Interestingly some of the translocation would create new regions where co-directional gene clusters diverge or converge (T 12–18, T 18–20, T 08–17) that may impact on gene expression.
PCR amplification of the junction between chromosomes 27 and 02 revealed an insertion of 21 bp at the junction of chromosome 27 and chromosome 02 in the 27–02 hybrid (S6B Fig). This rearrangement is clearly not similar to the translocation events characterized in this study and may correspond to exchange of telomeric sequences between chromosomes 27 and 02.
The MRE11-/-RAD50-/- parasites displayed a decreased number of reads mapping to chromosomes ends, suggesting sequences near telomeres were impaired in that strain. This phenomenon occurred for eleven chromosomes (Fig 7 and S8 Fig) and three were experimentally verified by Southern blot (Fig 7). Genomic DNAs from the WT, MRE11-/- and MRE11-/-RAD50-/- cells were hybridized with a chromosome 05 probe close to the telomeres (LinJ.05.0060) and hybridization intensities were compared with probe LinJ.05.0560 used as an internal control for DNA loading (Fig 7A). Hybridization intensities yielded a 0.7 fold-decrease for the MRE11-/-RAD50-/- strain compared to the WT or MRE11-/- cells. Similar analyses were performed with probes derived from chromosome 28 (Fig 7B) and chromosome 34 (Fig 7C) and when telomeric proximal probes were used the signal was consistently lower in the MRE11-/-RAD50-/- parasites, compared to either WT cells or the MRE11-/- mutant. Genomic DNAs from WT, MRE11-/- and MRE11-/-RAD50-/- cells were also digested with Sau3aI, AluI and RsaI and hybridized with a telomeric probe [60,61]. After hybridization, discrete bands were present in both the WT and MRE11-/- but in the MRE11-/-RAD50-/- cells we observed a smear (S9A Fig). When an internal probe far from telomere was used (PTR1 gene), a single band was observed in all three lines (S9B Fig). The results suggest that within the MRE11-/-RAD50-/- population there is considerable heterogeneity at the end of chromosomes in individual cells (explaining the smear when hybridized with a telomeric probe), in line with the decreased number of reads mapping chromosomes ends (Fig 7).
Gene rearrangement in Leishmania is genome wide [14,38] and can lead to extrachromosomal elements [13,55], to chromosomes in multiple copies and to mosaic aneuploidy [56]. It is thought that these events can lead to selective advantage [11–14] and our recent work has shed some light on the enzymes involved in these processes. RAD51 and at least one of its paralog (RAD51-4) are involved in the formation of extrachromosomal circles [14,48], while MRE11 is involved in the formation of linear amplicons [38]. MRE11 is partnering with RAD50 and NBS1 as part of the MRN complex [34,35]. MRE11 is one of the main sensor of DNA DSBs while RAD50 modulate the activity of the complex [8]. It was shown in Saccharomyces cerevisiae that the RAD50 coiled-coil domain is indispensable for MRE11 functions since truncation of this domain in RAD50 impaired telomere maintenance, meiotic DSB formation, HR and NHEJ, indicating its need for MRN activities [62]. We initiated this work to test whether MRE11 and RAD50 functions would overlap and whether these proteins are involved in the maintenance of genomic integrity. To test this we used gene inactivation and while we were able to obtain a MRE11-/- null mutant [38], it has been impossible to generate a RAD50-/- null mutant in a WT background (Fig 1 and S1 Fig). We could only inactivate both alleles if a rescue episomal copy of RAD50 was present but despite prolonged passages in absence of the selecting drug, we could not lose the episomal RAD50 copies, a strong suggestion that RAD50 is essential in Leishmania. A chromosomal copy of RAD50 was maintained upon gene inactivation if cells were complemented with a mutated RAD50K42A rescue plasmid indicating that a fully functional RAD50 is essential for cell survival (S2 Fig). In mammals, RAD50 and MRE11 are essential [41,42] but in yeast both proteins are dispensable [33,63,64], thus diverse organisms have different requirements for proteins part of the MRN complex. In both human cells and S.cerevisiae, introduction of mutations in the RAD50 ATPase domain impaired DNA binding and DNA unwinding [65] suggesting RAD50 is required for the stability of the DNA-MRN complex interaction [66,67]. We also tried to inactivate the RAD50 gene in MRE11H210Y nuclease-deficient cells but this did not lead to a RAD50 null mutant (S3 Fig). The MRE11H210Y protein is deficient in nuclease activity but still capable of DNA binding [38]. Our results provide good evidence that inactivation of RAD50 may only be possible in the absence of MRE11. There may possibly be a need for RAD50 when MRE11 is present, even if its nuclease domain is inactivated such as in MRE11H210Y. The inactivation of MRE11 nuclease activity in murine cells did not change the MRN complex formation [36], and possibly the presence of MRE11 forces the presence of RAD50 and the formation of MRE11 H210Y/RAD50 interactions. We infer that, in addition to its interaction with MRE11, the Leishmania RAD50 protein might also have important functions which would only occur in the presence of MRE11. Indeed inactivation of RAD50 was easily achieved in a MRE11-/- background (Fig 1B and 1C lanes 7). One hypothesis is that MRE11 inactivation leads to genetic compensation in Leishmania and this compensation makes RAD50 dispensable to any other putative important function that RAD50 may have. It is also possible that MRE11 is detrimental in the absence of a RAD50-mediated regulation that might happened through maintenance of the MRE11/RAD50 complex stoichiometry [68]. This hypothesis was plausible with previous observations showing that overexpression of MRE11 in Leishmania was detrimental for cell growth [38], possibly because of stoichiometry disruption.
Absence of the MRE11/RAD50 complex led to a growth defect, a sensitivity to MMS and an altered capacity for HR (S4A and S4B Fig). Although knockdown of individual components of MRN in human cells led to a decrease in the other two MRN members [69], our results suggest that RAD50 is normally expressed in the MRE11-/- strain at the RNA levels (S1C Fig) and functionally (S4C and S4D Fig). The MRE11 and RAD50 deficient cells had an incapacity of generating PTR1 linear amplicons upon MTX selection (Fig 2 and [38]). Whole genome sequencing indicated that translocations were observed in mutants lacking a fully functional MRE11/RAD50 complex (Table 1) and the only clear difference between the MRE11-/- and MRE11-/-RAD50-/- mutants was at the level of subtelomeric sequences where the number of sequenced reads was much lower in several subtelomeric loci for the MRE11-/-RAD50-/- mutant (Fig 7 and S8 Fig).
One new aspect of this work is the discovery of chromosomal translocations which have not been observed before in old world Leishmania species [70–72]. Studies in yeast have also indicated an increase of translocations events and chromosomal rearrangements when either MRE11 or RAD50 are mutated [2]. Translocations are likely to have occurred after DSBs which are usually repaired by either HR or NHEJ. Several components of NHEJ are absent in Leishmania and the parasite thus relies mostly on HR [23]. Since HR is diminished in the MRE11 and RAD50 mutants (S4E Fig), the cells may use alternative strategies to repair DNA. One of these alternative pathways used for repair of DSBs is based on MMEJ. Indeed, MMEJ has been implicated in chromosomal translocation in yeast [73], mammals [74], and in the related parasite T. brucei [75]. Mice defective in NHEJ have exhibited an increased level of translocations mediated by an alternative NHEJ that relied on microhomology [76]. The PCR reactions of the junctions created following translocations revealed that these events occurred via a mechanism of MMEJ where the microhomology is between 7 and 17 bp (Fig 6 and Table 1). There is no sequence specificity between the various translocation breakpoint sequences. We have shown previously that the Leishmania genome is filled with large repeated sequences [14] and we found that several of the microhomology sequences are either part of large repeated sequences (for T 12–17) or close to repeated sequences (for T 12–06 and T 12–18). It is well known that repeated sequences can be fragile sites and therefore more prone to DSBs and could explain translocation mediated by MMEJ [77–79]. A recent study done in Leishmania donovani has described the use of MMEJ for repair of Cas9-induced DSBs using the CRISPR-Cas9 system, even though the parasites mostly relied on HR for DSBs repair [80]. The Cas9/gRNA complex is, however, continuously generating DNA breaks in every chromosomal allele of the targeted region, complicating the search for intact homology by the HR machinery, hence favoring alternative end-joining mechanism such as MMEJ. When a template with sequence homology was provided to the parasites, the HR mechanism largely dominated the DSBs repair [80]. In our study, the homologous chromosomal allele is thought to be intact but the defect in HR probably led the cells to a MMEJ mechanism for DNA repair. Several studies done in yeast have shown the importance of resection by MRE11 for the first step of MMEJ [30,69,81–83] but we suggest MMEJ can be MRE11-independent in Leishmania. It is possible that upon genetic compensation in the knock-out strains, the expression of other nucleases is increased and could perform some of the activities usually carried out by MRE11. However, the other nucleases encoded by Leishmania (e.g EXO1, DNA2 [23]) are reputed for extensive DNA resection which is unfavorable for MMEJ that usually favors short length resections (performed by MRE11) [83,84]. Further experiments could help in deciphering the MRE11-independent MMEJ in Leishmania. Few chromosomes have been implicated in translocation and some of these chromosomes (e.g chromosomes 12, 17 and 18) have been implicated in more than one translocation. It is not clear whether the initial state of ploidy has a role to play seeing as chromosome 12 is tetraploid, chromosomes 17 and 18 are diploid but other chromosomes not involved in translocation are also polyploid like chromosomes 13, 31 and 32 (in L. infantum 263 WT). The rearrangements observed were also stable since re-sequencing of 3 clones from each null mutants after six months of continuous growth highlighted the same translocations and no additional one. Thus either Leishmania can support only few translocations or those are relatively rare events that can be maintained. Translocations and the formation of hybrid chromosomes change ploidy of specific regions (Figs 3C, 3D, 5C and 5D) but in general, a minimum of diploid state is conserved because of the overlap with the different translocations. Overall, none of the translocation breakpoints correspond to transcription initiation or termination sites and one aspect that was not studied is whether the formation of hybrids has consequences on the expression of genes in this novel context. This is particularly relevant as some of the rearrangements created regions where co-directional gene clusters diverge or converge (e.g T 12–18, T 18–20, T 08–17). Those divergent and convergent regions could represent regions where RNA polymerase can enter or exit, but transcription could also initiate or terminate within directional gene clusters [59,85,86]. Furthermore, when we compared our results to a recent study revealing that Leishmania chromosomes are replicated by a single origin (instead of multiple sites of replication origins as other eukaryotes) [87], we observed that the translocation events generated hybrid chromosomes that also contained a single origin of replication (coming from either one of the chromosome involved in the translocation). This observation suggest that even though genomic integrity was altered by the formation of hybrid chromosomes, the parasites were consistent in maintaining a single-origin of replication per chromosome.
In the MRE11-/-RAD50-/- mutant, chromosome 8 is smaller than in the WT cells and in the MRE11-/-parasites, chromosome 18 is also smaller than in the WT, suggesting that some rearrangements events (e.g deletions) happened in the mutants. While analysis of sequence reads did not allow us to confirm this deletion on chromosome 8, sequence reads has been highly useful in the past to detect changes in copy number [11,88]. We nonetheless conducted a reads depth analysis to detect either deletions or duplications and the bioinformatics analysis revealed the potential presence of several of them in the null mutants. However experimental validation by Southern blot only allowed confirming 1 out of 5 deletions and 1 out of 5 duplications deduced from the bioinformatics analyses. This suggests that the 5 kb window given by our bioinformatics pipeline might not be optimal for the detection of deletion and duplication events. Nevertheless our results suggest that there are more than translocation events as part of gene rearrangements in the nuclease null mutants.
One key change that reads depth analysis detected and that we could confirm experimentally is a reduction of reads of several subtelomeric sequences exclusively in the MRE11-/-RAD50-/- mutant (Fig 7 and S8 Fig), suggesting that the absence of RAD50 altered chromosome end stability as already observed in human cells [89]. The log2-transformed read counts would suggest that populations are not clonal but that several cells within the population have various levels of subtelomeres shortening including coding sequences (Fig 7). Decrease of sequence reads extends up to 100 kb from the telomeres in some of the cells although in general, the shortenings are smaller. The T. brucei subtelomeres harbor fragile sites [90] and subtelomeric regions are known to be more sensitive to DSBs that are processed differently than internal DSBs [91,92]. The decreased in sequence reads observed here is possibly due to an altered repair of DSBs in the MRE11-/-RAD50-/- mutant that lead to shortening of subtelomeric sequences. Indeed, a Southern blot of DNAs derived from MRE11-/-RAD50-/- hybridized with a telomeric probe revealed and hybridization smear suggesting considerable heterogeneity at chromosomes ends for individual cells within the population. RAD50 seems to have an important role in this since this phenomenon is not observed in the MRE11-/- mutant.
This study provides strong evidence that MRE11 gene knock-out is a prerequisite for RAD50 inactivation in Leishmania. Chromosomal translocations are observed in the cells lacking a fully functional MRE11/RAD50 complex, and subtelomeric regions stability is altered in the absence of RAD50. Moreover, we report for the first time in Leishmania a MRE11-independent alternative end-joining mechanism that relies on microhomology sequences. Overall, these results show a predominant role of the two DNA repair proteins MRE11 and RAD50 in chromosomal organization. Deciphering DNA repair mechanisms and maintenance of genomic integrity in Leishmania parasites may allow novel strategies for their control as they seem to rely on gene amplification and rearrangement for surviving the changing environment in which they grow.
Promastigotes of Leishmania infantum (MHOM/MA/67/ITMAP-263) and all recombinants were grown in SDM-79 medium at 25°C supplemented with 10% fetal bovine serum, 5μg/ml of hemin at pH7.0. Independent clones generated in this study were selected for methotrexate (MTX) resistance in M199 medium, using a stepwise selection starting from an EC50 of 100nM up to 1600nM of MTX. All chemical reagents were purchased from Sigma-Aldrich unless specified.
The L. infantum RAD50 null mutant (RAD50-/-) was obtained by targeted gene replacement. RAD50 flanking regions were amplified from L. infantum 263 wild-type (WT) genomic DNA and fused to blasticidin-S deaminase (BLAST), puromycin acetyltransferase (PURO) and neomycin phosphotransferase (NEO) genes using a PCR fusion based-method as described previously [93]. Briefly, 5’UTR of RAD50 was amplified using primers C and D for the BLAST cassette, primers C and E for the PURO cassette and primers C and F for the NEO cassette. The BLAST, PURO and NEO genes were amplified with primers G and H, I and J and K and L respectively. 3’UTR of RAD50 was amplified using primers M and N for all inactivation cassettes (see primer sequences in S1 Table). At least 3μg of the 5’UTR-BLAST-3’UTR, 5’UTR-PURO-3’UTR or 5’UTR-NEO-3’UTR linear fragments were transfected by electroporation (as described in [94]) in L. infantum WT, L. infantum MRE11-/- or L. infantum HYG/PUR-MRE11H210Y cells [38] to replace both RAD50 alleles. Recombinants were selected in the presence of 80μg/ml of blasticidin-S hydrochloride, 100μg/ml of puromycin dihydrochloride (Wisent) and 40μg/ml G418 (Geneticin; Sigma-Aldrich). After 4–5 passages, cells resistant to the drug selection were cloned in SDM-Agar plates (1%) in the presence of the same concentrations of drugs. PCR analysis of the recombinants was done using forward primer located in the MRE11 5’ flanking region with reverse primer inside the MRE11 gene (primers set aa’), and forward primer in the RAD50 5’ flanking region with reverse primers located inside the RAD50 gene (primers set bb’) (see primers sequences in S1 Table).
An episomal construct, Psp72-α-NEO-α-RAD50WT was designed to express RAD50 in the cells before inactivation of the second RAD50 genomic allele. Briefly, the RAD50 gene was amplified by PCR using primers O and P from L. infantum WT genomic DNA. The amplified product was first cloned in pGEM-TEasy vector and then subcloned in Psp72-α-NEO-α [95] in the HindIII and NdeI sites of the vector. Site-directed mutagenesis (Stratagene, Quickchange) was used to introduce the K42A mutation in the RAD50 ORF and generate the Psp72-α-NEO-α-RAD50K42A using primers Q and R (S1 Table). Both Psp72-α-NEO-α-RAD50WT and Psp72-α-NEO-α-RAD50K42A plasmids were then transfected by electroporation in the L. infantum BLAST RAD50-/+ mutants and cells were selected with 40μg/ml of G418 (Geneticin; Sigma-Aldrich). After inactivation of the second RAD50 genomic allele with the PURO cassette, attempts to lose the Psp72-α-NEO-α-RAD50 construct were performed by removing the G418 drug pressure up to 55 passages.
Genomic DNAs from clones were isolated using DNAzol as recommended by the manufacturer (Invitrogen). SacI or Sau3aI/AluI/RsaI digested genomic DNAs or separated chromosomes were subjected to Southern blot hybridization with [α-32P] dCTP-labeled DNA probes according to standard protocols [96]. All probes were obtained by PCR from L. infantum genomic DNAs except the telomeric probe obtained from a Psp72-PT4 [97]. Intact chromosomes were prepared from L. infantum promastigotes harvested from log phase cultures, washed once in 1X Hepes-NaCl buffer then lysed in situ in 1% low melting agarose plugs as described in [38]. Leishmania intact chromosomes were separated in 1X TBE buffer (from 10X TBE: 1M Tris, 1M Acid boric, 0,02M EDTA) by Pulsed-Field Gel Electrophoresis (PFGE) using a Bio-Rad CHEF-DRIII apparatus at 5V/cm and a 120° separation angle as described previously [47]. The range of chromosome separation was between 150 and 1500 kb.
Late log phase promastigotes (30ml) were pelleted at 3000 rpm for 5 minutes and pellets were washed once with 1X HEPES-NaCl, resuspended in suspension buffer (100mM EDTA, 100mM NaCl, 10mM Tris pH 8.0), then lysed in 1% SDS and 50μg/ml proteinase K at 37°C for 2 hours. Genomic DNA was extracted with 1 volume phenol, precipitated with 2 volumes 99% ethanol, washed with 70% ethanol twice and dissolved in 1ml 1X TE buffer. RNAse A (Qiagen) was added at 20μg/ml and DNA was incubated at 37°C for 30 minutes, followed by the addition of 50μg/ml of proteinase K and 0.1% SDS at 37°C for 30 minutes. DNA was extracted with 1 volume of phenol, precipitated and washed in ethanol, and dissolved in DNase free-water (Millipore) at 37°C overnight. Sequencing libraries were produced with the Nextera DNA sample preparation kit (Illumina Inc) according to manufacturer’s instructions.
Genome sequences were determined by Illumina HiSeq 2500 101-nucleotides paired-end sequencing. Reads from each strain were aligned to the reference genome Leishmania infantum JPCM5 (TriTrypDB version 8.0) using Burrows-Wheeler Alignment (bwa-mem) [51] with default parameters. The maximum number of mismatches was 4, the seed length was 32 and 2 mismatches were allowed within the seed. Several python and bash scripts were created for the detection of copy number variations. Briefly, chromosomes were divided into genomic windows of 5 kb and the number of reads mapping to each windows determined and normalized to the total number of reads before inter-strains comparisons. Alignments were also performed using the Lumpy-sv and Delly software [41,52] with default parameters for split-reads alignments and discordant read pairs and only translocations found with both software were kept for validation. PCR amplification of the new junction created in the translocation was performed using primers within 750 nucleotides from the translocation breakpoint on each involved chromosome. Optimal primer length was 20 nucleotides and optimal melting temperature (Tm) was 55°C. Primer sequences are presented in S1 Table.
Digestion of the Psp72-α-ZEO-α plasmid [98] using PciI and XbaI enzymes was performed and the isolated α-ZEO-α fragment was used to target the alpha-tubulin loci in order to monitor the integration efficiency. Briefly, 2x106 cells from WT, MRE11-/- and MRE11-/-RAD50-/- strains were transfected with 5μg of the linear α-ZEO-α construct. After 24h following electroporation, cells were plated on SDM-Agar plates (1%) containing Zeocin (Invitrogen) at 1mg/ml. All strains were also transfected with the plasmid Psp72-α-ZEO-α and with sterile water as controls. Colonies were counted after 10–15 days of plating.
RNAs were extracted using RNeasy plus mini kit (Sigma) according to the manufacturer recommendations. The cDNA was synthesized using Oligo dT12-18 and SuperScript II RNase H-Reverse Transcriptase (Invitrogen) and amplified in SYBR Green Supermix (Bio-Rad) using a rotator thermocycler Rotor Gene (RG 3000, Corbett Research). The expression level was derived from three technical and three biological replicates and was normalized to constitutively expressed mRNA encoding glyceraldehyde-3-phosphate dehygrogenase (GAPDH, LinJ.36.2480). The sequences of the primers used in this assay are listed in S1 Table.
L. infantum WT, RAD50-/- Psp-RAD50, MRE11-/- and MRE11-/-RAD50-/- were resuspended at a concentration of 5x106 cells/ml and exposed to increasing doses of MMS (Sigma–Aldrich). Cells were counted after 72h and reported in survival rate.
Reactions (10 μl) contained 40nM of Leishmania infantum RAD50 or RAD50K42A (purified by double affinity purification accordingly to [99]) in 50mM Tris-HCl pH 7.5, 1mM Mg(CH3COO)2, 1mM DTT and 100 μg/ml BSA supplemented with 50 nCi [ɣ-32P]ATP (3000 Ci/mmole; Perkin Elmer Life Sciences). Aliquots (5 μl) were removed, stopped by addition of EDTA, and the percentage of ATP hydrolyzed was determined by thin layer chromatography followed by quantification using a Fujifilm Phosphoimager.
The data set supporting the results of this article is available at the EMBL-EBI European Nucleotide Archive (http://www.ebi.ac.uk/ena) under study accession number PRJEB11440 with sample accessions ERS934506, ERS934507 and ERS934508 for L. infantum MRE11-/-RAD50-/-, L. infantum MRE11-/- and L. infantum JPCM5, respectively. L. infantum 263 WT sequencing data is available under the study ERP001815 and sample accession number ERS179382.
|
10.1371/journal.pgen.1004225 | An Insulin-to-Insulin Regulatory Network Orchestrates Phenotypic Specificity in Development and Physiology | Insulin-like peptides (ILPs) play highly conserved roles in development and physiology. Most animal genomes encode multiple ILPs. Here we identify mechanisms for how the forty Caenorhabditis elegans ILPs coordinate diverse processes, including development, reproduction, longevity and several specific stress responses. Our systematic studies identify an ILP-based combinatorial code for these phenotypes characterized by substantial functional specificity and diversity rather than global redundancy. Notably, we show that ILPs regulate each other transcriptionally, uncovering an ILP-to-ILP regulatory network that underlies the combinatorial phenotypic coding by the ILP family. Extensive analyses of genetic interactions among ILPs reveal how their signals are integrated. A combined analysis of these functional and regulatory ILP interactions identifies local genetic circuits that act in parallel and interact by crosstalk, feedback and compensation. This organization provides emergent mechanisms for phenotypic specificity and graded regulation for the combinatorial phenotypic coding we observe. Our findings also provide insights into how large hormonal networks regulate diverse traits.
| Insulin signaling is widely implicated in regulating diverse physiological processes ranging from metabolism to longevity across many animal species. Many animals have multiple insulin-like peptides that can regulate the activity of this signaling pathway. For example, while humans have ten, including the well-studied insulin hormone, the nematode Caenorhabditis elegans has forty such peptides. The similarity among these insulin-like peptides led to the predominant notion that widespread redundancy occurs among these peptides. Contrary to this notion, we find that the forty insulin-like peptides in the nematode C. elegans have specific and distinct effects on eight different physiological outputs that range from development, stress responses, lifespan and reproduction. Interestingly, we also find that these peptides regulate each other at the transcriptional level to form a signaling network. In addition, we observe that this network is organized into parallel circuits, whose activities are affected by compensation, feedback and crosstalk. Finally, the organization of the network helps to explain how different combinations of peptides generate specific outputs and captures the complexity of how these peptides orchestrate an animal's physiology through distinct peptide-to-peptide signaling circuits.
| The organization and integration of multiple signals endow intercellular regulatory networks with information processing capabilities. For example, hormones modulate physiology and maintain homeostasis in variable environments [reviewed in 1], and morphogens give rise to intricate patterns during development [reviewed in 2]. Nevertheless, how simple circuits are organized into complex networks that perform sophisticated functions is not fully understood.
The ILPs are a superfamily of hormones that regulate many processes, including development, cell proliferation, energy metabolism, neuronal function, reproduction stress resistance, and longevity [3]–[15]. Canonical ILP signaling is mediated by a receptor tyrosine kinase pathway that culminates in the regulation of FOXO transcription factors and other regulatory molecules [16]. In Caenorhabditis elegans, this occurs via the DAF-2 ILP receptor tyrosine kinase, which signals through DAF-16 FOXO [6], [17]–[19]. The importance of ILP signaling is underscored by the conservation of both the signal transduction pathway and the processes they regulate. Indeed, a C. elegans ILP (INS-6) resembles human insulin structurally and can bind and activate the human insulin receptor [20].
Most animal genomes encode multiple ILPs: humans have 10 [21]; Drosophila melanogaster has 8 [22]–[24]; and C. elegans has 40 [25]–[27]. Small-scale studies have shown that certain ILPs can regulate other ILPs [4], [24], [28]–[31], and that ILPs can act as either agonists or antagonists of their receptor to differentially affect multiple processes [5], [24], [27]. How do such simple interactions between these hormones generate complex functionality? Here, we address this question by an integrated analysis of the C. elegans ILPs during larval development, stress resistance, reproduction and lifespan. We systematically tested the function of C. elegans ILPs in the control of diverse phenotypes. In contrast to the common notion of broad redundancy among ILPs [32], we now provide evidence supporting a combinatorial code of action that maps the ILPs to multiple phenotypes. We also uncover the existence of a C. elegans ILP-to-ILP regulatory network that reveals the mechanisms through which multiple functionally diversified ILPs interact to regulate complex developmental and physiological traits. Thus, our analysis of the ILP-to-ILP network provides organizational principles for multiple-gene families and signaling networks.
As in many animals, the C. elegans daf-2 insulin/ILP receptor pathway affects multiple physiological processes, including development, aging, pathogen resistance, thermotolerance and reproduction [6], [8], [9], [17], [33]–[37]. The C. elegans ILP pathway also regulates entry into a specialized form of larval arrest known as dauer that forms preferentially under adverse conditions, such as high temperature, high population density, and low dietary sterols and food levels [reviewed in 6]. Under favorable conditions, animals exit the dauer stage to resume reproductive growth. Dauer exit is also regulated by the ILP pathway [34], [38], which suggests that ILPs function to regulate developmental plasticity in response to complex environmental cues [5], [31].
Previous studies, which focused on a few ILPs, suggest that different phenotypes are modulated by distinct ILPs [4], [5], [11], [26], [27]. These ILPs can exhibit complex functional interactions in the regulation of certain phenotypes [4], [5]. Together, these observations have raised the possible existence of an ILP combinatorial code in regulating physiology, in contrast to the prevailing notion of widespread redundancy as a feature of the ILPs and other gene families [32]. We tested this possibility by mapping the relationships between the 40 C. elegans ILPs, ins-1 to ins-39 and daf-28 [25]–[27], and their developmental and physiological outputs. We systematically tested mutants in 35 ILPs for 8 distinct developmental and physiological phenotypes (Figures 1 and S1). Thirty-four of these mutations delete part or all of the coding sequence of an ILP and are predicted to be null mutations. One mutation, ins-10(tm3498), contained a deletion in the genomic sequence and a duplication that overexpressed the intact coding sequence, which represents a gain-of-function allele (Table S5, Figure S2 and Materials and Methods). To minimize genetic background effects, all mutants were outcrossed 6 times to wild type.
We used well-established procedures to score the ILP mutants and applied several statistical criteria to classify the phenotypes as high or low confidence based on statistical significance and reproducibility (see Materials and Methods). We also confirmed the roles of many ILPs that showed new phenotypes or represented key conclusions in this study by rescuing their phenotypes with a transgene bearing the wild-type copy of the corresponding gene, as described in the following sections and in Table S3. We included our previously published work in the analysis for comparison (Figures 1 and S1, Table S2) [4], [5]. Importantly, we implicated distinct combinations of ILPs in every process tested and ascribed new functions to more than half of the C. elegans ILPs (Figure 1): 66% (23/35) of those tested showed at least one high-confidence phenotype, and 89% (31/35) showed high- or low-confidence phenotypes. We focused our analysis on the high-confidence hits.
ILP-to-ILP signaling regulates several physiological processes [4], [24], [30]. To investigate its global nature, we used quantitative real-time PCR (qPCR) to identify changes in the mRNA levels of all 40 ILPs in each of 35 ILP mutants (Figure 2A, S2, and Table S5). Surprisingly, we found that ILP-to-ILP signaling extends to many members of this family, demonstrating the presence of an ILP-to-ILP regulatory network (Figure 2A). Out of a possible 1190, we observed only 101 ILP interactions (Figure 2A), which suggests that the inter-ILP regulation is sparse. These regulatory relationships also appear specific and diverse: each ILP is wired to a unique combination of regulators and targets, and regulation could be either negative (52%) or positive (48%) in a target-specific manner. These relationships showed an intermediate modularity of 0.49, reflecting a mix of cross-regulation and compartmentalization in ILP gene expression. Thus, like the phenotypic screens, the qPCR data show that the diversification of C. elegans ILPs beyond functional redundancy also extends to their gene expression.
For comparison, we also analyzed the changes in expression of all 40 ILPs in mutants that impair the ILP signaling pathway, using daf-2(e1368) (a reduction-of-function allele), and daf-16(mu86) (a null allele) [18], [34]. Many ILPs were up-regulated in the daf-2(e1368) background, suggesting compensation. Many of the ILPs that were regulated by other ILPs were also affected in the daf-2(e1368) and daf-16(mu86) backgrounds, suggesting that these changes were mediated through the canonical ILP signaling pathway. In general, daf-2(e1368) and daf-16(mu86) tend to cause larger effects on gene expression, suggesting that they might be closer to the upper limit of the gene expression changes, as might be expected if the central pathway for ILP signalling is disrupted. Some ILPs that were regulated by other ILPs were not affected by daf-2(e1368) or daf-16(mu86); this difference could be due to residual signaling activity retained in daf-2(e1368) [34] or the use of alternative pathways for inter-ILP regulation.
To understand inter-ILP communication, we built a network based on these qPCR results for graph theory analysis, treating each ILP as a node and each regulatory interaction as an edge (Figure 2B). In this network, the edges are directed (reflecting the regulation of one ILP by another) and signed (indicating positive or negative regulation) to represent the flow of information.
We discovered three major properties of this network. First, the ILP network had “small world” properties defined by two key parameters: the characteristic path length that measures the average minimal number of edges between all possible pairs of ILPs, and the clustering coefficient that measures the density of local interconnections [50], [51]. Compared with random networks with the same number of edges and nodes, the ILP network has a short path length, 3.17, and a high clustering coefficient, 0.13 (Figures S3A to S3C). Respectively, these properties might suggest that within these genetic circuits, signals can be communicated relatively efficiently from one ILP to another because they are separated by very few intervening ILPs, and that information is processed by local genetic circuits. These are consistent with the parallel processing we observed in the dauer entry sub-network, which is discussed below.
Second, the ILP expression network displayed hierarchical regulation. Plotting the number of regulators (in-degree) versus the number of targets (out-degree) of each ILP (Figure 2C) reveals a regulatory hierarchy where several ILPs had an exceptionally high number of regulators or targets. This organizational feature suggests different functional attributes for the ILPs. ILPs with few inputs and many outputs are putative upstream regulators; ILPs with similar numbers of inputs and outputs likely act in relays or processing circuits; and ILPs with many inputs and few outputs could serve as downstream integrators or effectors.
Third, important nodes for network communication tend to affect more processes. We calculated the betweenness centrality for each ILP, which measures its importance as a link between other ILP pairs in the network (Figure 2D) [52]. ILPs with higher betweenness centrality were more likely to be pleiotropic (Figures 2D to 2E), similar to protein-interaction networks where proteins with high betweenness centrality tend to be essential [52]. Thus, ILPs with high betweenness centrality may act as bottlenecks during information flux in a wider range of processes.
Our network analysis was robust to missing edges, such as those from subtle gene expression changes that did not rise to statistical significance. The top ranked ILPs for each network parameter were similar despite the addition or removal of 25% of random edges (Figures S3H to S3K), indicating that we have sampled the network sufficiently.
To relate ILP function to network organization, we mapped the high-confidence ILPs identified in each screen onto the network, which provided three global observations. First, the ILPs with phenotypes were spread over the network (Figure 3A), suggesting that signaling across many parts of the network was important for its overall function. Second, the ILPs with more specific phenotypes from the non-sensitized screens were segregated into different locations (Figures 3B to 3F and 3J), consistent with the observations that gene expression defects in these ILP mutants do not propagate over the entire network (Figure 2A). The separation of critical nodes in the network could limit the number of physiological defects when one ILP is perturbed. Third, our sensitized screens for dauer entry revealed another functional level of non-critical ILPs distributed over much of the network (Figures 3F to 3I). This suggests distributed processing, which could reduce the severity of a phenotype by providing alternate routes of communication. Together, these mechanisms contribute to functional specificity, which is an aspect of the ILP combinatorial code.
To address how ILPs combinatorially regulate a specific process, we analyzed genetic interactions among deletion mutations of ILPs involved in dauer entry. We tested 56 double mutant combinations by selecting a diverse subset of 13 ILPs identified from each of the three dauer entry screens, encompassing ILPs showing high and low penetrance (Figures 1B to 1D). To classify genetic interactions, we first determined how the fraction of dauer entry in the double mutant differed from the expected fraction in an additive model based on the single mutant phenotypes (Materials and Methods). We then subdivided the interactions based on whether the corresponding single mutants had the same or opposite phenotypes (Figure 4, Table S6). This analysis revealed a level of diversity in gene interactions not predicted by simple redundancy.
Diverse genetic interactions (defined in Figure 4) were observed in 47% (26/56) of the double mutants, of which 38% (10/26) were additive or synergistic. This result indicates that while the choice between dauer arrest and reproductive growth is binary, the likelihood of a given choice is specified by a graded combination of ILP activities. Strikingly, 9 of these 10 additive or synergistic interactions were seen in double mutants with null mutations in either ins-35 or daf-28, suggesting that these ILPs are important genetic hubs in dauer entry, consistent with their strong dauer entry phenotypes. The remaining 53% (30/56) of the double mutants showed no effect or no interaction (Figure 4, Table S6), indicating that ILPs are not promiscuous in their interactions during dauer entry, even with other ILPs involved in the same process. These results reveal how signals from pairs of ILPs are integrated to regulate dauer entry. Our findings also demonstrate functional differences among ILPs that regulate dauer entry, and indicate that the effect of an ILP depends on genetic background.
Information processing is strongly influenced by the signaling motifs within the network and the overall network architecture [53]. While regulatory interactions serve as a roadmap for information flow among ILPs, genetic interactions between ILPs reflect how their activities are integrated to generate a physiological outcome. To assess information flow and processing, we combined regulatory and functional data for the ILPs whose genetic interactions were extensively defined for the dauer entry phenotype (Figure 5).
The connectivity and synergistic or additive genetic interactions indicate parallel signaling in the dauer entry sub-network (Figures 5A to 5B). The major signals that inhibit dauer entry come from three main branches (daf-28, ins-6/ins-33 and ins-6/ins-35), because mutants in these branches have the strongest phenotypes (Figure 1C). To generate graded probabilities of dauer entry, signals from these three branches are integrated in an additive or synergistic manner based on their genetic interactions (Figures 4 and 5B). This network organization was supported by the phenotypes observed when we disrupted the daf-28, ins-6/ins-33 and ins-6/ins-35 branches using combinations of null mutations. In the ins-33 and daf-28 double deletion mutant, we observed a strong synergistic response with a high proportion of dauers even at 25°C (Figure 5D, Table S7). Strikingly, in the ins-33; daf-28; ins-35 triple deletion mutant, up to 80% dauers were observed at 25°C (Figure 5D, Table S7), which is nearly comparable to daf-2 mutants. These results reinforce the idea that the daf-28, ins-6/ins-33 and ins-6/ins-35 branches are major pathways for regulating dauer entry.
The different connectivities within each branch of the ILP network suggest that they use different information processing strategies (Figure 5). In the daf-28 branch, daf-28 inhibits ins-26, which likely serves as a compensatory regulation based on their synergistic interaction (Figures 4 and 5B). The effect of this compensation is likely to be regulation of ins-5 as both daf-28 and ins-26 inhibit ins-5. In contrast, the ins-6 branches have a bifurcated topology where ins-33 and ins-35 process inputs from ins-6. A non-additive interaction was observed between ins-6 and ins-35, as well as between ins-6 and ins-7, which is downstream of ins-35 (Figures 4 and 5C); while an additive interaction between ins-6 and ins-33 indicates compensation (Figures 4 and 5B; see below). At a downstream level, non-additive or non-synergistic interactions occur within the ins-33 or ins-35 branches, but not the daf-28 branch. Crosstalk occurs between the daf-28 and ins-6 branches (Figure 5A), which may coordinate their signaling activities.
ins-3 is likely to act as a negative modulator providing feedback to the dauer entry sub-network at multiple levels; such circuits are associated with noise reduction and homeostasis. Unlike most ILPs, the ins-3 mutation decreased dauer entry in several backgrounds (Figures 1D and 4). ins-3 expression was activated by ins-6; while ins-3 in turn inhibited ins-6 expression, as well as other ILPs in the daf-28 branch (Figures 2A and 5A).
While both ins-14 and ins-17 show high and low-confidence dauer entry phenotypes, respectively, they are likely to act separately as modulators in the main dauer entry sub-network (Figure 5) for two reasons. First, they are not directly connected to the ins-6 and daf-28 branches of the expression network (Figures 5A to 5C). Second, they have weaker interactions with the genes in the daf-28 and ins-6 branches (Figure 4A). One exception is an additive interaction between ins-35 and ins-14 (Figures 4 and 5B), which might represent cross-talk at the downstream level.
ILPs could also exert either strong or weak effects (Figure 1). For example, although ins-6 and daf-28 both regulate dauer entry and exit, ins-6 null mutations had a stronger effect on dauer exit, whereas daf-28 null mutations had a stronger effect on dauer entry [5]. Our results reveal that this feature is common in the whole ILP system (Figure 1). These specificities are not due to some ILPs being generally strong signals, while others are generally weak, because the relative effects of the ILPs can be reversed depending on the phenotypes.
Our integrated analysis provided a mechanistic explanation for the phenotypic specificity of daf-28 and ins-6 (Figure 1C) during dauer entry. Loss of daf-28 is compensated by ins-26, because ins-26 was up-regulated in daf-28 mutants (Figures 2A and 5A) and because ins-26; daf-28 double mutants have a more severe phenotype than either single mutant (Figures 4 and 5B). However, ins-26 is a weak compensator, as indicated by its weak phenotype (Figure 1C). Additionally, daf-28 mutants up-regulate ins-5 (Figures 2A and 5A), an ILP that can promote dauer entry (Figures 1 and 4). As opposed to compensation, increased ins-5 expression contributes to the mutant phenotype of daf-28, because removing ins-5 suppressed the daf-28 mutation (Figures 4 and 5C). These two targets of daf-28 therefore contribute to its strong dauer entry phenotype.
In contrast, ins-6 is compensated by daf-28 and ins-33, because both daf-28 and ins-33 were up-regulated in ins-6 mutants (Figures 2A and 5A), and both ins-6; daf-28 and ins-33; ins-6 double mutants had a more severe phenotype than the respective single mutants (Figures 4 and 5B). Both daf-28 and ins-33 were strong compensators, as indicated by their strong phenotypes (Figure 1C). Thus, the weak ins-6 phenotype could be explained by compensation from two strong regulators.
Together, these results show that connectivity within the ILP network serves as an important determinant of functional differences among ILPs.
Most animals, including humans, encode multiple ILPs in their genomes, which regulate multiple processes [4], [5], [14], [15], [23], [24], [26], [27], [54]–[58]. However, the biological function of large ILP ensembles remains an open question. Our systematic analysis of C. elegans ILPs revealed that they are organized into an ILP-to-ILP network that provides several regulatory mechanisms for graded signaling, functional diversity, robustness to gene perturbation and information flow. In turn, these functional properties of the ILP network generate aspects of a combinatorial code that links ILPs to developmental and physiological outputs. Thus, our findings challenge the notion that broad redundancy is the central feature of the C. elegans ILP family.
Large gene families are often proposed to employ a combination of redundancy and diversity to regulate biological processes [59]. Here, we reveal the specific implementation of an ILP combinatorial code that coordinates aspects of development and physiology (Figure 1A). Different ILPs generally affect different combinations of processes, which support the idea that redundancy is not evolutionarily stable unless the genes have additional functions [59], [60]. The high-confidence phenotypes indicate that many single ILPs can significantly contribute to different phenotypic outputs. This combinatorial coding of phenotypes therefore argue against simple redundant mapping between ILPs and their outputs, but show that the complexity of these gene-phenotype relationships is generated at least in part by inter-ILP communication.
The intermediate modularity of the ILP phenotypes raises the possibility that multiple ILP signaling centers exist in the animal, which could provide differential contributions to different processes. In addition to the regulatory connectivity that underlies phenotypic specificity, spatial specificity in ILP signaling could also be a complementary mechanism in achieving the specific patterns of ILP phenotypes. This model will need to be tested in the future by tissue- or cell-specific rescue of the ILPs, coupled with the elucidation of their downstream target tissues where the DAF-2 ILP receptor acts.
Undirected networks have been recently used to group the C. elegans ILPs based on similarities in their expression patterns [31]. Here we show that the C. elegans ILPs are organized at the level of ILP-to-ILP regulation in a directed regulatory network, where signals in different branches are processed differently and modulated by cross-talk. This is exemplified in the different connectivities between the ins-6 and daf-28 branches of the dauer entry subnetwork, whose distinct signals are ultimately integrated to set the probability of dauer entry. This network organization thus contributes to the graded nature of the ILP combinatorial code. This property generates different probabilities of dauer entry that result in different fractions of developmentally arrested dauers versus reproductive adults within a population. Dauers can survive environmental insults that kill reproductive adults and can thus serve as a hedge at the cost of delayed reproduction. Therefore, the advantage of this graded response provided by the parallel circuit organization is the ability to optimize the trade-off between fast reproduction versus survival in response to variable environments.
These findings further underscore how circuit organization in a network contributes to the phenotypic outputs of a multi-gene family. Compensation and distributed, parallel processing in the ILP network provide robustness against gene or network perturbation. Robustness in preventing dauer entry allows for rapid reproduction, ensuring that animals develop as dauers only in extreme conditions, such as when the environment impinges on more than one ILP.
In addition, the connectivity of the ILP network show that specific compensatory circuits are organized to generate strong and weak regulators, an important component of the combinatorial code. Extensive genome-wide studies in yeast indicate that complete or partial functional redundancy can occur among duplicated gene pairs [61], [62] where the loss of one gene can be compensated by responsive circuits that increase the expression of a second homologous gene [63]. Although compensatory circuits are often hypothesized as a feature of gene families that lead to redundancy, we show that its actual implementation can lead to more complex outcomes than previously proposed. Instead of global redundancy, the gradation provided by the ILP network is consistent with the idea that partial redundancy, as well as overlapping and distinct functions, could serve to encode diverse inputs [59], [60].
ILP-to-ILP signaling in diverse animals uses similar signaling motifs, such as feedback, compensatory inhibition and feedforward circuitry [4], [24], [28]–[30], [64], [65], which may provide similar biological functions despite component differences [53]. Our findings suggest how simple circuits can be organized to generate complex network functions; like signaling motifs, these principles may also apply to networks in general. Because our results indicate the importance of specificity versus redundancy in multi-gene families is a consequence of network organization, we propose that large-scale connectivity-based approaches have general utility in dissecting the regulatory mechanisms employed by different families of intercellular signals in different animals.
In summary, we have delineated the C. elegans ILP-to-ILP regulatory network based on functional criteria, which provides a distinct approach to existing ILP networks based on expression similarities [31]. This ILP-to-ILP regulatory network, coupled with our systematic genetic analyses, serves as a mechanistic framework for understanding information processing by ILPs. Our findings suggest that the multiple ILPs provide the ability to organize circuits into a network with diverse points of regulation, which in turn produces an intricate combinatorial code to orchestrate development and physiology. Together, this represents a new avenue to understand how hormonal systems compute the development and physiology of the organism.
C. elegans were cultivated at 20°C under standard conditions except where otherwise stated. The strains used are listed in Table S1. All ILP deletions were independently confirmed using PCR from genomic DNA with primers different from those used by the C. elegans Knockout Consortium to isolate the mutation. ins-10(tm3498) had increased expression of the coding region from our qPCR experiments (below). PCR using genomic DNA from 6× outcrossed ins-10(tm3498) mutants with primers that annealed to the start and end of the ins-10 coding sequence amplified a genomic fragment that contained the full ins-10 coding sequence which was verified by sequencing (data not shown). Because ins-10(tm3498) also contained a deletion in the endogenous ins-10 locus, which we verified independently from the C. elegans Knockout Consortium, these results indicate that ins-10(tm3498) involves at least a deletion and duplication of the ins-10 coding region that led to ins-10 overexpression.
All mutant strains used in this study were obtained from the Knockout Consortium [66]. Double and triple mutants were generated by standard genetic methods. See Table S1 for strain list. Deletions were regularly verified using PCR.
All the phenotypic assays were conducted on fresh NGM plates seeded with fresh OP50 unless specified otherwise, using animals that were well fed for at least 2 generations. The lifespan and dauer assays were replicated in different labs. The identity of each strain was blinded for most assays.
We generated plasmids to rescue the phenotypes of the ILP mutants. These plasmids contain the entire coding region of the gene of interest and the 5′ and 3′ intergenic regions up to the next open reading frame. Genomic regions for ins-3, ins-4, ins-5, ins-14, ins-15, ins-21, ins-23, ins-26, and ins-27 were subcloned using a recombineering method [69] from the corresponding fosmids into the pQL60, a vector derived from the original pPUB in which the unc-119 marker was removed. Genomic regions for ins-31, ins-33 and ins-35 were amplified by PCR and subcloned into pCR-Blunt TOPO (Invitrogen). The transgenic lines bearing extrachromosomal arrays were generated by microinjection of the rescue construct at different concentrations (see Table S1) as well as ofm-1::gfp as a coinjection marker (25 ng/µl) and pBluescript as a carrier DNA up to a final concentration of 100 ng/µl of DNA.
For ins-12, a mini-gene was synthesized, subcloned into the MosSCI plasmid pCFJ352 [70] with the corresponding the 5′ and 3′ intergenic regions up to the next open reading frame and integrated into the QL35 strain using MosSCI [71].
Modularity of the ILP-to-phenotype and the mRNA-to-ILP matrices were estimated by rearranging the rows and columns of the matrix to find highly interconnected groups and then assessing matrix-wide the ratio of the number of inside to outside group connections. We used the adaptive BRIM (Bipartite Recursively Induced Modules) algorithm [49], [76], which is a heuristic method, implemented in MATLAB [49], [76] to maximize a bipartite modularity value Q. This Q value is dependent on modularity of the matrix; by definition, a perfectly modular matrix is comprised of clusters of completely isolated groups (), and modularity declines as the number of cross-group connections increases (). Because the modularity calculation is based on a stochastic algorithm that produced different matrix arrangement each time the algorithm is run, we performed the calculation 30 times and took the average of the modularity. The average modularity value of ILP-to-phenotype matrix is (highly reproducible) and that of mRNA-to-ILP is To evaluate the statistical significance of the modularity, we utilized two null models. The first model is a Bernoulli random null model in which the null matrix has the same total number of interactions as the original matrix, albeit randomly positioned. The second is a probabilistic degree null model in which each interaction in null model is assigned a probability. The ILP-to-phenotype and mRNA-to-ILP matrices are significantly different against the Bernoulli random null model (p<0.001 in both cases); however, when compared against the probabilistic degree null model, which is a stronger statistical test, the p-values of both matrices are greater than 0.05. These results suggest that both matrices are weakly modular.
|
10.1371/journal.pntd.0006372 | Risk factors for human acute leptospirosis in northern Tanzania | Leptospirosis is a major cause of febrile illness in Africa but little is known about risk factors for human infection. We conducted a cross-sectional study to investigate risk factors for acute leptospirosis and Leptospira seropositivity among patients with fever attending referral hospitals in northern Tanzania.
We enrolled patients with fever from two referral hospitals in Moshi, Tanzania, 2012–2014, and performed Leptospira microscopic agglutination testing on acute and convalescent serum. Cases of acute leptospirosis were participants with a four-fold rise in antibody titers, or a single reciprocal titer ≥800. Seropositive participants required a single titer ≥100, and controls had titers <100 in both acute and convalescent samples. We administered a questionnaire to assess risk behaviors over the preceding 30 days. We created cumulative scales of exposure to livestock urine, rodents, and surface water, and calculated odds ratios (OR) for individual behaviors and for cumulative exposure variables.
We identified 24 acute cases, 252 seropositive participants, and 592 controls. Rice farming (OR 14.6), cleaning cattle waste (OR 4.3), feeding cattle (OR 3.9), farm work (OR 3.3), and an increasing cattle urine exposure score (OR 1.2 per point) were associated with acute leptospirosis.
In our population, exposure to cattle and rice farming were risk factors for acute leptospirosis. Although further data is needed, these results suggest that cattle may be an important source of human leptospirosis. Further investigation is needed to explore the potential for control of livestock Leptospira infection to reduce human disease.
| Leptospirosis is an under-recognized but important cause of febrile illness and death in Africa. The bacteria that cause leptospirosis have their usual life cycle in animals; humans are infected as accidental hosts. There is considerable variation between countries as to which reservoir animals and human activities are important for transmission of leptospirosis to humans. In many tropical countries flooding and rodents are the dominant sources of human infection. However, in Africa it is unknown which sources of leptospirosis are most responsible for human infection and what behaviors put people at risk for infection We performed a prospective cross-sectional study, to identify risk factors for acute leptospirosis and sources of human infection. We identified contact with cattle and work in rice fields as risk factors for acute leptospirosis. Our findings indicate that cattle may be an important source for human leptospirosis, and therefore control of leptospirosis in livestock may help prevent leptospirosis in people. Further work is needed to isolate Leptospira from humans and livestock. Rice farming was an uncommon activity in our study, but strongly associated with acute leptospirosis. Efforts are warranted to prevent infection in rice farmers living in Africa.
| Leptospirosis is a zoonotic bacterial infection and is increasingly recognized as an important cause of fever in Africa [1]. Leptospirosis was a leading cause of severe febrile illness in a study conducted in northern Tanzania during 2007–8, where it was diagnosed in 8.8% of participants [2]. The annual incidence of severe acute leptospirosis in northern Tanzania is high, but has fluctuated during surveillance over two time periods: from 75–102 cases per 100,000 people in 2007–08 to 11–18 cases per 100,000 people in 2012–14, suggesting dynamic transmission patterns [3]. An understanding of major animal reservoirs, sources, and modes of transmission to humans is required to inform leptospirosis control.
Animals infected by Leptospira may become carriers and excrete Leptospira in urine leading to environmental contamination. Humans can be infected following direct exposure to the urine of infected animals or through contact with contaminated surface water or moist soil [5]. Portals of entry include mucous membranes and broken skin [5]. While the major reservoirs, sources of human infection, and modes of transmission of infection are established on a global scale, there is substantial variation by location reflecting the diverse ecology of Leptospira. In many tropical countries, rodent species are considered the most important animal reservoir for human infection [4]. As such, dominant risk factors for leptospirosis in many tropical countries include activities that expose individuals to rodent urine, such as living in urban slums, proximity to sewers, and exposure to flood waters [4, 6, 7]. In Tanzania and most other African countries, the risks factors for human infection are not well characterized [1, 4], and there is some evidence that the risk factors may differ from other tropical countries. In northern Tanzania there is evidence that leptospirosis is more common in rural areas where both livestock and rodents could be important sources of human infection [8], and previous Leptospira exposure studies have identified livestock farmers as a high-risk group for Leptospira seropositivity [9]. Serogroup reactivity patterns of acute human leptospirosis infections have also suggested that livestock may be reservoirs for human cases [8], and studies of livestock have found high proportions that were seropositive or with leptospiruria [10–12]. To inform leptospirosis control in Tanzania, we aimed to identify risk factors for acute leptospirosis and Leptospira seropositivity, and identify sources of human Leptospira infection.
We conducted a cross-sectional study at Kilimanjaro Christian Medical Centre (KCMC), a 450-bed zonal referral hospital and, Mawenzi Regional Referral Hospital (MRRH) a 300-bed regional referral hospital, both in Moshi. Moshi (population ~180,000) is the administrative capital of the Kilimanjaro Region (population ~1.6 million) of Tanzania. Moshi is situated at approximately 890 meters above sea level and has a tropical climate with rainy seasons from October through December, and March through May. Agriculture in northern Tanzania includes smallholder systems involving mixed crop and livestock farming, as well as pastoralism.
We enrolled pediatric and adult patients presenting to KCMC and MRRH from February 2012 through May 2014. From Monday through Friday, we screened all patients in the adult medical ward at KCMC and the adult and pediatric medical wards at MRRH within 24 hours of admission, as well as patients presenting to the outpatient department at MRRH. We enrolled consecutive eligible inpatients and every second eligible outpatient. Patients were eligible to participate if they had an axillary temperature of >37.5°C or a tympanic, oral, or rectal temperature of ≥38.0°C at presentation. Inpatients were also eligible if they reported a history of fever within the past 72 hours. After obtaining informed consent, a trained study team member completed standardized clinical history and risk factor questionnaires. The risk factor questionnaire included questions on socio-demographic characteristics, participant living environment, and daily activities performed over the past 30 days, focusing specifically on animal-related activities, exposure to surface water and to rodents (S1 Text). The questionnaire was designed to include established risk factors for leptospirosis from studies done in other settings [4, 6, 7, 13–15], and was piloted prior to use. For participants who lived in the Kilimanjaro Region, study personnel visited participant households to record Global Positioning System (GPS) coordinates of participants’ dwellings. Clinician diagnoses were recorded. Participants were asked to return 4–6 weeks after enrollment for collection of a convalescent serum sample.
Blood was allowed to clot for between 30 and 60 minutes. It was then centrifuged for 15 minutes at 1,126–1455 relative centrifugal force to separate serum. Serum was stored at -80°C. Serum specimens were batch shipped on dry ice from Moshi, Tanzania to Atlanta, GA, United States of America for testing. Serology for leptospirosis was performed at the US Centers for Disease Control and Prevention using the standard microscopic agglutination test (MAT) with a panel of 20 Leptospira serovars belonging to 17 serogroups [16]. These included: Australis (represented by L. interrogans serovar Australis, L. interrogans serovar Bratislava), Autumnalis (L. interrogans serovar Autumnalis), Ballum (L. borgpetersenii serovar Ballum), Bataviae (L. interrogans serovar Bataviae), Canicola (L. interrogans serovar Canicola), Celledoni (L. weilii serovar Celledoni), Cynopteri (L. kirschneri serovar Cynopteri), Djasiman (L. interrogans serovar Djasiman), Grippotyphosa (L. interrogans serovar Grippotyphosa), Hebdomadis (L. santarosai serovar Borincana), Icterohaemorrhagiae (L. interrogans serovar Mankarso, L. interrogans Icterohaemorrhagiae), Javanica (L. borgpetersenii serovar Javanica), Mini (L. santarosai serovar Georgia), Pomona (L. interrogans serovar Pomona), Pyrogenes (L. interrogans serovar Pyrogenes, L. santarosai serovar Alexi), Sejroe (L. interrogans serovar Wolffi), and Tarassovi (L. borgpetersenii serovar Tarassovi). MAT was performed beginning at a dilution of 1:100, with subsequent two-fold dilutions. Positive and negative controls were included with each run.
We defined leptospirosis cases as participants with either a four-fold rise in agglutinating antibody titers between acute and convalescent serum, or a single reciprocal titer of ≥800 [17]. Seropositivity was defined as a single positive reciprocal titer of ≥100 from either sample. Controls were participants with negative titers on both acute and convalescent serum samples. The predominant reactive serogroup for cases and seropositive participants was defined as the serogroup containing the serovar with the highest titer.
For each participant, village population density was calculated from the 2012 Tanzania Population and Housing Census [18]. For the purpose of analysis, a priori zone classifications were applied to each village [19]. Villages with a population density of 10 inhabitants/km2 were classified as urban; villages ≤15km distance from urban areas with a population density ≥3 and < 10 inhabitants/km2 were classified as peri-urban; and villages ≥15km distance from an urban area with a population density of <3 inhabitants/km2 [19]. Georeferenced mean annual rainfall and soil type data were obtained from the 2002 Kenya International Livestock Research Institute report [20]. Land use data were obtained from the 2010 National Geomatics Center of China report [21]. Daily rainfall data were obtained from the Tanzania Production Company (TPC) rainfall stations located near Moshi.
Patient history, questionnaire, and MAT data were entered using the Cardiff Teleform system (Cardiff, Inc., Vista, CA, USA) into an Access database (Microsoft Corporation, Redmond, WA, USA). Geospatial data were managed using QGIS, version 2.8.3 (Free Software Foundation, Boston, MA, USA). Spatial scan statistics were calculated using a Bernoulli model to assess evidence of spatial clustering of cases using SatScan version 9.0 (www.satscan.org) [22]. All other analyses were performed using Stata, version 13.1 (StataCorp, College Station, TX, USA).
This study was conducted in accordance with the Declaration of Helsinki. It was approved by the KCMC Research Ethics Committee (#295), the Tanzania National Institute for Medical Research National Ethics Coordinating Committee (NIMR1HQ/R.8cNo1. 11/283), Duke University Medical Center Institutional Review Board (IRB#Pro00016134), and the University of Otago Human Ethics Committee (Health) (H15/055). Written informed consent was obtained from all participants or their guardians.
Of 15,305 patients admitted and 30,413 presenting to the outpatient department, 2,962 met eligibility criteria and 1,416 (47.8%) were enrolled. Of 1,293 participants who completed the risk factor questionnaire and had serum tested, 24 (1.9%) met the study criteria for acute leptospirosis, 252 (19.5%) were seropositive, and 592 (45.8%) were classified as controls (Fig 1). The remaining 449 (34.7%) were seronegative but provided only a single serum sample and so were excluded from analysis. The frequency with which participants were predominantly reactive to different serogroups is shown in Table 1. Participant characteristics are shown in Table 2. Clinicians did not diagnose leptospirosis in any study participant. Four (25.0%) of 16 leptospirosis cases with discharge diagnoses recorded were diagnosed with malaria despite negative blood parasite examinations.
Bivariable logistic regression of individual risk factors are included in S2 Table. There was a strong association between behaviors involving a single livestock species. For example having cleaned cattle waste was associated with having fed cattle with an OR 324.1 (95% confidence intervals 96.6–1087.0). There was some association between behaviors involving different livestock species. For example having cleaned cattle waste was associated with having cleaned goat waste with an OR 28.8, 95% confidence interval 12.0–69.1. There was a small magnitude association between rodent contact variables and livestock related variables. For example owning cattle was not associated with seeing rodents frequently in the house, compound or fields, and had a low magnitude association with seeing rodents in the kitchen or food store (OR 1.5, 95 confidence intervals 1.1–2.1).
Results for the logistic regression analysis of individual behaviors are shown in Table 3. On bivariable regression, variables associated with acute leptospirosis included working in rice fields (OR 14.6, 95% confidence intervals (CI) 2.9–59.5); cleaning up cattle waste (OR 4.3, CI 1.2–12.9); feeding cattle (OR 3.9, CI 1.3–10.3) and working as a farmer (OR 3.3, CI 1.3–8.2).
Nine (42.9%) of 21 experts (three livestock field officers, four veterinarians, and two zoonotic disease epidemiologists provided internally consistent multiple pairwise rankings of the relative exposure to livestock urine from the behaviors listed in Table 4. Four (100.0%) of four experts (one water engineer, one water and sanitation epidemiologist, and two zoonotic disease epidemiologists) provided consistent multiple pairwise rankings of the relative exposure to surface water. Three (75.0%) of four experts (one rodent ecologist, one veterinarian, and one zoonotic disease epidemiologist) provided consistent multiple pairwise rankings of the relative exposure to rodent urine. The individual behaviors evaluated for each exposure scale and the geometric means of the weights assigned to each are listed in Table 4. The results of pairwise comparisons, and calculated weights for each behavior are presented in S3 Table, S4 Table, and S5 Table. The distributions of participants’ exposure scores on each scale are shown in Fig 2. Overall, 534 (69.3%) of participants had no evidence of exposure to cattle urine, 563 (73.0%) had no exposure to goat urine, 241 (31.2%) had no exposure to rodent urine, and 262 (34.0%) had no exposure to surface water. There was limited correlation between cattle urine exposure and both goat urine exposure (r2 = 0.21) and pig urine exposure (r2 = 0.04). In addition there was little correlation between livestock urine exposure scores and rodent urine exposure (for example, cattle urine exposure and rodent urine exposure, r2 = 0.04), livestock exposure scores and surface water exposure (for example cattle urine and surface water (r2 = 0.02), and between rodent urine exposure and surface water exposure (r2 = 0.02). All exposure scales had a linear relationship with log odds of acute leptospirosis
Our bivariable logistic regression (Table 5) found that increasing exposure to cattle urine (OR 2.3, CI 1.1–4.7) and exposure to rodents (OR 1.7, CI 1.1–2.8) were both associated with increased odds of acute leptospirosis. In multivariable logistic regression (Table 5), no exposure scale was independently associated with leptospirosis. As shown in S6 Table, there were no significant interactions. The largest variance inflation factor was 1.33.
GPS co-ordinates were available for houses of 649 (84.2%) participants. No two or more participants lived at the same household. Land use designation could be determined from participant’s self-reported village of residence for an additional 79 (10.2%) participants. There was no evidence of clustering in the spatial distribution of cases. Results of the bivariable logistic regression analysis of geo-referenced variables and rainfall, and acute leptospirosis are shown in Table 6. There were no statistically significant associations.
Results of the logistic regression of individual risk factors for Leptospira seropositivity are listed in Table 7. Working in rice fields (OR 3.6, 95% CI 1.5–9.0); slaughtering goats (OR 2.3, 95% CI 1.0–4.8), working as a farmer (OR 1.8, 95% CI 1.3–2.5), and frequently seeing rodents in the kitchen (OR 1.5, 95% CI 1.1–2.1) were significant risk factors (p < 0.05) on bivariable regression. We fitted an initial multivariable model using the risk factors shown in Table 8. As shown in S6 Table, we did not identify any significant interactions between variables. In our final multivariable model, working as a farmer (OR 1.6, CI 1.1–2.3), working in the rice fields (OR 2.7 CI 1.0–7.2), or seeing rodents in the kitchen ≥ once per week (OR 1.5, CI 1.0–2.1) were all independent risk factors for Leptospira seropositivity. Walking barefoot (OR 0.7, CI 0.5–0.9) and owning dogs (OR 0.6, CI 0.4–1.0) were associated with reduced odds of Leptospira seropositivity.
The logistic regression models of the exposure scales and Leptospira seropositivity are shown in Table 9. Increasing exposure to rodent urine (OR1.2, CI 1.0–1.5) was associated with Leptospira seropositivity on bivariable logistic regression, but not on multivariable regression.
Results of the bivariable logistic regression analysis of rainfall and Leptospira seropositivity are shown in Table 10. There was an inverse association with mean annual rainfall >1,600mm per year (OR 0.56, 95% CI 0.33–0.93). We fitted an initial multivariable model using household elevation, mean annual rainfall, maximum daily rainfall in the preceding 30 days, and total rainfall in the preceding 30 days. The final model contained elevation (OR 0.99 per 10m, CI 0.98–1.0, p = 0.06), and total rainfall in the preceding 30 days (OR 1.2 per 100mm, CI 0.95–1.5, p = 0.13) but neither association was statistically significant. An analysis of the risk factors for seropositivity against Leptospira serogroup Icterohaemorrhagiae is included as S6 Table.
We identified multiple associations between exposure to cattle and acute leptospirosis, suggesting that cattle may be important sources of human leptospirosis in northern Tanzania. We also identified work in rice fields as an important risk factor for human leptospirosis. These findings must be interpreted with caution, as they were based on a small number of cases, and were present in only bivariable regression. Despite this, our findings have implications for the control and prevention of leptospirosis in Tanzania.
On bivariable regression, exposure to cattle was associated with acute human leptospirosis both when we evaluated individual behaviors and scales of cumulative exposure to cattle urine. These findings support other data from northern Tanzania that indicate that livestock may be an important source of human leptospirosis [31]. Among cattle slaughtered for meat in the Moshi area, 7.6% of cattle tested were carrying pathogenic Leptospira spp. in their kidneys [31]. Furthermore, seroreactivity against serogroups Australis and Sejroe, the two dominant serogroups among human cases in our study, was also frequently observed among cattle slaughtered for meat in the Moshi area in 2014 [12]. Our findings are also consistent with studies examining risk factors for Leptospira seropositivity in Africa. Leptospira seropositivity was common among abattoir workers in Kenya and Tanzania [11, 27]. In rural Uganda, livestock skinning was reported as a risk factor for seroreactivity and human seropositivity to livestock-associated Leptospira serovars was common [28]. In a global context, cattle have also been identified as a key risk factor in other rural livestock-farming communities in Central America and South Asia [14, 15], suggesting that strategies to reduce either livestock leptospirosis or transmission of leptospirosis from livestock to humans may be important global public health interventions.
Rodent exposure is an important risk factor for leptospirosis in the tropics, particularly in urban areas of Asia and South America [4, 29, 30]. In our study, an increasing score on the exposure to rodent urine scale was associated with acute leptospirosis in bivariable regression. However, the only individual component of the scale for which we found an association on bivariable regression was smallholder farming. Since smallholder farming may involve substantial exposure to both livestock and rodents, and other rodent related variables were not associated with leptospirosis the role of rodents in this association is uncertain. We also found that frequently sighting rodents in the kitchen or food store was associated with Leptospira seropositivity. Rodents could transmit leptospirosis to humans, or act as a reservoir that transmit Leptospira to livestock. However, recent work in the Kilimanjaro Region found no evidence of Leptospira urinary shedding, or renal infection among 393 wild rodents [31]
Although practiced by few participants, we found an association between working in rice fields, and both acute leptospirosis and Leptospira seropositivity. In some areas of northern Tanzania rice farming is practiced intensively, and there are active efforts to increase irrigated, continuously flooded rice farming across Tanzania [32]. In Asia rice farming is an established risk factor for leptospirosis. In Asia humans are infected through prolonged contact with water that may be contaminated by infected animal hosts [4, 29]. Further work is needed to evaluate possible sources of contamination of rice paddies in Tanzania and promote personal protective measures among rice farmers.
We did not find associations between acute leptospirosis and rainfall, or environmental risk factors around the home. The small number of cases available for analysis, and the relative lack of resolution of geo-referenced data meant that this result must be interpreted with caution. The lack of association with heavy rainfall differs from findings of studies from other locations [33, 34]. We found that seropositivity was associated with lower elevation and lower rainfall. While we did not have household level slope data, the topography of the study area includes steeply sloping terrain on the flanks of Mount Kilimanjaro that may not favor surface water accumulation. The lack of association between leptospirosis and home location may indicate that the workplace is an important site for infection [9, 11]. Future studies should collect data regarding workplace location.
Clinicians did not diagnose leptospirosis during the study period, and over-diagnosis of malaria was common. At the time of our study, there were no locally available, accurate diagnostic tests for leptospirosis. In addition, despite the high incidence in the region, clinician awareness of leptospirosis and other zoonotic diseases remains low [35]. This highlights the need for clinician education and evaluations in Africa of inexpensive point-of-care diagnostic tests.
We found that risk factors and the pattern of predominant reactive serogroups among leptospirosis cases was markedly different from those in seropositive individuals, for whom the febrile illness concurrent with enrollment was unlikely to be leptospirosis. In particular, reactivity to serogroup Icterohaemorrhagiae was common among seropositive participants, but there were few acute cases associated with this serogroup. These results may indicate that a serovar from the Icterohaemorrhagiae serogroup was circulating in this region [36], causing only mild disease not requiring tertiary medical care. Elsewhere, a difference in severity of disease has been linked to variability of infecting Leptospira species [37], Alternatively, the presence of Icterohaemorrhagiae seropositivity but absence of acute cases could indicate historic circulation of this serogroup that has since declined. Other results suggest that leptospirosis has a dynamic epidemiology in this area with the emergence and decline of specific serovars over time [3]. Cross reactivity between serogroups, and non-specific reactivity are other possible explanations [38].
Our study had several limitations. First, the prevalence of acute leptospirosis was lower than anticipated [8], potentially curtailing our ability to detect important associations. Conversely, associations of individual activities and leptospirosis identified by this study were sometimes based on only a few cases and should be interpreted with caution, especially given the multiple statistical tests. In addition, changes in leptospirosis incidence in the study area might also reflect changes in predominant sources and modes of transmission over time [3]. Second, the associations for acute leptospirosis were seen only on bivariable analysis, and these associations may be due to confounding from unobserved behaviors. Due to the complex interconnection between individual behaviours, we also consider that confounding may influence the multivariable logistic regression model of individual behaviours and Leptospira seropositivity. For example, the inverse association of walking barefoot and leptospirosis is puzzling, and we think it is likely to be influenced by an association with some protective factor, despite not identifying such an association among the behaviors we investigated. Diagnostic test limitations may have also introduced classification errors of participant cases or controls into our analysis. Leptospirosis is notoriously difficult to diagnose, particularly in the acute stages of illness and all currently available diagnostic tests for leptospirosis, including MAT [39], are imperfect. The sensitivity of MAT on paired serum samples is approximately 80% and the specificity close to 100% [40]. Specifically, not all participants with leptospirosis will seroconvert [40], and it is not possible to differentiate between historic and recent infection based on a single high titer [41]. We chose MAT for our case definitions since MAT on paired serum samples, while imperfect, remains the reference standard [40]. Furthermore, culture, nucleic acid amplification and point-of-care IgM serology lack sensitivity in our setting [12, 42, 43], and reports from other settings have been mixed [39, 44–46]. Our MAT panel comprising 20 serovars covered the major Leptospira serogroups that cause human disease, and all those within which African isolates are grouped [1]. We did not use locally isolated serovars and this may have influenced identification of cases. However, studies on the use of local isolates in MAT reference panels have shown that they do not necessarily perform better than other serovars from the same serogroup [47, 48]. Our analysis of acute leptospirosis was limited to cases across all serogroups. We acknowledge that risk factors may vary by infecting serovar, and pan-serogroup analyses may mask important associations.
We developed scales for use in our analyses for dimension reduction due to the unanticipated low number of cases. We suggest that cumulative exposure scales may have a future role in assessing sources of acute leptospirosis, as they allow assessment of cumulative exposure that may be important in assessing individual risk of disease. The analytic hierarchy process was an appropriate method of creating these scales, as it is an effective tool for quantifying multi-dimensional qualitative knowledge [24]. While we acknowledge that there is scope to improve our cumulative exposure scales, our scales that quantify expert opinion offer more biologically plausible groupings than statistical methods of dimension reduction. Key areas for future development of cumulative exposure scales are to validate them across multiple groups of experts, and to formally compare their effectiveness against purely statistical dimension reduction. Since our questionnaire sought exposures over a 30 day period, recall bias may have influenced our findings. Finally, we enrolled only 47.1% of eligible patients. We found no bias towards particular ethnic or occupational groups. However, we cannot rule out the possibility that the enrollment pattern influenced our results. Despite these limitations, the consistency of the association of the livestock related variables strengthens our confidence in the interpretation of their role in transmitting leptospirosis to people in our region.
Our results have implications for control of leptospirosis. Transmission of leptospirosis within rice fields, and from livestock to people is amenable to control through personal protective equipment for those performing high risk activities [49]. In addition, Leptospira vaccines are available for use in livestock against some Leptospira serovars. In some countries such vaccines have contributed to successful control of leptospirosis [49]. However, before a vaccination program is considered it is essential to understand reservoir structure and predominant infecting serovars.
Our study identifies associations between cattle contact and work in rice fields with acute leptospirosis. Our findings suggest that cattle may be a source of human leptospirosis in northern Tanzania. Further work is needed to determine if these findings are stable over time, and to investigate the link by isolating infecting serovars from humans and animal hosts. The development of local MAT capacity, or use of nucleic acid amplification or point-of-care IgM tests that have sufficiently high sensitivity would enable real-time diagnosis and allow testing of potential animal hosts living in proximity to humans with acute leptospirosis. Nonetheless, our findings suggest that control of Leptospira infection in livestock could play a role in preventing human leptospirosis in Africa.
|
10.1371/journal.pcbi.1000179 | Discarding Functional Residues from the Substitution Table Improves Predictions of Active Sites within Three-Dimensional Structures | Substitutions of individual amino acids in proteins may be under very different evolutionary restraints depending on their structural and functional roles. The Environment Specific Substitution Table (ESST) describes the pattern of substitutions in terms of amino acid location within elements of secondary structure, solvent accessibility, and the existence of hydrogen bonds between side chains and neighbouring amino acid residues. Clearly amino acids that have very different local environments in their functional state compared to those in the protein analysed will give rise to inconsistencies in the calculation of amino acid substitution tables. Here, we describe how the calculation of ESSTs can be improved by discarding the functional residues from the calculation of substitution tables. Four categories of functions are examined in this study: protein–protein interactions, protein–nucleic acid interactions, protein–ligand interactions, and catalytic activity of enzymes. Their contributions to residue conservation are measured and investigated. We test our new ESSTs using the program CRESCENDO, designed to predict functional residues by exploiting knowledge of amino acid substitutions, and compare the benchmark results with proteins whose functions have been defined experimentally. The new methodology increases the Z-score by 98% at the active site residues and finds 16% more active sites compared with the old ESST. We also find that discarding amino acids responsible for protein–protein interactions helps in the prediction of those residues although they are not as conserved as the residues of active sites. Our methodology can make the substitution tables better reflect and describe the substitution patterns of amino acids that are under structural restraints only.
| Identification of residues responsible for a specific function of a protein can provide clues about the mechanism of action. Computational approaches to identifying functional residues have emerged as low-cost alternatives to experimental methods by providing fast and large-scale analyses. Moreover, the demand for such approaches is increasing as more sequences become available from genome sequencing projects. Here, we focus on the use of CRESCENDO to identify functional residues in proteins of known structure by comparing the amino acid substitutions observed in a family of proteins with those predicted on the basis of the protein structure. CRESCENDO uses Environment Specific Substitution Tables, or ESSTs, which define the way that accepted amino acid substitutions are influenced by the local structural environment. We describe how the calculation of ESSTs can be improved by using only amino acids that are not involved in catalytic activity, metal or ligand binding, nucleic acid or protein interactions, and other molecular functions. Our new substitution table can better describe the degree of amino acids substitutions that are under structural restraints. It should be of value in all applications of ESSTs, including their use in sequence–structure homology recognition, structure validation, and structure prediction in addition to their use in the identification of functional residues. These approaches should enhance the understanding of protein structure and function, which is critically important in the postgenomic era.
| Proteins existing in living organisms have been selected through the process of evolution. However, much of the amino acid variation between orthologues appears to be selectively neutral [1] as far as the whole organism is concerned and accepted amino acid substitutions result in equal fitness. It has been long understood that the rate and nature of accepted mutation or substitution is different for the 20 amino acids in a protein [2]–[5]. Indeed the different substitution rates and patterns for the 20 amino acids were first quantified by Margaret Dayhoff as the PAM (Percentile Accepted Mutation) matrix in 1970s [2], which measures the point mutation for every 100 amino acids. The methodology was further developed by Henikoff et al. [3] to reflect more divergent relationships of protein sequences. The BLOSUM62 is now recognized as a standard measure of substitution rate for the 20 amino acids in the sequence comparisons. Jones et al. [4] introduced a fast and automated approach based on a maximum parsimony counting method and Whelan et al. [5] applied a maximum-likelihood method to estimate the rate for amino acid replacement. All these substitution models are based on the sequence alignments of closely related protein families.
Orthologous protein families (or superfamilies) are assumed to be diverged from a common ancestor by accepting mutations that are selectively neutral. The rate of evolution [1] is assumed to be constant over evolutionary time [6],[7] and so evolutionary distances can be measured by analysing the substitutions of amino acids. The degree of conservation and the nature of substitutions of amino acids will be under many evolutionary restraints. One of those is dependent on the need to retain the protein tertiary structure and usually expressed as a tendency to maintain the local structural environments of individual amino acids [8].
The Environment Specific Substitution Table (ESST) is a substitution table that considers structural restraints in the calculation of substitution patterns. Overington et al. [9],[10] first calculated ESSTs from a set of homologous protein families whose three-dimensional structures were available. The rationale behind ESSTs is that the acceptance of substitution of an amino acid in an orthologous family is subject to its local tertiary environment. The local structural environments of amino acids include (1) main-chain conformation and secondary structure, (2) solvent accessibility, and (3) hydrogen bonding between side-chain and main-chain. 64 ESSTs can be derived from a combination of structural features; four from secondary structures (α-helix, β-strand, coil and residue with positive φ main-chain torsion angle), two from solvent accessibility (accessible and inaccessible), and eight (23) from hydrogen bonds to main-chain carbonyl or amide or to another side-chain. These combinations of structural features restrict possible substitutions of an amino acid and give rise to distinct patterns of substitution.
The ESST was improved and updated by Shi et al. [11] in 2001 by the use of the following features: (1) a clustering scheme to correct sampling bias, (2) a smoothing procedure to correct data sparsity, (3) using only high resolution structures in the alignments as a source of substitution matrices and (4) reduction of the bias caused by non-structural restraints. The last feature was designed to separate functional restraints from structural restraints when generating ESSTs. Because ESSTs take into account only structural environments, substitutions where the amino acids are conserved for functional reasons should not be counted in the calculation of matrices. Shi et al. took two kinds of functional residues into account to eliminate non-structural restraints which may cause a bias in the ESST. They were (1) residues involved in domain-domain interactions and (2) those interacting with ligand. Such residues were masked in the alignment files and were not taken into account in the substitution counts. However, the masking appeared to have very little impact on the performance of FUGUE [11]. Chelliah et al. [12] further developed ESSTs by introducing functional restraints, particularly in enzymes, on amino acid substitutions as a new environment in addition to 64 structural environments. They measured the Euclidean distance between every amino acid and the known functional residues and compared the degree of conservation in terms of the proximity with the functional residues. Their ESST, known as the function-dependent ESST, showed improvements in sequence to structure homology recognition.
Compared with traditional substitution tables (PAM, BLOSUM) derived from sequence information only, ESSTs were shown to give more precise and discriminating measures of substitution probabilities [13]. ESSTs have been shown to be useful in applications to secondary structure prediction [13] and sequence-structure homology recognition [14],[15]. Recently, CRESCENDO [8] has been successful in prediction of functional residues by comparing the observed substitution patterns for amino acids which are under both functional and structural constrains with those that are predicted on the basis of structure alone.
Here we investigate the impacts of various functional restraints on the conservation of amino acids in three-dimensional structures. The functional residues are divided into four categories. They are residues involved in (1) protein–protein interaction, (2) protein–nucleic acid interaction, (3) protein–ligand interaction, and (4) catalytic reaction at enzyme active sites. Such residues will be under greater pressure to be conserved throughout the evolution process where they remain critically important to the activity of protein and thus the selective advantage of the organism. We measure the degree of functional residue conservation by masking the locations in the alignment file and then discarding them in the calculation of substitution probabilities. The substitution models are compared with the non-masking model which counts those functional residues in the calculation of substitution probabilities. We measure relative contributions of four categories of functional residues by making several masking tables in combinatorial fashion. We test our substitution models by performing computational experiments using CRESCENDO [8] which is a program predicting functional residues from known three-dimensional structures of proteins and which should be more sensitive to the accuracy of the predicted substitution tables than FUGUE [11]. We show that our new ESST can find 16% more functional residues compared with the ESST of Shi et al. [11] for the same test-set. The new ESST is different from previous ones in that we cover a broader range of protein families, we take into account more three-dimensional structures and we consider a wider variety of functional residues which may bias amino acid substitution patterns.
Four categories of functional residues are considered in this study (Table 1). The first category of functional residues comprises catalytic residues of enzyme active sites, which are strongly conserved in orthologous families and often across superfamilies. CSA [16] and “ACT_SITE” records in UniProt [17] were used. The Catalytic Site Atlas (CSA) is a database of enzyme active sites and catalytic residues of enzymes whose 3D structures are available. It provides two types of entries: (1) original hand-annotated entries derived from the primary literature and (2) entries homologous to one of the original entries by sequence similarity. We took into account only the hand curated entries for reasons of reliability. The second category comprised amino acids involved in protein–protein interactions. Data concerning protein interactions were retrieved from InterPare [18] which is a database for interacting interfaces between protein domains. InterPare uses SCOP [19] as a domain definition and detects interacting domain pairs if there are at least five pairs of residues which fall within 5 Å distance between two adjacent domains. Residues interacting with nucleic acids comprise the third category. BIPA (S. Lee, unpublished) and “DNA_BIND” records in UniProt were used for this category. BIPA is a database for protein–nucleic acid interactions, which defines the atomic interactions using a distance threshold of 5 Å for van der Waals contacts, and HBPLUS [20] default options for hydrogen bonds and water mediated hydrogen bonds. The final category comprises the ligand-binding residues. For this information, the following UniProt feature annotations were used: “BINDING”, “METAL”, “NP_BIND”, and “CA_BIND” (see Table 1 for details).
The data from InterPare, CSA and BIPA are based on three-dimensional structures of proteins. Hence, those functional residues can be easily identified and mapped into PDB entries using chain and residue numbers as unique identifiers. However, as the functional feature annotations from UniProt are based on sequence information, they are required to be mapped into their corresponding PDB entries. For this purpose, we developed a mapping protocol named “double-map” to align a sequence from UniProt with that of PDB at the residue level. This mapping protocol is critically important as we should find and mask the exact functional residues from the structural alignment. The detailed algorithm of double-map is described in Material and Methods.
The new Environment Specific Substitution Table (ESST) was built based on the alignments of three-dimensional structures of proteins which belong to the same protein family. We used PDB as a source for the three-dimensional structures of proteins and SCOP as the definition of protein families and domains. SCOP version 1.71, which was used in this study, classifies 3004 families and 75930 domains from 27599 PDB entries. For each SCOP family, domains were clustered with sequence identity of 80% or more, after pre-processing the structure data (see Materials and Methods for details). Within a cluster defined in this way, a structure having the best resolution was selected as a representative for the structure alignments. This process yielded 1187 SCOP families having 5833 domains from 4309 PDB entries. These final alignments, which are shown as “ALL” in the matrix type of Table 2, were used as a source for the calculation of substitution tables.
Table 2 shows 17 ESSTs and compares the numbers of structures and the functional residues masked from the alignments. There are four matrix types which differ in the alignment source; OLD, ENZ, NOENZ and ALL. “OLD” is based on the 177 HOMSTRAD families, from which the ESST of Shi et al. [11] was derived. “ENZ” is for the 221 enzyme-specific SCOP families whose members contain at least one “ACT_SITE” residue or CSA hand-curated entry. “NOENZ”, the opposite of “ENZ”, does not contain any “ACT_SITE” annotations or CSA entries at all. These two matrix types are prepared in order to assess the effect of alignment sources in the substitution patterns of amino acids. “ALL” is based on 1187 SCOP families described above. SCOP families that belong to ENZ and NOENZ are subsets of ALL type and do not overlap as they include different SCOP families. Each matrix type is further divided into several subtypes (A, B, C, and D) which differ in the masking sources of functional residues (see Table 1). This is to investigate the effect of a specific category of functional residues by comparing the differences in the substitution patterns. For example, the effect of masking enzyme active sites can be measured by calculating the difference between two matrices D and X, because X does not mask any functional residues whereas D masks only active site residues. We made random-masking models (R), in order to assess the value of masking models in benchmarking the new ESSTs. Our new ESSTs mask more functional residues than the ESST (J) of Shi et al., because our models take into account a broad range of structural families and functional residues. ESSTs and structure alignments in Table 2 are available from http://www-cryst.bioc.cam.ac.uk/ESST.
Our new ESSTs differ from those of Shi et al. [11] in the source of structure alignments and the categories (and the number) of functional residues removed from the alignments. The differences between 17 substitution tables were measured and investigated in terms of 1) the conservation probability of amino acids (PCONS) and 2) the distance (DIST) between ESSTs (see Materials and Methods). We first looked at the different sources of structure alignments to assess their effects on the amino acid conservation in the substitution table. For this purpose, the non-masking models (X) from four alignment sources (OLD, ENZ, NOENZ and ALL) were compared. Figure 1A plots the PCONS of 21 amino acids (PCONS in Table S1). The conservation probability in the figure is averaged over the diagonal entries (i.e. those amino acids which are not substituted) from 64 ESSTs for each model. The overall degree of conservation is 28.93, 29.10, 32.08, and 36.73% for NOENZ, ALL, ENZ and OLD respectively (see Table S1 for details). All the amino acids in OLD-type are more conserved than those of ALL-type. We are aware that the number of structures and families in the alignment may affect the PCONS. In addition, the definition of protein families and domains of HOMSTRAD is more stringent than those of SCOP. This will make the sequences less divergent and the alignments more conserved. The distance of substitution tables (Table S2) shows that NOENZ and ENZ are the most distant (507) among four tables and NOENZ and ALL are the closest. This is clear as NOENZ and ENZ do not share nay families but all the families in NOENZ belong to ALL. Figure 1A shows that amino acids R, K, H and S of ENZ-type are more conserved than those from NOENZ by 17, 14.2, 8.5 and 7%, respectively. However, C and W from ENZ are less conserved than those of NOENZ by 24% and 9%.
Figure 1B shows PCONS of amino acids from the same source of alignment (ENZ) but having different masking types (A, B, C and D), being compared with non-masking (X), random-masking (R) and ESST of Shi et al. (OLD-J). Overall, the differences of PCONS among the tables are less clear than the differences shown in Figure 1A. In addition, Table S2 shows that the distances (DIST) between tables of different masking types, but having the same alignment source, are smaller than the distances of tables from the different alignment sources. This explains why the variations of PCONS and DIST between tables are more affected by the source of alignments than the masking sources. However, the relationship between PCONS (or DIST) and the number of masking residues (%Mask) could be clearly understood by the Spearman's rank correlation between two (see Table 3). The more we mask functional residues (%Mask) from the alignments, the smaller PCONS gets and the greater the difference as measured by DIST between the substitution tables. We found that the correlation between PCONS and %Mask (−0.3) was not made more distinctive by removing residues involved in protein–protein interactions. A-type masks 13.4% and 16.9% many more residues than B-type in ENZ and ALL, respectively, where the discrepancies lie in the protein–protein interactions as B does not include InterPare as masking sources. However, the average PCONS of A is bigger than B, although A masks much more residues than B. This becomes much clearer on looking at the PCONS of A and D where the difference is in residues annotated as CSA and ACT_SITE. The PCONS of D is bigger than A, although D masks many fewer residues than A. The result shows that the residues involved in protein–protein (or domain-domain) interactions are not as conserved as residues responsible for the catalytic activity of enzymes. From PCONS of ENZ-D and ENZ-X (Table S1), which differ in active sites as masking source, we observe that active site residues J, D, H and E are most conserved throughout enzyme families, where H is the most abundant amino acid annotated as ACT_SITE or CSA followed by D, E, and J.
The performance of the new ESSTs was benchmarked by using CRESCENDO [8], which is a program for predicting functional residues given a three-dimensional structure. The rationale behind CRESCENDO is to distinguish functional restraints from structural restraints, both of which give rise to the conservation of amino acids in the evolutionary process. For example, amino acids in the core region of a protein are conserved or conservatively varied in order to maintain an appropriate structure (and ultimately function) whereas the catalytic triad of a protease, such as CYS-HIS-ASP, is conserved to maintain the functional properties of the enzyme family. CRESCENDO quantifies the degree of amino acid conservation by measuring (1) the observed value based on the alignment to which a queried protein sequence belongs and (2) the expected value calculated by using ESST. Note that the first value reflects both structural and functional restraints, whereas the latter only reflects the structural restraints because ESST, by definition, only takes structural environments into account. The overall difference between the two is converted into Z-score (or CRESCENDO score) which can represent extra restraints—probably functional—on the process of evolution. Hence, the more accurate the ESST, the less good the agreement between the probabilities of conservation observed and that predicted on the basis of the structure of the protein alone. CRESCENDO can be a good benchmarking tool for the evaluation of new ESSTs, because more functional residues are masked than the old ESST. In addition, we can identify relative contributions of four masking resources on the performance of ESSTs. The benchmarking was designed to investigate the following two questions. (1) How well can a new ESST identify functional residues compared with the ESST of Shi et al. which is used currently as the default by CRESCENDO? (2) If there is any improvement, what makes the improvement?
From 221 enzyme-specific SCOP families for ENZ in Table 2, one third (73 SCOP families) was selected as a test-set and the rest were used to make benchmarking-ESSTs for ENZ. The test-set consists of 339 SCOP domains having 81,410 residues in total. Out of 81,410 residues, 602 residues are active sites (ACT_SITE or CSA), 11,917 residues are annotated by InterPare, 194 residues for nucleic-acid interactions and 1,348 residues are involved with ligand interactions. They are the true functional residues that we are trying to predict using CRESCENDO in order to evaluate the performance of our new ESST. In our analysis we took only the first cluster as the predicted residues. The performance of our new ESST was compared with that of the old in terms of detecting functional residues. Note that, for both ENZ and ALL types, the 73 SCOP families in the test-set were removed from the original ESST. The benchmarking ESSTs were renamed as At, Bt, Ct, Dt, Rt, and Xt to distinguish them from the original new ESSTs which are A, B, C, D, R, and X, respectively. This is in order to make our benchmarking an unbiased blind test by removing sequences in the test-set which might affect the benchmarking results. In the case of OLD and NOENZ, the original masking types were used in the benchmarking process as they do not contain SCOP families in the test-sets. The test-sets and benchmark results are accessible from http://www-cryst.bioc.cam.ac.uk/ESST.
Table 4 shows the average Z-score of CRESCENDO for 602 active sites, 11,917 PPI residues, 194 residues for protein–nucleic acid interactions (PNI) and 1348 residues responsible for interaction with ligands (PLI) along with the P-values for the predicted residues. The P-value demonstrates that the Z-score of the predicted residues is different from the randomly selected residues with a 0.09 level of significance. In other words, we can say that the predicted residues of CRESCENDO are far from the random within 0.09 error rate. The Z-scores for all the residues (81,410) in the test-sets are compared with those of functional residues predicted by CRESCENDO. The average Z-score of all the residues is near zero, regardless of masking types, which means there are no differences between the probabilities of residue conservations observed in the alignments and those predicted by ESST. However, the Z-scores for 602 active sites range between 0.48 and 0.93 depending on the matrix types and the masking sources. This observation suggests there are extra restraints which make the active sites more conserved in families of homologous proteins. The Z-scores of 1,348 PLI (Protein–Ligand Interaction, see Table 4) residues also imply that they are under extra restraints other than structural reasons. On the other hand, the average Z-scores for PPI and PNI residues are much smaller than that of 602 active sites. This may suggest that residues at protein–protein interfaces are under less strong restraints than residues responsible for the catalytic activity. However, there is strong evidence that sub-regions in protein interfaces—so called hot spots—are energetically more important and may be under stronger restraints in evolution [21],[22].
In Table 5, the performance of 17 ESSTs is compared in terms of recognizing 602 active-site residues. SENS, SPEC and COV were measured using the ratios of TP (true positive), FP (false positive), FN (false negative) and TN (true negative) (see Material and Methods for the definitions). The Z-score and SENS are plotted together in Figure 2; they are highly correlated having 0.95 Spearman's rank correlation score (Table 3). As shown in Figure 2, the average Z-scores and SENS of non-masking (X) and random-masking (R) models are always less than those from masking-models (A, B, C, and D) within the same matrix type. This clearly shows that the position of masking is significant and discarding the substitution counts of functional residues from the substitution table can increase the performance of CRESCENDO by making ESST less dependent on the substitution patterns of the residues under functional restraints. This result is clearer from the rank correlation (0.45) between %Mask and SENS in Table 3. In addition, our new masking models (A, B, C and D) outperform the ESST of Shi et al. (J) and even the non-masking model (ENZ-X, NOENZ-X and ALL-X) outperform J (see Figure 2 and Table 5). This can be explained in terms of PCONS and SENS; the average PCONS is highest in the order of J, followed by ENZ-X, ALL-X and NOENZ-X, but the performance (SENS) is exactly the reverse order of PCONS. Figure 3A shows an example of predicting active sites of a SCOP domain d1evua4 (a domain in the A chain of PDB 1evu, [23]) which is a cysteine proteinase containing three active site residues annotated by UniProt. Three active site residues (CYS-314, HIS-373 and ASP-396) could be identified only by ALL-type ESSTs (ALL-B and ALL-C) which are highly ranked in Figure 2. This is probably because PCONS of ALL is lower than that of ENZ and OLD for the local environments of the three catalytic residues.
Table 6 shows the recognition performance for 11,917 PPI residues with the same measurements (TP, FP, FN, and TN) in Table 5. Four masking substitution tables of ALL-matrix could detect more PPI residues than that of Shi et al. (J), but not all tables in ENZ-matrix outperform J. Regardless of matrix types and masking types, the sensitivity (SENS) of detecting PPI residues is much lower than those for detecting active site residues. We think that this arises from the average Z-score for PPI residues (see Table 4) which is close to zero, suggesting less strong evidence for extra restraints. Figure 3B shows an example of predicting PPI residues of a SCOP domain d1i7kb_ (B chain of PDB 1i7k, [24]) which is a ubiquitin conjugating (UBC) enzyme containing 14 residues interfacing with the A chain. Using ALL-A, CRESCENDO predicted 12 residues of which five were correct PPI residues (true positive, coloured in pink in Figure 3B). Among the nine missing residues (orange), PRO-30, SER-87, TYR-91, GLU-120 and LYS-121 were highly accessible (more than 50 Å2) to solvent in the complex whereas five true positives had relatively small solvent accessible area (see Figure 3B for details). Thus, as expected, residues within the protein–protein interaction interface which are partially accessible are less conserved and more difficult to identify by CRESCENDO. Table S3 contains benchmark results for detecting residues interacting with nucleic acids and ligands. The sensitivity is better than the benchmarking results of recognizing PPI residues but still less than that of detecting active site residues. Figure 3C and 3D show examples of predicting residues interacting with nucleic-acids and ligands, respectively (see Figure 3 for details).
We found that the number of functional residues masked and discarded (%Mask) from the substitution table does not always guarantee the best performance (SENS) of ESST in detecting functional sites using CRESCENDO. The rank correlation between %Mask and SENS is 0.45 (see Table 3). Hence, it is very evident that masking-models outperform non-masking and the ESST of Shi et al. as described above. However the category of functional residues does matter and affects the performance. Figure 2 shows the performance of 17 ESSTs on the predictions of 602 active sites of the test-sets. Regardless of the alignment source, the performance (Z-score and SENS) of table B (no-PPI mask) is always better than table A (all mask), which means discarding PPI residues is not effective in the recognition performance of enzyme's active sites. In addition, OLD-B also outperforms OLD-J by 5% in the sensitivity, where the difference lies in the PPI residues as well. However, in the case of recognizing PPI residues, table A of ALL-matrix outperforms table B by 5.2% in terms of TP (Table 6). Interestingly, table C, which does not mask active sites, ranked as second highest and the performance of table D, which masks only active sites, is worse than the random-masking (R) substitution table (see Table 6). This result indicates that discarding PPI residues can increase the recognition performance of PPI residues but does not improve predictions of active sites of enzymes. This observation probably arises from the fact that the interfacial interactions differ in nature from those residues in catalytic sites and therefore masking of catalytic residues has little impact on those in interfaces.
We have shown that discarding functional residues from the calculation of the substitution table improves the detection of functional residues when the new substitution table is used with CRESCENDO. We considered four categories of functional residues in this study (Table 1) and found that functional residues can be best predicted when the relevant category is discarded from the calculation of the substitution table. Our new masking models outperformed non-masking, random masking and the old ESST (Shi et al., [11]) not only in terms of true positives but also sensitivity. However, as shown in Tables 5 and 6, false positives (FPs) and false negatives (FNs) are relatively high compared with the number of true positives (TPs). The reason for high FPs is expected to arise from the restricted definition of functional residues. As shown in Figure 3A, FPs, coloured in green, are clustered around the catalytic triad (CYS-HIS-ASP) of the cysteine protease shown here. Some of these residues will be important for the local architecture of the active site and may even be buried; the substitutions accepted at these positions will therefore be restrained. Others will be directly involved in binding and positioning the substrate for catalysis. We have previously shown that CRESCENDO identifies such residues in predicting the active site [8]. Furthermore we have shown that the degree of residue conservation is significantly higher the closer the residues are to the active site and that geometrical proximity to the known active sites can be considered to constitute a new environment of ESST [12]. A reason for some high FNs is that we took only the first cluster predicted by CRESCENDO into account as positive results in the benchmark analysis; however CRESCENDO is expected to predict all regions under functional restraints and occasionally those critical for protein interactions, allostery, metal binding, post-translational modification and so on will be as conserved and score as high or higher than the active site residues. In addition, the annotations of functional residues might not be complete, which makes both FPs and FNs relatively high.
Other than CRESCENDO, there are several computational approaches to detecting possible functional regions of a protein in a fast and low-cost manner. Among them, the Evolutionary Trace method (ET), introduced by Lichtarge et al. [25] in 1996, is widely used and very successful in identifying functional regions, for example of SH2, SH3, and DNA binding domains. ET differs from CRESCENDO in that it identifies conserved residues only on the protein surface and exploits the use of a phylogenetic tree to identify local patterns of conservation unique but distinct amongst different branches which constitute protein subfamilies. Hence, the performance of ET highly depends on the quality of a phylogenetic tree which is determined by a set of sequences to which a query protein belongs. If the set of sequences were recently diverged, the branch-specific conservation could not be detected because the substitutions were not accumulated enough to construct a reasonable phylogenetic tree. CRESCENDO does not explicitly use the phylogenetic tree (although it could well do so), but will also not work well if the degree of divergence is low. It will, however, gain from local conservation of buried residues in the active site, for example the threonine of the aspartic proteinase catalytic triad. It also gains from a careful definition of the expected substitution patterns in any local environment and for this the proper treatment of functional residues when deriving substitution tables is of critical importance.
New ESSTs were derived from the structure alignments of SCOP families [19]. Baton (D.F. Burke, unpublished, Table S4), which is a successor of COMPARER [26], was used as a structure alignment program. The domain boundary and classification scheme of protein families were adopted from SCOP 1.71 as of this writing. PDB [27] was used as a source for protein three-dimensional structures. SCOP class F, which contains membrane and cell surface proteins, was not included in the alignment process as their amino acids can be in environments which differ from those in the cytoplasm. Also, non-canonical SCOP classes, H, I, J, and K, which are coiled-coil proteins, low resolution protein structures, peptides, and designed proteins, respectively, were removed from the alignment sources.
To guarantee the best alignment quality, the following three filtering conditions were applied. (1) Filtering by resolution: NMR structures and structures having resolution worse than 2.5 Å were not included in the alignment procedures. (2) Filtering by sequence identity: For each SCOP family, protein domains were clustered by running CD-HIT [28] with sequence identity of 80% or more. Within a cluster, a protein structure having the best resolution was selected as the representative. This is to remove any bias arising from the majority sequences of proteins in a SCOP family. (3) Filtering by sequence length: Within a SCOP family, the average sequence length is maintained by removing any domains having sequence below (1−0.3)*mean-length and above (1+0.3)*mean-length. Single member SCOP families were removed as they can not provide multiple alignments for the substitution calculation.
To take advantage of UniProt annotations in terms of three-dimensional structures, we developed a mapping protocol, “double-map”, which aligns a sequence of UniProt with that of PDB at residue level. Three sequences are required for every PDB chain; 1) one from SEQRES record of a PDB file, 2) another from the residue (SEQ) in ATOM record of a PDB file, and 3) the third (SP) from the corresponding UniProt entry of a PDB chain. Double-map makes two alignments from the three sequences (so the name “double-map”). The first is an alignment between SEQ and SEQRES and the second is between SEQRES and SP. Using SEQRES as a reference, SP can be aligned with SEQ and the locations of UniProt residues can be mapped onto three-dimensional structures. Ideally, the alignment between SEQ and SP is enough to locate UniProt residues in PDB. However, residues in the sequence (SEQRES) can be absent and sometimes different from the coordinate section (SEQ) for various reasons (e.g., the position in space is undetermined) and this makes the direct alignment between SEQ and SP incomplete. Double-map uses two sequence alignment programs; EXONERATE [29] and BL2SEQ of NCBI blast package [30]. If EXONERATE fails to run for a short sequence around 10–15 amino acids, BL2SEQ succeeds to complete the alignment.
The program SUBST (http://www-cryst.bioc.cam.ac.uk/kenji/subst), written by Dr Kenji Mizuguchi (unpublished software, Table S4), was used in the calculation of substitution table. SUBST takes structural templates as inputs which can be generated by JOY [31], a program to identify the local structural environments of amino acids in the structure alignment files. The Euclidean distance between two ESSTs, X and Y, (DIST(X·Y)) was calculated as; , where and is the probability of amino acid j to be substituted by k from the ESST of X and Y under the structure environment of i. Note that there are 64 structure environments (4*2*8 from the secondary structures, solvent accessibility and H-bonds, respectively) and 21 amino acids (Cysteine and half-cysteine using one-letter code J and C, respectively).
CRESCENDO [8] was used to benchmark new ESSTs based on the predictions of four categories of functional residues: (1) catalytic residues of enzyme active sites, (2) residues involved in protein–protein interactions, (3) protein–nucleic acid interactions, and (4) protein–ligand interactions (see Table 1 for the source). The divergent score was used as it is more sensitive to the environments and it better discriminates functionally conserved residues from structurally conserved residues. The CRESCENDO scores (Z-score) were smoothed and contoured using Kin3Dcont [32]. CRESCENDO returns several clusters of predicted residues based on the size of grid points contoured using the Z-score. Residues only in the first cluster were used as the predicted residues of functional residues in the analysis. The details of the equation can be found in the original paper [8]. The P-value of the predicted residues is calculated using a one-tailed test under the standard normal distribution.
The performance ESSTs were assessed by measuring sensitivity (SENS), coverage (COV) and F-measure. These measurements were calculated based on the ratios derived from TP (true positives), FP (false positives), FN (false negatives), and TN (true negatives), which are defined as follow.
TP is the number of residues correctly predicted by CRESCENDO. If the residues predicted by CRESCENDO are the same as those annotated by the reference database, they are counted as being correct. FN is the number of real functional residues where CRESCENDO failed to predicted. FP is the number of false hits that CRESCENDO predicted as functional residues but not actually annotated by the references. TP, FP, FN, and TN are exclusively determined by the ESST used in CRESCENDO.
The Spearman's rank correlation (ρ) was calculated as follows; , where di is the difference between each rank of corresponding values and n is the number of pairs of values.
|
10.1371/journal.pntd.0001420 | Activation of the Innate Immune Response against DENV in Normal Non-Transformed Human Fibroblasts | When mosquitoes infected with DENV are feeding, the proboscis must traverse the epidermis several times (“probing”) before reaching a blood vessel in the dermis. During this process, the salivary glands release the virus, which is likely to interact first with cells of the various epidermal and dermal layers, cells which could be physiologically relevant to DENV infection and replication in humans. However, important questions are whether more abundant non-hematopoietic cells such as fibroblasts become infected, and whether they play any role in antiviral innate immunity in the very early stages of infection, or even if they might be used by DENV as primary replication cells.
Fibroblasts freshly released from healthy skin and infected 12 hours after their isolation show a positive signal for DENV. In addition, when primary skin fibroblast cultures were established and subsequently infected, we showed DENV-2 antigen-positive intracellular signal at 24 hours and 48 hours post-infection. Moreover, the fibroblasts showed productive infection in a conventional plaque assay. The skin fibroblasts infected with DENV-2 underwent potent signaling through both TLR3 and RIG- 1, but not Mda5, triggering up-regulation of IFNβ, TNFα, defensin 5 (HB5) and β defensin 2 (HβD2). In addition, DENV infected fibroblasts showed increased nuclear translocation of interferon (IFN) regulatory factor 3 (IRF3), but not interferon regulatory factor 7 (IRF7), when compared with mock-infected fibroblasts.
In this work, we demonstrated the high susceptibility to DENV infection by primary fibroblasts from normal human skin, both in situ and in vitro. Our results suggest that these cells may contribute to the pro-inflammatory and anti-viral microenvironment in the early stages of interaction with DENV-2. Furthermore, the data suggest that fibroblast may also be used as a primary site of DENV replication and provide viral particles that may contribute to subsequent viral dissemination.
| In this work, we demonstrate that that both human whole skin and freshly isolated skin fibroblasts are productively infected with Dengue virus (DENV). In addition, primary skin fibroblast cultures were established and subsequently infected with DENV-2; we showed in these cells the presence of the viral antigen NS3, and we found productive viral infection by a conventional plaque assay. Of note, the infectivity rate was almost the same in all the primary cultures analyzed from different donors. The skin fibroblasts infected with DENV-2 underwent signaling through both TLR3 and RIG-1, but not Mda5, triggering up-regulation of IFNβ, TNFα, defensin 5 (HB5) and β defensin 2 (HβD2). In addition, DENV infected fibroblasts showed increased nuclear translocation of interferon (IFN) regulatory factor 3 (IRF3), but not interferon regulatory factor 7 IRF7, when compared with mock-infected fibroblasts. Our data suggest that fibroblasts might even participate producing mediators involved in innate immunity that activate and contribute to the orchestration of the local innate responses. This work is the first evaluating primary skin fibroblast cultures obtained from different humans, assessing both their susceptibility to DENV infection as well as their ability to produce molecules crucial for innate immunity.
| Dengue virus (DENV) has become one of the most important arthropod-borne viral infections of humans, with approximately 100 million cases per year. The etiological agent is a positive-sense, single-stranded RNA virus that belongs to the Flaviviridae family, of which there are four antigenically related serotypes (DENV-1, DENV-2, DENV-3 and DENV-4) [1], [2].
Cumulative data have demonstrated that both the innate and adaptive immune responses participate in the control and pathogenesis of Dengue disease, which has a wide spectrum of clinical forms ranging from weak Dengue Fever (DF) to severe dengue disease, such as dengue shock syndrome and hemorrhagic Dengue (SSD/DHF). A rapid initiation of the innate host defense may be the critical limiting step in the infection because the Dengue virus must overcome all barriers mediated by innate immunity before arriving at the regional lymphoid tissues [3].
The symptoms generally appear 4–7 days afterward; at this point, when the adaptive immune response may be ongoing, the patients usually present for medical care. As a result, much work has been devoted to understanding the adaptive immune response to DENV, but not enough information has been raised about the very early steps of interaction with the host [4]–[6].
One of the most intriguing questions about dengue is the identity of the non-hematopoietic cells that may play a crucial role in the innate antiviral immune response to DENV in the early stages of infection. Given the fact that mosquitoes inoculate DENV into human skin while they are feeding, the potential target cells for dengue infection should be localized in the dermis and epidermis, which constitute the first level of defense [7].
The more abundant cells present in the inoculation site are keratinocytes and fibroblasts. These cells could act either as a limiting step or like a jumpstart of the replication cycle, depending on inoculation multiplicity, and intrinsic host variability. Early studies infecting human forearm dermal fibroblasts have suggested the participation of skin fibroblasts in the immune response against DENV [8].
Although dendritic cells and monocytes in the skin have been suggested as important targets of DENV infection, the number of these cells is considerable low compared with fibroblasts [9], [10].
The skin is not only a physical protective barrier; it also participates in the rapid initiation of innate host defenses that might represent a limiting step to DENV infection. In the skin, both the infiltrating cells (such as macrophages, neutrophils, dendritic cells and lymphocytes), and the resident cells, such as the keratinocytes and fibroblasts that are abundantly localized in the epithelia, participate in the production of various types of cytokines, establishing a pro-inflammatory microenvironment with antimicrobial activity against arthropod borne pathogens such as enveloped viruses [11]. Although some of these elements have been exhaustively reviewed by Nielsen et al. [12], little is known about the events that occur in the skin in the very early stages after mosquito feeding.
It is conceivable that DENV has evolved mechanisms to evade the early innate host responses, which are initiated by pattern recognition receptors (PRRs). Among the PRRs the family of membrane–bound Toll-like receptors (TLRs) [13] and the RIG-I like receptors (RLR); Retinoic acid-inducible gene (RIG-I) and the melanoma differentiation-associated gene 5 (Mda5) [14]. PRRs triggering lead to a signaling cascade that converges on the activation of latent transcription factors, such as interferon response factor 3 (IRF3), interferon response factor 7 (IRF7) and nuclear factor kB (NF-kB) activating transcription factor 2 (ATF-2)/c-Jun. The transcription factors undergo subsequent nuclear translocation and bind to the type I interferon's (IFNα/β) gene promoters. Then activated homo- and hetero-dimers of IRF3 and IRF7 bind to the IFN-stimulated response elements (ISRE) located in the promoter regions of various ISGs, including ISG54, ISG56 and ISG15, thereby enabling direct IFN-independent activation of these ISGs. The synthesis of type 1 IFNs, in turn leads to the development of an antiviral state in the surrounding cells and to the activation and modulation of the adaptive immune system [15]–[17].
Cumulative data have shown that DENV is sensed by both TLR3 and TLR7. At this respect Warke et al., demonstrated that plasmacitoid dendritic cells (pDCs) constitutively express both TLR7 and through IRF7 lead to IFNα/β production in response to DENV. By contrast, HUVEC cells or U937 cells that have internalized DENV produce IL-8 and IFNα/β after viral recognition through TLR3 [18]–[20].
Cytoplasmic molecules, such as RIG-I (which senses short double-stranded blunt-end 5′-triphosphate RNA) and MDA5 (which recognizes long dsRNA), have been shown to play an important role in sensing different flavivirus, including DENV. Recent experiments using single and double RIG-I/MDA5 knockout, fibroblast cell lines have shown that DENV triggers the interaction of both molecules with the IPS-1 molecule to activate IRF3, IkB kinase and phosphatidylinositol-3 kinase (PI3K) [21], [22]. In addition, Conceicaoi et al., showed that TLR3, TLR8, RIG-I and MDA5 mRNA are up-regulated, along with the type I interferon IFNβ and pro-inflammatory cytokines, in HepG2 cells infected with DENV [23].
The IFN response is one of the early mechanisms of host defense that contributes significantly to innate immunity. The IFN system includes cells that synthesize IFN in response to viral infection. The induction of IFNα or β is one of the early events that follows viral infection, and it is widely accepted as the most immediate and important antiviral host response to many viral infections [24].
Indeed, mice deficient in IFNα/β and IFNγ receptors are more susceptible to mortality following intraperitoneal inoculation of mouse-adapted DENV [25], [26]. A strong IFNα-mediated inhibitory effect on DENV replication was also observed when different cell types were treated with IFNβ prior to virus exposure [27].
Recently, antimicrobial peptides have been considered as a key element in the defense mechanisms of the skin. Cumulative data had shown that defensins modulate the immune response against both enveloped and non-enveloped virus, by inducing cytokine and chemokine production and inflammatory and immune cell activation [28], [29]. In this study, we demonstrated that skin fibroblasts freshly released from healthy human skin and subsequently infected with DENV showed positive viral antigen at 48 h post-infection.
Considering that fibroblasts are one of the most abundant dermal cells in the host during early contacts with this virus, primary skin fibroblast cultures were established to address the contribution of these cells in innate immunity. Our findings revealed that fibroblasts sense the DENV through TLR3 and RIGI which then signal to produce IFNβ, TNF, and β defensins. We believe these are previously unrecognized features of importance in the immunobiology of DENV infection.
The DENV-2 clinical isolate that we used has been described previously [5]. Mosquito C6/36 cells derived from Aedes albopictus were grown in MEM supplemented with 10% fetal bovine Serum (FBS) (Gibco Carlsbad, CA) at 34°C. Baby hamster kidney (BHK-21) cells were cultured at 37°C in the presence of 5% CO2 in MEM supplemented with 10% FBS, 1 IU penicillin/mL, 1 µg/mL streptomycin and 2.4 ng/mL of amphotericin B at final pH of 8. The virus stock was prepared by infecting a C6/36 cell monolayer in 75 cm2 tissue culture flasks at 75%–85% confluence. When the infected monolayer showed cytopathic effects, the cells and supernatant were homogenized and diluted in a 40% polyethylene glycol solution in 2 M NaCl (Sigma-Aldrich St. Louis, MO) and incubated at 4°C overnight. The suspension was centrifuged at 6000 rpm for 1 h. The virus was resuspended in 1/15 of the total volume with a glycine buffer (Tris 50 mM, Glycine 200 mM, NaCl 100 mM and EDTA 1 mM) and 1/30 of the total volume of FBS. The virus was homogenized, aliquoted and frozen at −70°C until use.
The virus was titrated by the standard plaque-forming assay technique using BHK-21 cells as described elsewhere. Briefly, ten-fold serial dilutions of virus stock in Hank's salt solution (Gibco Carlsbad, CA) were used to infect monolayers of BHK-21 cells in 24-well plates. After incubation at 37°C for 1 h, the infected cells were overlaid with MEM Eagle modified medium (Gibco Carlsbad, CA) with 3% carboxymethyl cellulose (Sigma-Aldrich St. Louis, MO). After 5 days, the resulting plaques were stained with naphthol blue-black solution to quantify the plaque forming units (PFUs).
Freshly prepared, non-cadaveric, healthy human skin was obtained at the Surgery Department of the Hospital General de Ticoman (the protocol was approved by the hospital ethical committee) from patients submitted to general surgery; the tissue was obtained from the wound area during the surgical procedure.
For the histology, the biopsies were washed extensively with Dulbecco containing penicillin (200 mg/mL) and streptomycin (200 U/ml) as described by Limon-Flores et al [5]. The tissue was then cut into pieces of approximately 1.0 cm2, the underlying fat was carefully removed and each piece was placed (epidermis side up) into individual wells in a 12-well culture plate (Costar corning, NY, USA). Skin explants were gently inoculated with 1×106 PFU of the virus in a total volume of 30 µL using a 30-gauge insulin syringe needle (HSW, Tuttlingen, Germany) in the upper side of the epidermis, causing penetration into the dermis without passing through the explants. Control explants were inoculated with either UV-inactivated DENV or with PBS for the manipulation control.
The inoculated explants were incubated in a 24-well plate for 24 and 48 h with DMEM containing 15% FBS and supplemented with antibiotic, antimycotic, non-essential amino acids and L-glutamine (Gibco Carlsbad, CA) in an incubator with a humid atmosphere and 5% CO2 at 37°C. At the end of the incubation, the skin was mounted in tissue freezing medium (Jung 0201 08926). The histological sections were obtained as described elsewhere [5].
The explants where cut into 10 µm frozen sections. The tissue sections were fixed with 4% paraformaldehyde (Sigma-Aldrich St. Louis, MO) for 1 h. After 4 washes with PBA, the sections were blocked with goat serum at 10% in PBA for 1 h. The primary and secondary antibodies were incubated for 1 h. After 5 washes between each antibody, the samples were mounted and analyzed using a confocal microscope (Olympus FXM).
The fat tissue was removed carefully from the normal skin explants with a scalpel. The tissue was then cut into pieces of approximately 1 mm2 and washed exhaustively with DPBS (Sigma-Aldrich St. Louis, MO) supplemented with an antibiotic solution (3 IU/mL of penicillin, 3 µg/mL of streptomycin and 7.2 ng/mL of amphotericin B). Afterward, the tissue was incubated with a collagenase/dispase cocktail at 37°C in agitation. The cells in suspension were collected at 12 h and 24 h post-treatment. All cells were split into two 24-well plates over a slide treated with poly L-Lysine. They were then incubated for 12 h at 37°C in a 5% CO2 atmosphere. The cells were infected with 1×106 PFU of the virus and the immunofluorescence was performed as described below. The cells were double-stained for fibroblasts (Sigma-Aldrich St. Louis, MO) and for the dengue envelope protein (Chemicon Millipore Billerica, MA). Mouse anti-human IgM-rhodamine (Jackson Pennsylvania, Phi), and goat anti mouse IgG2a-FITC (Southern Biotech Birmingham, AL) were used as the secondary antibodies.
The explants where cut into pieces of approximately 2 mm2, and the tissue were incubated for 4.5 hours in a cocktail of collagenase and dispase (0.5 U/mL and 4 U/mL, respectively). After this treatment, the epidermis was mechanically removed from the dermis and incubated separately in a 0.1% trypsin solution (Sigma-Aldrich St. Louis, MO) for 30 minutes. The cell suspension and tissue fragments were then resuspended in DMEM medium (Gibco Carlsbad, CA) containing 20% fetal calf serum (FCS) and 10 ng/mL epidermal growth factor (Gibco Carlsbad, CA) in a 25-cm2 flask. After incubation for 5 to 10 days, when adherent cells covered the plastic flask, the tissue was removed and the cells were maintained in DMEM medium (Gibco Carlsbad, CA) containing 15% FCS and 10 ng/mL epidermal growth factor (Gibco Carlsbad, CA). The culture was passaged until confluence was reached. In addition, surface fibroblast protein marker expression was checked for each established primary culture. The cultures were used between the passages 2–10.
Infection of the skin fibroblasts was analyzed by flow cytometry. Briefly, 2×105 primary human skin fibroblasts were infected with 5 PFU/cell of DENV-2. Six hours before harvesting, the cells were treated with brefeldin A (Sigma-Aldrich St. Louis, MO) to interrupt the vesicular traffic. After 3 washes, the cells were fixed and treated with a permeabilizing solution (Becton Dickinson, Franklin Lakes, N.J) for 45 minutes. The cells were analyzed at 6, 12, 24 and 48 hours post-infection. All the cells were treated for 30 minutes with a non-related immunoglobulin solution (10% goat serum in PBS) to block the Fc receptors. The primary antibodies were diluted in blocking solution (PBS-10% Goat serum) at the corresponding dilutions: E protein (Chemicon Millipore Billerica, MA), IFNβ (Santa Cruz CA, USA), APC-TNFα (Serotec Kidlington UK), human β defensin 2 (Santa Cruz CA, USA), human defensin 5 (Santa Cruz, CA, USA), TLR3 (ebiosciences San Diego, CA), RIG-I (Santa Cruz CA, USA), Mda5 (Santa Cruz CA, USA) and fibroblasts surface protein (Sigma-Aldrich St. Louis, MO). The non-conjugated antibodies were marked with the corresponding secondary antibodies: anti-mouse IgG H+L (Caltag Invitrogen Carlsbad CA), sheep IgG H+L (Invitrogen, Carlsbad CA), anti-rabbit IgG H+L (Invitrogen, Carlsbad CA), and anti- goat IgG H+L Zymed Invitrogen Carlsbad CA). For flow cytometry, the cells were resuspended in 0.01% EDTA in PBS. Data were collected using a FACS Calibur flow cytometer (Becton Dickinson, Franklin Lakes, N.J) and analyzed using FlowJo software (Tree Star, Inc.). All the results were statistically analyzed using t student test with graph pad Prism 5 software (CA, USA).
Total RNA was extracted at the indicated times from mock- or DENV-infected cells Poly I:C (Amershan USA) transfected cells (as positive control) using TRIZOL reagent (Gibco Carlsbad, CA), according to the manufacturer's instructions. The RNA concentration was measured by spectrophotometry, and the quality of the purified RNA was analyzed. Total RNA (2 µg) was used for DNase (Invitrogen, Carlsbad CA) digestion, and RT for the synthesis of the first strand of cDNA was performed using SUPERSCRIPT (Invitrogen, Carlsbad CA) in a final volume of 30 µl. Five percent of the first-strand reaction was used for the PCR analysis. The PCR amplification was performed using specific primers for RIG-I (5′ GCATATTGACTGGACGTGGCA-3′ and 5′ CAGTCATGGCTGCAGTCC TGTC-3′), TLR3 (5′-CCCTTGCCTCACTCCCC-3′ and 5′-CCTCTCCATTCCTGG CCT-3′), TLR7 (5′-CCTCAGCCACAACCAACTG-3′ and 5′-TTGTGTGCTCCTG GCCCC-3′) and GAPDH (5′-GACCCCTTCATTGACCTCAAC-3′ and 5′-GTCCATGCCCATCA CTGCCAC-3′). The PCR products were separated by 1% agarose gel electrophoresis. The assays were performed at least three times from different RNA preparations.
The cells were seeded on glass cover slips (6×104) (Bellco NJ USA). After 24 h, the culture medium was removed and monolayers were infected with DENV-2 active or UV-inactivated virus both at 5 PFU per cell; they were then incubated at 37°C and analyzed by immunofluorescence at different times. Briefly, the cells were fixed with 4% para-formaldehyde (Sigma-Aldrich St. Louis, MO) in PBS for 20 minutes at room temperature; the cells were then permeabilized with 0.1% Triton-×100 in PBS and blocked with 10% normal goat serum. The cell monolayer was treated for 60 minutes with primary antibody: E protein (Chemicon Millipore Billerica, MA), NS3 [30], NFkB p50 (Novus Biologicals), IRF3 (Santa Cruz CA, USA), IRF3 ser396 (Cell Signaling 4D46 4947S), IRF7 (Santa Cruz CA, USA) and fibroblasts surface protein (Sigma-Aldrich St. Louis, MO) followed by fluorochome-conjugated secondary antibody mouse IgG H+L (Caltag Invitrogen Carlsbad CA), rabbit (Invitrogen, Carlsbad CA), and goat (Zymed Invitrogen Carlsbad CA). An irrelevant isotype antibody that matched the monoclonal antibody was used as a negative control.
Finally the nucleus was labeled with DAPI (1 µg/mL) in PBS for 10 minutes and the slides were mounted with Vectashield (Vector). The images were captured using two different confocal microscopes (Leica SP2 and OLYMPUS FVX).
Primary skin fibroblasts were seeded and infected at 5 MOI as described previously. The supernatants were harvested at 12 h post-infection for IFNβ (Interferon source, NJ USA) and at 24 h for TNFα (Biolegend NJ USA). The cytokine levels were measured according to the manufacturer's instructions. Absorbance at 450 nm was measured using the ELISA reading equipment (Sunrise Tecan, Salzburg, Austria).
Apoptosis-induced DNA strand breaks were end-labeled with dUTP terminal deoxynucleotidyltransferase using a commercial kit, according to the manufacturer's instructions (In Situ Cell Death Detection kit, TMR red; Roche Indianapolis, IN) Briefly, the cells were fixed with para-formaldehyde [2% in phosphate-buffered saline (pH 7.4)] for 60 min at room temperature and permeabilized in 0.1% Triton X-100–0.1% sodium citrate for 2 min in an ice bath. The TUNEL reaction was performed using TMR red dUTP at 37°C for 60 min, and the labeling was analyzed by fluorescent microscopy. TUNEL assays with infected and mock-infected cells with different DENV 2 proteins were performed at 24 and 48 h post-infection. As a positive control, human primary cultured skin fibroblasts were exposed to 10 µg/mL DNase for 10 min at room temperature.
To investigate whether healthy human fibroblasts from the skin are infected in situ with DENV-2, we employed a model previously reported by our group using non-cadaveric fresh skin explants [5]. The tissue samples were either infected with 1×106 DENV-2 or mock-infected, and 24 hours later analyzed by immunofluorescence. The DENV-2-infected tissue showed cells with viral antigen (E protein in green) in the dermis area; these cells were likely dermal fibroblasts due to their morphology and location. By contrast, no expression of E protein was detected in skin explants incubated with the UV-inactivated virus (mock 48 h), as shown in Figure 1A.
To verify this result, healthy human skin was processed as described in Materials and Methods to obtain fresh skin cell suspensions. These cells were incubated for 12 hours before infection to allow the re-expression of cell receptors that might have been lost during the colagenase/dispase treatment. The Skin cell suspensions were then infected with 1×106 PFUs and analyzed 48 h post infection. Skin fibroblast infection was assessed by immune detection of E protein (green); the monoclonal antibody 1B10, which is specific for a fibroblast surface protein, was used to identify fibroblasts (red). Figure 1B clearly shows that the fibroblasts from healthy skin are permissive to DENV-2 infection; in contrast to mock-infected dermal suspensions, where only the fibroblast marker (red) was observed. Of note, 7% of infected cell suspension that were positive for fibroblast marker, also showed a positive labeling for DENV antigen E, Thus corroborating the above results in which we also detected viral antigens in the in situ zone where fibroblasts are located Figure 1C. These data warranted further investigation regarding fibroblasts, as these may be one of the cell types that first support dengue virus and other flavivirus infections in vivo after the bite of flavivirus-infected mosquitoes.
It is well known that established immortalized cell lines may have multiple mutations that can affect the permissiveness to viral infections [31], [32]. In addition, fibroblasts are one of the more abundant cell types in the skin where the mosquito introduces the DENV. Thus, primary skin fibroblast cultures should be an appropriate in vitro model to provide insights about the role of if these cells in skin.
We established 10 primary cultures of skin fibroblasts from 10 different donors who were free from any known infectious disease and especially had no dermatological disorders. Skin biopsies were obtained from these donors who consented to participate in this study. The characteristics of the surgery, age, and the permissiveness of DENV infection as well as general information about the donors are included in the Figure S1A.
Each cell culture was first characterized by flow cytometry, assessing the expression of fibroblasts cell surface protein marker with the mAb 1B10, (Figure S1B). This finding confirmed that the cultures of established skin fibroblasts consisted of a homogeneous population of cells expressing the protein marker recognized by 1B10 antibody. HMEC-1 cells (endothelial cells) and isotype control staining were used as the corresponding controls. All the experiments were carried out on fibroblasts with less than 10 passages, because according with our data after more than 15 passages the cells become highly permissive to DENV infection, indicating changes regarding the original cells. Our data show that infection of primary skin fibroblasts from passages 2–14 did not exhibit overt changes in the percentage of infection (data no shown).
Once the primary cultures were established and well characterized, they were infected with DENV-2 at MOI of 5. The cells were then harvested and the expression of E protein was analyzed by flow cytometry at different post-infection times. To address whether the origin of the subjects skin biopsy may have influenced the susceptibility of the fibroblasts to infection with DENV, ten different primary cultures obtained from ten different subjects were infected or mock-infected, and the data are shown in the Figure S1C. Figure 2A shows the FACS results of 4 representative infected primary cultures of dermal fibroblasts expressing the DENV viral E protein with data from 3 independent experiments. The graphics corresponding to the rest of the cultures are shown in the Figure S1C; according to the number of viral E protein antigen-positive cells, the range of infection fluctuates from 21.4% to 31.7% regardless of the donor source. No E protein expression was detected in the mock-infected cells (Figure 2A). To further investigate whether the DENV replicated in the primary fibroblasts, different cultures were infected or mock-infected and then analyzed by immunofluorescence using a mouse monoclonal antibody to the NS3 protein [30]. Figure 2B shows that only the fibroblasts inoculated with active DENV-2 showed positive staining for NS3. By contrast, no staining was observed in the mock-infected cells or the isotype control cells. Furthermore, a plaque assay was performed to demonstrate that the fibroblasts supported active DENV viral replication.
Previous works have shown variation in the susceptibility to infection among cell types, using different DENV-2 strains. While some cells are susceptible to infection by all DENV2 strains assayed, human foreskin fibroblasts showed differences in the susceptibility to the prototype viral Thai isolates and Nicaraguan strains, depending on the passage [33]. In order to address this point we additionally infected four different established skin fibroblasts with a reference type virus (DEN4 H241). There were no significant differences in the viral titers obtained from different fibroblast cultures, and each line showed similar viral replication kinetics regardless of the donor (Figure 2C).
The results presented here strongly support the notion that skin fibroblasts must be one of the primary cell types that can hold up initial dengue virus infections in situ, during or immediately after the bite of a DENV-2-infected mosquito.
Previous reports have demonstrated that the TLR3 and TLR7 molecules are involved in flavivirus recognition. These molecules are predominantly observed in the intracellular compartments. Meanwhile, RIG-I and MDA5 are cytoplasmic sensors involved in the recognition of RNA. Furthermore, the above mentioned molecules were demonstrated to have a role in the innate immunity against DENV-2 in a mouse embryonic fibroblast model [22]. However, the variation among virus strains/serotypes or cell types may suggest distinct cellular responses. Furthermore in this study, we addressed whether human primary skin fibroblasts sense DENV through different PRRs than have been reported for different cell lineage, including MEFs.
Semi-quantitative RT-PCR analysis was performed in DENV-infected fibroblasts (Figure 3A). The mRNA levels of both TLR3, and RIG I showed changes at 18 hr post-infection and densitometry analysis was performed on these data. The results suggest that both TLR3 and RIG signaling are required for activating of innate response and for establishing an antiviral state in skin fibroblasts in response to DENV-2 infection. To verify the RT-PCR results, a flow cytometry assay was performed in DENV infected primary skin fibroblasts. We confirmed RIG-I up-regulation in skin fibroblasts at 12 h post-infection with a maximum at 36 h post infection, whereas there were no significant changes in RIG, TLR3 and Mda5 expression in the mock-infected cells or the non-treated cells (Figure 3D, C, D). By contrast, changes in the infected cells regarding expression of TLR3 molecule were observed as early as 6 h post-infection, with the maximum level occurring at 12 hours post-infection. Subsequently, both RIG-I and TLR3 protein levels were stabilized, and no important changes were observed in the Mda5 molecule. Our results suggest that both RIG-I and TLR3 may act in concert to detect DENV RNA in primary skin fibroblasts cultures (Figure 3D and 3B).
In keeping with the other evidence presented in this article that a fresh skin fibroblast primary culture is highly permissive to active infection with DEN-2 virus, we evaluated the ability of DENV-2 to elicit a functional antiviral response. The induction of type I interferon (IFNα/β) is an early protective event that occurs within hours of viral infection and is widely accepted as one of the most immediate and important antiviral host response [34]; therefore, we investigated whether human skin fibroblasts produce IFNβ upon infection with DENV-2. A flow cytometry assay was performed, and Figure 4A shows primary skin fibroblasts from four different donors (D1, D2, D3 and D4); three independent experiments were performed with each culture and the results are presented as the mean fluorescence intensity (MFI). Figure 4A clearly shows differences among the donors. However, all primary cultures analyzed consistently showed that the infected fibroblasts expressed significant levels of IFNβ within 6–12 h of infection (Figure 4A); the expression decreased at 24 h in inverse proportion to the amount of virus. Indeed, in some of the primary cultures, the MFI of the IFNβ produced by infected fibroblasts (50 MFI) was higher than that of the IFNβ produced by poly I:C-stimulated cells (20 MFI), used as a positive control. These results contrast to the low levels observed in the uninfected and mock-infected cells.
Furthermore, we analyzed the levels of IFNβ in the culture supernatants of primary skin fibroblasts of four donors (D1–4) (Figure 4B), and we found that the cells of D2 and D4 donors produced high level of IFNβ (4800 pg/mL 3600 pg/mL) which partially corroborated the cytometry results. Although, all of the skin fibroblasts showed similar IFNβ expression kinetics, the total population was displaced in the histograms, showing a clear paracrine effect because not all the cells were infected with DENV. There were clear differences in the magnitude of the response, although the susceptibility to infection was almost the same in different fibroblast cultures.
TNFα has been shown to induce inflammation and apoptosis, thereby limiting viral infection, through a wide variety of mechanisms. Indeed, TNFα can have a direct effect on viral tropism by altering the expression of the cell surface receptors used by viruses; and depletion of this cytokine using TNFα blockers might facilitate the development or reactivation of the viral infection [35]–[37]. Thus, we evaluated the expression of TNFα by the fibroblasts at various times after infection. Figure 5A shows the results of two representative primary cell cultures analyzed by flow cytometry; the results are presented as MFI. The fibroblasts expressed significant levels of TNFα 24 h post infection. In contrast, cells treated with UV-inactivated DENV-2 (mock-infected) and uninfected cells did not show any changes. Similar trends were seen in all cell lines.
Antimicrobial peptides have been demonstrated to be an important early element of innate immunity. Emerging studies indicated that certain defensins can block viral infections. Thus, we decided to investigate whether human primary fibroblasts were able to produce defensins during DENV infection. The production of defensins was evaluated at different times post-infection. The infected cells were able to produce HβD2 as early as 12 h post infection, reaching maximum production at 24 h (Figure 5B). DENV-2 infection also induced the production of HD5 at 24 h; however, the MFI was about half of the corresponding value for HβD2 (Figure 5C).
The transcriptional activation of innate immunity molecules is dependent on the activation of a family of transcriptional factors from the NFkB and IRF families. Although IRF3 is ubiquitously expressed in most cell types, the expression of IRF7 is differentially regulated, depending upon both the pathogen and host cells. Because IFNβ was produced by skin fibroblasts in response to dengue infection, we decided to assess the regulatory mechanisms of this IFN induction. Our first step was to evaluate the role of the master transcriptional factors IRF3 and IRF7.
For this, human skin fibroblasts were infected with 5 MOI of DENV-2. Cellular localization of IRF3 and IRF7 was analyzed by immunofluorescence at different times post-infection. Figure 6A and 6B show that fibroblasts undergo nuclear accumulation of IRF3 following infection with DENV-2. While in mock and uninfected cells, IRF3 distribution was predominantly cytoplasmic. By contrast, DENV infection of skin fibroblasts failed to induce IRF7 nuclear translocation at 24 or 48 h post-infection (Figure 6E), the location of the molecule was mainly cytoplasmic. At 72 h, when some changes were detected, the cells were already damaged and highly granular (Figure 6D).
According with the data described above IRF-3 is likely activated since 12 h post-infection. However IRF3 has to be phosphorylated to dimerise before translocating into the nucleus. To asses this, immunofluorescence was performed again with an anti-IRF3 antibody to detect active associated phosphorylation (SER 396). Nuclear IRF3 was observed mainly in the infected cells as early as 6 h post-infection increasing at 24 h post-infection, contrasting with the mock-infected or the uninfected cells where no nuclear signal was observed (Figure 6C)
Activation of the NFkB transcription factors are one of the early steps during viral infections, the activation of the pathways involving the homo and hetero-dimmers of p50 have implications on induction of a pro-inflammatory set of genes and in the regulation of the inflammatory response [38]. To evaluate if the infection with DENV induces transcriptional activation of the pro-inflammatory transcription factor, we evaluated the localization of p50 by IF in infected cells. At 48 h post-infection, p50 was observed in the nucleus of both infected and in neighbor cells with no evident infection, suggesting once again a paracrine effect as consequence of the antiviral/pro inflammatory micro environment by the infected fibroblasts. Figure 6F.
Some viruses direct the apoptosis of infected cells. The induction of apoptosis may be a viral strategy to aid dissemination. Furthermore, apoptosis has recently been associated with the activation of innate immunity pathways in DENV disease via TRADD caspase activation [39]. To evaluate whether Dengue virus serotype 2 is able to induce apoptosis in human primary fibroblast cultures, we infected primary skin fibroblasts with 5 MOI of DENV and then analyzed them at different times. Twenty-four hours after infection, 34% of the 60% of skin primary fibroblasts that expressed viral NS3 protein, showed a TUNEL-positive signal as detected by immunofluorescence (Figure 7). The maximum infected cell count (68% of the antigen-positive cells) was reached at 48 h after infection. Mock-infected cells do not shown viral antigen (NS3).
Cumulative invaluable data has been generated regarding the adaptive immune response against Dengue, focusing in characterizing the most important changes during the disease, such as those involving antibodies, T cells and cytokines [40], [41]. However, insufficient information is available regarding the early events of the innate immune response when the virus is introduced through the skin, the natural route to acquire the infection.
Given that mosquitoes inoculate DENV-2 into human skin while they are feeding, some research have suggested that DENV initial replication should occur in the skin and that potential target cells for dengue infection should be the cutaneous DCs localized in the epidermal and the dermal layers. [5], [9], [42] Recent evidences suggest that Langerhans and/or dendritic cells may take up antigens for processing and presenting them to the adaptive immune system, instead of working as reservoir cells for the dengue virus [43], [44].
Clearly other non-hematopoyetical cutaneous cells may potentially be infected; fibroblasts are one of the most abundant cell types at the Dengue virus arriving site. In consequence these cells may be physiologically relevant for in vivo infection when inoculation occurs by any virus-bearing tick or mosquito.
In this article we have shown that fresh human fibroblasts obtained from the healthy skin explants are permissive to DENV infection. These data were generated after double staining with markers for fibroblasts and DENV. This experimental model was used to more closely approximate an in vivo infection. Based on these data, we established human primary dermal fibroblasts cultures with a reduced number of passages, and we clearly demonstrated productive infection in all of primary cultures. Thus, healthy human skin fibroblasts are likely to be involved in the development of innate immune responses to arthropod or thick-borne viruses rather than being only target cells whose primary role has been proposed to serve as replication depot.
However it is important to consider that cumulative work raised using a wide range of models, with different hosts, mosquito species and arthropod-borne viruses showing that, mosquito saliva and/or feeding is associated with a potentiation or control, affecting virus transmission, host susceptibility, viremia and disease progression impairing the antiviral immune response [45], [46].
In contrast in other experimental data, the expression of anti-dengue effectors molecules in the distal-lateral lobes of A. aegypti salivary glands has shown to reduce prevalence and mean intensities of viral infection [47]. Moreover, Thangamani et al. demonstrated important differences in the activation of immune response to CHIKV due the route of transmission showing an up-regulation of TL3 in both uninfected and CHIKV infected mosquito bites [48]. This data showed contrasting immune activation in response to CHIKV. However other studies have shown no effect on arbovirus infection due to mosquito transmission
In summary the studies investigating the effects of mosquito saliva upon the vertebrate immune response remain unclear and all together may suggest that at high concentrations some salivary proteins are immmunosuppressive, whereas lower concentrations of this modulate the immune response [49].
Understanding the potential role of mosquito saliva over the DENV infected fibroblasts is a point that must be considered in the future.
Double-stranded RNA is produced during DENV viral replication. This molecule can be detected by Toll-like receptors (TLRs) 3 and 7with in the endosomes, while the RNA helicases detect viral dsRNA in the cytoplasm. It has been described the possibility that DENV is detected by RIG-I, TLR3 and TLR7 (depending on the cell type) triggering the IFN response [16], [18], [19], [21]–[23]. These detection systems are not mutually exclusive and the expression of the PRRs depends on the cell lineage or origin.
Our data showed that in the primary skin fibroblasts, both TLR3 and RIG-1 (but not Mda5) molecules are up-regulated at different times suggesting that both molecules may trigger a coordinate induction of the antiviral response against DENV. TLR3 may “prime” the early response while RIG-I might regulate the amplified response at later time points. This concerted response has been observed in a model of MEFs infected with WNV, in which RIG-I mediated the initial detection of WNV infection in cells of non-immune origin. But RIG-I-deficient cells were still capable of responding to WNV, suggesting that other PRRs are also involved in mediating the innate antiviral response.
During DENV detection by skin fibroblasts, none of the established cultures showed mRNA for TLR7 after DENV infection. These data may suggest the absence of these PRRs in such cells. Similarly, Paladino et al. analyzed the TLR expression in virally infected fibroblasts of different origins and found absence of TLR7 [10].
The differences in the molecules involved in DENV detection suggest a differential cytokine profile that may contribute and maintain the microenvironment collaborating with dendritic cells (DCs) and possibly other resident cells when the virus arrives. The activation of the resident DCs depends on viral replication itself and also on the presence of important mediators such as IFNβ, TNFα and defensins, and they may be modulated by activation of other signaling pathways, such as the production of double-stranded RNA.
During viral infection various cytokines play a role both in viral clearance and in tissue damage mechanisms. The type I interferon's (IFNs) is regulated by a large family of multifunctional immunoregulatory proteins such as IRF3 and IRF7. IRF3 possesses a restricted DNA binding site specificity and interacts with CBP, while IRF-7 has a broader DNA binding specificity that contributes to its capacity to stimulate IFNα subtype expression. Both IRF3 and IRF7 play distinct and essential roles in the IFNα/β response to eliminate the viral infection. At the same time the activation of transcription factors during the infection depends on the signal given to the PRRs. Here we observed activation of IRF3, and we also observed that even if the secretion of IFNβ maintains the temporal kinetics, the magnitude of the response varies between donors. Moreover, studying skin samples from different donors, races and ages, would give valuable information of the innate immune system profile that may correlate with the susceptibilities of some hosts.
On other hand if the skin fibroblasts produces type I interferons this event becomes very important in terms of contribution to an antiviral state, in the infection site, more when considering that DENV is a strong inducer of type I IFN responses.
The production of defensins by the infected fibroblasts may increase the recruitment of immune cells such as neutrophils and monocytes during the DEN infection; otherwise these molecules may act directly causing virolisis.
Preliminary experiments in our laboratory evaluating the potential activity of human defensins over the viral particle suggests a direct activity against DENV (unpublished data).
RIG-I activation induces the association with IPS-1 (Cardiff, Visa or MAVS). This interaction induces the activation of the inflammatory/antiviral response. However, the exposure of the Caspase recruitment domains (CARD) of RIG-I and IPS-1 may also recruit TRADD, RIP-1 and FADD, a molecular complex known as the TRADDosome, which induces the activation of caspases 8 and 10 provoking cell death [39]. We detected cell death early during the infection with DENV (Figure 7) and an increment of the RIG-I protein (Figure 3D). The apoptosis activation may be caused by the activation of the innate immune system; however, more extensive studies are needed to test this hypothesis. These data corroborate findings previous findings by our group [50].
Clearly most studies addressing innate immunity in Dengue have been performed using human or mouse cell lines; however, not enough work has been performed by using primary human cultures of cells from the natural host and from the first anatomical site in contact with DENV when it enters through the natural route. These cells produce soluble mediators such as IFNs, TNFα and defensins. The magnitude and the variety of this response will depend on the amount of virus load inoculated and in consequence it might limit the infection in situ. Some studies suggest that the localization of the fibroblasts may confer different phenotype and gene expression. [51], [52] However we used healthy skin obtained from arm, abdomen and leg among others, and no differences in terms of permissively to DENV infection were detected.
Whether the mediators produced by dermal fibroblasts contribute mainly to pro-inflammatory anti-viral responses, or whether these cells might be used as in vivo reservoirs where an initial DENV replication occurs, is still uncertain and deserves further scrutiny.
As far as we know this is the first study where primary skin fibroblast cultures from different individuals have been evaluated for their susceptibility to DENV infection and for the elements of innate immunity that may contribute to DENV antiviral state. Moreover, this is the first report showing that defensins may be up-regulated by skin cells in response to DENV infection.
|
10.1371/journal.pbio.1001031 | Direct Observation of the Myosin Va Recovery Stroke That Contributes
to Unidirectional Stepping along Actin | Myosins are ATP-driven linear molecular motors that work as cellular force
generators, transporters, and force sensors. These functions are driven by
large-scale nucleotide-dependent conformational changes, termed
“strokes”; the “power stroke” is the force-generating
swinging of the myosin light chain–binding “neck” domain
relative to the motor domain “head” while bound to actin; the
“recovery stroke” is the necessary initial motion that primes, or
“cocks,” myosin while detached from actin. Myosin Va is a processive
dimer that steps unidirectionally along actin following a “hand over
hand” mechanism in which the trailing head detaches and steps forward
∼72 nm. Despite large rotational Brownian motion of the detached head about
a free joint adjoining the two necks, unidirectional stepping is achieved, in
part by the power stroke of the attached head that moves the joint forward.
However, the power stroke alone cannot fully account for preferential forward
site binding since the orientation and angle stability of the detached head,
which is determined by the properties of the recovery stroke, dictate actin
binding site accessibility. Here, we directly observe the recovery stroke
dynamics and fluctuations of myosin Va using a novel, transient caged
ATP-controlling system that maintains constant ATP levels through stepwise
UV-pulse sequences of varying intensity. We immobilized the neck of monomeric
myosin Va on a surface and observed real time motions of bead(s) attached
site-specifically to the head. ATP induces a transient swing of the neck to the
post-recovery stroke conformation, where it remains for ∼40 s, until ATP
hydrolysis products are released. Angle distributions indicate that the
post-recovery stroke conformation is stabilized by ≥5
kBT of energy. The high kinetic
and energetic stability of the post-recovery stroke conformation favors
preferential binding of the detached head to a forward site 72 nm away. Thus,
the recovery stroke contributes to unidirectional stepping of myosin Va.
| Myosin Va is a “two-legged” ATP-dependent linear molecular motor that
transports cellular organelles by “stepping” along actin filaments
in a processive manner analogous to human walking, the two “feet”
alternating between forward and backward positions. During stepping, the lifted
leg undergoes rotational Brownian movements around a free joint at the
leg–leg junction. Although these movements are random, the lifted foot
lands preferentially on forward sites and rarely steps backward. This
directional bias arises in part from the forward movement of the junction
bending the “ankle” of the attached leg. Here, we show that the
lifted foot also plays a role in the direction of stepping by controlling the
orientation of its actin-binding site (the “sole”), which dictates
the accessibility of potential stepping positions. We observed the ATP-dependent
foot orientation and its stabilizing on individual myosin Va molecules in real
time under an optical microscope; we show that the lifted foot of walking myosin
Va is oriented in a “toe-down” conformation so that binding to a
forward site on actin is preferred largely over backward or adjacent sites.
Thus, the great kinetic and energetic stability of the myosin Va lifted foot
conformation contributes to unidirectional stepping along actin filaments.
| Myosin is an ATP-driven linear molecular motor that produces force and unidirectional
movement along actin filaments. The “swinging lever arm” hypothesis
proposes that small nucleotide-dependent movements at the catalytic ATPase active
site are amplified by rotation of the myosin “lever arm,” or
“neck,” light chain–binding domain that extends from the motor
domain, or “head” [1],[2]. In the myosin chemomechanical cycle, the lever arm swing
that propels the myosin motor forward along actin is referred to as the “power
stroke” and is accepted as a general mechanism for myosin contractility. The
“recovery stroke” is the essential motion that primes, or
“cocks,” the lever arm in the pre-power stroke position while myosin is
detached from actin. These strokes are the basis for the physiological functions of
all characterized myosin motors.
Myosin Va is a cargo transporter in cells [3] that has two heads, each
connected to a long and relatively stiff neck [4] reinforced with six calmodulins
(Figure 1A). Myosin Va moves
processively along actin filament and takes unidirectional “steps” [5] in which it
alternately places its two heads in forward positions ∼72 nm away from a
previous binding site [6], analogous to human bipedal walking. A mechanism for
unidirectional stepping has been investigated and proposed as follows (Figure 1A). When a head is
detached off actin, the detached neck undergoes rotational Brownian fluctuations
around a free joint at the neck–neck junction [7],[8]. Although the fluctuations are
random [7],
the power stroke of the bound head [4],[9] tilts the neck via “lever action” and moves
the junction (i.e., the pivot point for the fluctuations) forward, thereby favoring
binding of the detached head to a forward site.
This mechanism explains how a detached head can access a forward site, but not why it
binds preferentially to a forward site 72 nm away as opposed to other accessible
sites as, for example, a site adjacent to the bound head. For a detached head to
bind actin, the actin-binding site of myosin must be properly oriented with respect
to the actin filament. Therefore, since the position of the neck–neck junction
relative to the actin filament is constrained by the bound neck, the orientation
(angle) and stability of the detached head relative to its neck (head–neck
angle) dictate the binding site along a filament. The detached head orientation is
determined by the recovery stroke that occurs after ATP-induced detachment from
actin. If the role of the recovery stroke were just to prime myosin, the
head–neck angle could fluctuate significantly. This could allow for the
unbound head to bind to a site near or adjacent to the bound head as well as to a
site 72 nm away with similar frequency. Such a distribution of the step size,
however, has never been observed in the absence of applied external load [5],[6],[10]. Therefore,
another mechanism must exist.
We anticipated that the recovery stroke plays a critical role in orientating the
unbound head so that binding to a ∼72-nm forward site occurs preferentially
[11],[12]. In addition, it
has been reported that myosin Va moves forward under ∼2 pN of backward load
[5],[10] which would bring
the junction back beyond the neutral position [13] or reverse the power stroke
[14], and
cancels the bias introduced by the attached head power stroke. The additional role
of the recovery stroke above can be another bias for forward stepping even in the
presence of the load. Thus, the properties of the recovery stroke are critical for
the myosin Va stepping mechanism.
Several recent structural and kinetic studies have demonstrated the existence and
implications of the myosin recovery stroke. High-resolution crystal structures of
muscle myosin II [15] identified different nucleotide-dependent head–neck
angles in the absence of actin; these are thought to correspond to pre- and
post-recovery stroke angles. Bulk Förster resonance energy transfer assays of
myosin II revealed two [16] or three [17] nucleotide-dependent (averaged) transient angle
distributions. In addition, electron microscopic analysis of myosin Va [18] showed two
different orientations (i.e., projection angles) of heads relative to the neck,
depending on the nucleotide in solution. These observations have contributed to a
general model in which ATP binding triggers the recovery stroke, and phosphate
(Pi) release after hydrolysis leads to
relaxation of the recovery stroke (i.e., generation of the power stroke). However,
the energetic and kinetic angle stability of the pre- and post-recovery stroke
conformations of myosin (Figure
1A) and the manner in which they contribute to actin binding specificity
during processive stepping of myosin Va remains unknown.
We present in this study, to the best of our knowledge, the first direct observations
of the myosin recovery stroke (angle change at head–neck junction) in real
time and at the single molecule level. We developed a novel light-induced
ATP-concentration controlling system and single motor molecule assay that enables
the direct observation of the nucleotide-dependent dynamics and fluctuations of the
myosin motor domain. Our observations and analysis indicate that the myosin Va motor
conformation adopted after the recovery stroke is kinetically and energetically
stable, which allows for the detached head to bind preferentially to a forward site
72 nm away, thereby providing the grounds for biased forward stepping of myosin Va
along actin filaments.
We constructed an optical microscope observation system (Figure 1B) to directly visualize in real time
the nucleotide-dependent swings (i.e., strokes) and fluctuations of the myosin
head–neck angle using an engineered monomeric (single-headed,
“S1-like”) myosin Va (Figure S1). We anticipated that a monomeric
myosin Va molecule with a 50-nm bead (gray) attached at its neck (configuration
depicted in Figure 1B) would
permit transient swinging of a 0.29-µm bead duplex (cyan) attached to the
distal head region. To determine how the head orientation, assayed from the bead
position of the 0.29-µm bead duplex, responds to ATP, we included 200
µM caged ATP and 1.7 mU µl−1 apyrase in the
solution, such that ultraviolet (UV) irradiation generated an ATP transient that
was rapidly removed (hydrolyzed to AMP) by the apyrase with a time constant of
2–3 s (Figures S2 and 2B). We imaged a duplex (or a larger aggregate) of beads, and
initiated a full-intensity (∼2 nW µm−2, defined as
100%) UV pulse for 0.1 s that yielded a peak ATP concentration
([ATP]peak) of ∼2 µM (Figure S2).
Approximately 0.1% of duplexes made a distinct angular (>30°
judged in real time) swing within several seconds of the UV flash. Such a low
frequency is not unexpected given the low probability of an unobstructed
configuration, as illustrated in Figure 1B (drawn to scale in Figure S4).
Myosin Va predominantly adopts two distinct conformations during an experiment: a
resting angle in the absence of ATP (i.e., before UV irradiation; cyan in Figures 1C, 3A, and S5) and a
metastable angle (yellow) accessible only after ATP generation (save rare
excursions driven by Brownian fluctuations), interpreted as the post-power
stroke and pre-power stroke conformations of myosin Va, respectively.
A large fraction (∼50%) of the beads that swung returned to the
original angle in less than 2 min, and the UV-induced transient swings could be
repeated multiple (>2) times (Figures 1C and 3A; Video S1). We monitored 15 such duplexes (i.e., myosin Va molecules)
and analyzed a total of 121 swing–return pairs as detailed below. A subset
(∼20%) made two return swings and then detached from the surface or
remained immobile. The remaining ∼30% did not return or did not
respond to the second UV flash.
Excursions to the post-recovery stroke “state” (see Figure 1C legend) are ATP (UV
flash)–dependent. Every UV irradiation lasting 0.1 s at 100%
intensity ([ATP]peak ∼ 2 µM) induced a bead swing
within a few seconds (0.78 s on average; 29 flashes in six molecules; e.g.,
Figure 1C). Shorter
and/or weaker irradiations yielded longer delays before a swing (Figure S5A and
S5B) or no bead swings. UV irradiation while the bead duplex was in
the post-recovery stroke state, in contrast, never induced a swing: 37 flashes
of 0.1- or 0.2-s duration at 100% intensity failed to induce bead
rotation in six molecules (Figure S5C). Thus, swings from the
pre-recovery stroke state are initiated and limited by ATP binding, and myosin
Va in the pre-recovery stroke state prior to a swinging event is free of bound
nucleotide.
To quantitate the ATP dependence of swings, we developed a new technique that
generates a nearly constant level of ATP in a chamber using caged ATP, which was
first applied to a biological system by Trentham and colleagues [19], and
evaluated the method using the rotary F1-ATPase (GT mutant) motor
[20]–[22] (Figure
S3A). Stepwise UV pulse sequences with pulse width modulation (Figure 2A) of varying
intensity (14%, 4.4%, and 0.7%) repeatedly generated
intensity-dependent rotations of a given F1-ATPase molecule (Figure 2B). Averaged traces of
rotations are smooth, indicating that the UV pulse sequences generate nearly
constant ATP levels in the sub-second time scale (Figure
S3B). Rotational rates in the presence of known [ATP] (Figure S3C)
yielded UV intensity–dependent ATP concentrations (Figure 2C). These calibrations for constant
ATP level allow us to analyze the kinetics of ATP-induced myosin Va
swinging.
In the myosin Va swing assay, we turned on the sequence at different UV
intensities (i.e., [ATP]) until a swing occurred (orange bars in Figures 3A and S5D). The
time before a swing was inversely proportional to the [ATP] (Figure 3B), yielding an
apparent ATP binding rate constant of 2.5×106
M−1s−1, comparable to the value of
1.7×106 M−1s−1 measured
in solution (Protocol S1). These qualitative measurements strongly suggest
observed swinging events are those of functional myosin motors.
Except for occasional, short reversals in the post-recovery stroke state (e.g.,
arrow heads in Figures 1C
and S6;
discussed below), the post-recovery stroke state is characterized by
exponentially distributed dwell times with an average of 40±4 s (standard
error) (Figure 4). Note that
the post-recovery stroke state is quite stable kinetically, particularly in
comparison to the stepping intervals of 60–80 ms at physiological
[ATP], during which Pi release is
accelerated by binding to actin [10],[23]. Our bulk, biochemical assays indicate that ATP is
rapidly (>100 s−1; [23]) hydrolyzed into ADP and
inorganic Pi is released with a rate constant of
∼0.02 s−1 (τ = ∼50 s) (Figure S7).
Measurements with a shorter-neck, 1IQ construct [23],[24] indicated a
Pi release rate constant of ∼0.02
s−1 and a subsequent ADP release rate of ∼1.2
s−1. Collectively, these measurements indicate that myosin
Va in the post-recovery stroke conformation has ADP and
Pi bound in its active site and that
Pi release limits the return swing (i.e.,
power stroke off actin), consistent with bulk Förster resonance energy
transfer assays with myosin II in solution [16] and electron microscopy of
myosin Va bound to actin [25].
The angular fluctuations in both the pre-recovery stroke and post-recovery stroke
states are well fitted to Gaussian distributions (Figures 5A and S8), with a
peak separation yielding an average swing amplitude (θswing) of
85°±19° (standard deviation [s.d.] for 15 molecules).
The measured amplitudes reflect projections in the image plane, and thus the
actual amplitudes will differ if out-of-plane swinging occurred. However, the
observation that the appearance of most (∼2/3) of the bead amplitudes is
independent of the swing angle (e.g., Figure 5B), as confirmed by the constancy of
the axial ratio (Figure 5C),
indicates that the recorded swings used in the analysis were in a near
horizontal plane. Electron micrographs of myosin Va without actin show
comparable (∼90°) nucleotide-dependent angular changes [18], thereby
strengthening the interpretation that transitions between pre-recovery stroke
and post-recovery stroke conformations of myosin Va are being observed.
Both the pre-recovery and post-recovery stroke conformations display considerable
conformational flexibility, as indicated by the standard deviation of the
angular fluctuations (Figures
5A and S8). The magnitudes of fluctuations in both
states are comparable, with the Gaussian width (s.d., σ) averaging
24°±10° (s.d. for 15 molecules;
σpre/θswing = 0.29±0.09)
for the pre-recovery stroke conformation and 26°±9°
(σpost/θswing = 0.31±0.10)
for the post-recovery stroke conformation. These observed fluctuations include
contributions from flexibility in the myosin–bead junctions as well as
experimental image noise, so they represent an upper limit, with actual angle
fluctuations being smaller.
The Gaussian width of the thermally driven fluctuations (σ) measured here,
with the equipartition principle [26],(1)where
kBT (4.1 pN•nm) is thermal
energy and k is the myosin Va head–neck joint stiffness
(spring constant), allows us to determine the spring constants,
kpre = 23
pN•nm•rad−2, and
kpost = 20
pN•nm•rad−2.
With this spring constant, the energy required for bending of the head in the
post-recovery stroke conformation to the pre-recovery stroke conformation (i.e.,
the energy needed to bend the spring by θswing) expressed in
terms of the elastic potential energy
(E),(2)is
5.2 kBT. The post-recovery stroke
conformation is stabilized at least to this extent: because the experimental
σ in equation 1 includes the fluctuations of other components described
above, the spring constant k for the head–neck junction
must be underestimated, and thus the energy difference, E, of
5.2 kBT between pre- and
post-recovery stroke conformations is a lower limit.
There were occasions where we observed momentary swings back to the pre-recovery
stroke angle in the post-recovery stroke state (e.g., arrow heads in Figures 1C and S6). These
are unlikely to be purely Brownian excursions, because the bead tended to remain
at the pre-recovery stroke angle for a second or longer. A natural return
followed by immediate ATP binding that would induce a second swing is also
unlikely, because [ATP] must be negligibly low and these momentary
swings happened irrespective of the time after UV irradiation. The observed
momentary returns may represent reversal of the reaction responsible for the
swing to post-recovery stroke conformation, ATP hydrolysis [23], or subsequent myosin
isomerization [16]. We note that the return frequency in the absence of
drag from the attached beads could possibly be higher.
The natural assumption is that the detached head accessing a forward site in the
post-recovery stroke conformation will have its actin binding site properly
oriented for productive binding to actin (Figure 1A). Conversely, when the detached
head goes back to the post-recovery stroke conformation, the actin binding site
is predicted to be oriented incorrectly, thereby precluding actin binding (Figure 6A). The kinetic
stability of the post-recovery stroke state observed here indicates that this
proper head orientation is maintained for ∼40 s, much longer than the
stepping intervals. Even if the head in the post-recovery stroke state
accidentally touches a backward site at a moment when the head adopts a near
pre-recovery angle by fluctuation or momentary reversal, the binding should be
unstable by at least by 5 kBT
compared to forward binding. Thus, the kinetic and energetic stabilities of the
post-recovery stroke state together ensure forward binding of an unbound
head.
Momentary binding of a head with incorrect orientation will be unstable from
intramolecular strain [11],[12]. Consistently, a quantitative model has shown that
the lever arm (neck) elasticity and its strain influence the position of the
next binding site on actin, therefore the detached head preferentially binds to
the forward site [27]. This model assumes that the unbound neck with bound
ADP–Pi rigidly takes post-recovery
stroke conformation, which we report here.
The key for directional movement is to bias the completely random Brownian
rotations of a detached neck toward forward binding. The power stroke and its
angle stability of the attached rear head contribute approximately half of the
bias by moving the pivot for the Brownian rotation of the unbound neck forward,
which allows the detached head to access positions 36- to 72–nm distant on
an actin filament [4],[7] (Figure
1A). The remaining bias between positions ∼36 nm and ∼72 nm
from the detached site is provided by the recovery stroke and its stability.
Under a high backward external load, the power stroke would fail to produce a
bias: owing to the compliance in the neck and/or neck–head junction [13],[14], the
neck–neck junction would be pulled back to the neutral position,
immediately above the bound head (Figure 6B). Even under this circumstance, the bias by recovery
stroke still works, favoring forward binding. Therefore, for ∼72-nm discrete
unidirectional steps of myosin Va, the recovery stroke and its angle stability
of the detached head contributes to the bias, in addition to the power stroke
and its angle stability of the attached head. This mechanism may contribute to
transport cargos in a cell since some cellular components could be obstacles to
hinder the movement of cargo at times.
An alternative mechanism has recently been proposed for myosin VI [28], which is
thought to function as a force sensor as well as a transporter [29]: stable
lead head binding is facilitated by a backward load on the head, and hence
internal strain between the two necks promotes forward binding of an unbound
head. Myosin VI is the only reverse motor known to date, moving in the direction
opposite to all other myosins studied so far. It is of interest to study whether
the other myosins, including myosin Va, also adopt a similar, strain-dependent
binding for forward bias.
The stability of the post-recovery stroke conformation would also be important
for muscle myosin II, which can produce tension without contraction (isometric
tension) by repeatedly “scratching” actin. Forward binding is
required for efficient force production, but the base of the necks does not move
in this situation, and thus myosin II may rely entirely on the head orientation
being stabilized in the post-recovery stroke state. Other linear motors may also
rely on an effective swing to the post-recovery stroke conformation [11],[12].
To study ligand-dependent motion of molecules, caged nucleotides (uncaged by UV
irradiation) have been combined with microscopic observations. UV pulse
irradiation allows one to trigger motion of the molecular motor and to clearly
show its nucleotide dependence [30],[31], and modulation of UV irradiation time allows one to
control motor velocity and total movement [32]. This assay design has the
advantage over conventional flow/mixing assays in that solution conditions
(e.g., nucleotide concentration) can be altered rapidly and with minimal
perturbation. However, caged nucleotide measurements have been limited to
kinetic analysis because the concentration of uncaged compound can change
significantly during the course of an experiment, particularly if consumed by
the system being examined (i.e., diffusion, enzyme–substrate interaction,
or apyrase). We have developed a new technique to keep ATP level constant in
which the concentration and time evolution can be modulated by light intensity
and irradiation time (Figures
2, 3A, S3, and
S5D).
This method for visualization of a nucleotide-linked conformational change in a
motor protein under the controlled delivery of ATP should be generally
applicable to ligand-induced conformational changes of macromolecules.
Monomeric Gallus gallus myosin Va truncated at Leu-909
(containing all six IQ motifs) with an N-terminal myc tag (EQKLISEEDL)
positioned directly replacing Met-1 and a C-terminal FLAG tag (DYKDDDDK) with a
single glycine linker (Figure S1) was co-expressed with Lc-1sa in
Sf9 cells and purified by FLAG affinity chromatography in the presence of excess
calmodulin as previously described [24],[33]. The calmodulins on the
expressed protein were exchanged for 6× his-tagged calmodulin, expressed
in Escherichia coli, as previously reported [34] and
modified [7]: the his-tagged calmodulin and monomeric myosin Va at
the molar ratio of 6∶1 were mixed and incubated for 10 min on ice in 20 mM
imidazole-HCl (pH 7.6), 4 mM MgCl2, 100 mM KCl, 0.04 mM EGTA,
0.5% (v/v) β-mercaptoethanol, and 400 µM CaCl2. The
reaction was terminated by the addition of 4 mM EGTA followed by >20 min
incubation on ice. Monomeric myosin Va carrying his-tagged calmodulin was mixed
with an anti-his monoclonal antibody (Clontech Laboratories) at the
antibody:myosin molar ratio of 17∶1 in buffer A (25 mM imidazole-HCl
[pH 7.6], 4 mM MgCl2, 100 mM KCl, 1 mM EGTA, 5 mM DTT), and
incubated at room temperature for >5 min to allow binding.
A flow chamber, in all experiments under a microscope, was made of two coverslips
separated by two spacers of ∼100-µm thickness, and, after the last
infusion, the chamber was sealed with silicone grease or nail liquid. The
following infusions (2–3 chamber volumes), all in buffer A, were made with
1–2 min of incubation in between: 2 mg ml−1
unphosphorylated α-casein for surface blocking, buffer A for washing,
5.6% (w/v) 0.05-µm silica beads (Polysciences), buffer A for
washing, monomeric myosin Va (10 nM) complexed with anti-his antibody (for
binding to the silica beads through the antibody) or myosin Va alone without the
calmodulin exchange (for direct binding), 2 mg ml−1
unphosphorylated α-casein, 25 µg ml−1 biotinylated
anti-myc monoclonal antibody (Millipore), and buffer A for washing. Finally,
0.29-µm streptavidin-coated beads (Seradyn), washed three times by
centrifugation in buffer A, were infused together with 200 µM caged ATP
(Dojindo), 1.7 mU µl−1 apyrase (Sigma), 1.1 mg
ml−1 unphosphorylated α-casein, and 0.5% (v/v)
β-mercaptoethanol.
The purpose of the anti-his antibody was to let it serve as a cushion between the
myosin neck and a silica bead so as to keep the myosin intact. Direct binding,
though, worked as well, and some results, e.g., in Figures 1C and S5A–S5C, were obtained with direct binding. In both cases,
most of the 0.29-µm beads on the surface were bound to the head of myosin
Va through a biotin–avidin linkage, because the bead density decreased
significantly without myosin, with non-biotinylated anti-myc antibody instead of
the biotinylated one, or by mixing excess biotin with the streptavidin-coated
beads before infusion. When we infused short actin filaments instead of the
0.29-µm beads, they attached (presumably) to myosin Va, and a flash of
100% UV light for 0.2 s released >97% of them from the surface
within a few seconds.
We used an Olympus IX70 microscope equipped with a 100× objective
(UPLSAPO100× O IR, N.A. 1.4, Olympus), a stable sample stage (KS-O,
ChuukoushaSeisakujo), a dual-view system [35] for simultaneous
observation of fluorescence and bright-field images [36], a regular epi-fluorescence
port, and an additional UV excitation port consisting of a mercury lamp, an
extension tube (IX2N-FL-1) that forms an intermediate (conjugate) image plane
outside the microscope body, and a computer-controlled shutter with 5-ms
open–close time (Uniblitz). Fluorescence of Alexa 488 was excited at
475–490 nm, and images at 500–535 nm were captured with an
intensified (VS4-1845, Video Scope) CCD camera (CCD-300-RCX, Dage-MTI).
Bright-field images (650–730 nm) were recorded with another CCD camera. UV
excitation (300–400 nm) for uncaging ATP was confined in a circle of
diameter ∼90 µm at the image plane. A mask was placed on the conjugate
plane in the extension tube such that the central ∼30-µm square in the
image plane did not receive UV light. The swing assay was always made near the
center of the masked area to protect myosin from possible UV damage, although we
found that direct UV irradiation at the maximum intensity (see below) for tens
of seconds did not affect the motile activity of myosin Va. The rotation speed
of F1-ATPase (for estimation of ATP concentration; Figures 2 and S3) did not
depend on the position in, and even outside, the masked area, and short actin
filaments bound to myosin Va were released by a UV flash with indistinguishable
kinetics at all positions. Note that oblique UV beams illuminated the solution
above the masked area except for the immediate vicinity of the coverslip
surface. To record correlation of events and UV irradiation, a part of the UV
beam was recorded with the intensified CCD camera above, or with the camera for
bright field at an edge of the image. The UV power was measured above the
objective lens, and the estimated intensity in the image plane was ∼2 nW
µm−2 for unattenuated (maximal) excitation (defined
as 100% intensity). Observations were made at 23 °C.
The orientation of a bead duplex was determined as previously reported [7]. When
another bead came nearby, the orientation was judged by eye or abandoned.
Ellipticity of a bead image was estimated as the ratio of the long axis length
to the short one, calculated from the second moments of the intensity
distribution as
<Ix2>1/2/<Iy2>1/2
where x and y are pixel coordinates measured
along the long and short axes and with the origin at the image centroid,
I is the pixel intensity minus a threshold value, and
<> denotes averaging.
UV-generated ATP concentrations were estimated by both gliding bead assay for
native myosin Va and rotational assay for F1-ATPase (Protocol
S1).
ATP binding rate and Pi release rate of myosin Va
were measured using stopped flow apparatus (Protocol
S1).
|
10.1371/journal.ppat.1005555 | FleA Expression in Aspergillus fumigatus Is Recognized by Fucosylated Structures on Mucins and Macrophages to Prevent Lung Infection | The immune mechanisms that recognize inhaled Aspergillus fumigatus conidia to promote their elimination from the lungs are incompletely understood. FleA is a lectin expressed by Aspergillus fumigatus that has twelve binding sites for fucosylated structures that are abundant in the glycan coats of multiple plant and animal proteins. The role of FleA is unknown: it could bind fucose in decomposed plant matter to allow Aspergillus fumigatus to thrive in soil, or it may be a virulence factor that binds fucose in lung glycoproteins to cause Aspergillus fumigatus pneumonia. Our studies show that FleA protein and Aspergillus fumigatus conidia bind avidly to purified lung mucin glycoproteins in a fucose-dependent manner. In addition, FleA binds strongly to macrophage cell surface proteins, and macrophages bind and phagocytose fleA-deficient (∆fleA) conidia much less efficiently than wild type (WT) conidia. Furthermore, a potent fucopyranoside glycomimetic inhibitor of FleA inhibits binding and phagocytosis of WT conidia by macrophages, confirming the specific role of fucose binding in macrophage recognition of WT conidia. Finally, mice infected with ΔfleA conidia had more severe pneumonia and invasive aspergillosis than mice infected with WT conidia. These findings demonstrate that FleA is not a virulence factor for Aspergillus fumigatus. Instead, host recognition of FleA is a critical step in mechanisms of mucin binding, mucociliary clearance, and macrophage killing that prevent Aspergillus fumigatus pneumonia.
| Inhaled Aspergillus fumigatus conidia are effectively eliminated from the lung by the coordinated actions of mucociliary clearance and macrophage killing, but the mechanisms of attachment of Aspergillus fumigatus (A. fumigatus) conidia to the airway mucus gel are unknown. In addition, the mechanisms of phagocytosis of conidia by macrophages are incompletely understood, because inhibition of Dectin-1, mannose receptor, and TLR-2/4 does not completely prevent phagocytosis. A fucose-binding lectin (FleA) expressed on the surface of Aspergillus conidia has recently been described, but its function is unknown. In order to reveal FleA’s function, we carried out combined in vitro and in vivo studies using several novel reagents, including recombinant FleA, FleA deficient (∆fleA) conidia and a potent fucopyranoside inhibitor of FleA. In vitro studies found that FleA mediates binding of A. fumigatus conidia to airway mucins and phagocytosis of conidia by lung macrophages. In in vivo studies we found that mice infected with ∆fleA conidia develop invasive aspergillosis whereas those exposed to WT conidia do not. Based on our findings, we propose a novel host defense mechanism against A. fumigatus in which FleA expression on conidia is recognized by lung mucins and macrophages to promote mucociliary clearance and macrophage killing and protect from invasive pulmonary aspergillosis.
| Aspergillus fumigatus (A. fumigatus) is an ubiquitous opportunistic pathogen that causes invasive and often fatal lung infection, particularly in immunocompromised patients [1]. Aspergillus fumigatus produces small hydrophobic conidia that are easily inhaled into the lungs and require robust host defense mechanisms to prevent infection. The mechanisms of clearance of conidia from the lung are incompletely understood but phagocytosis by macrophages is known to be important [2–5]. Macrophages express Dectin-1, a C-type lectin that recognizes β-1-3 glucan on the surface of A. fumigatus conidia. Although the amount of surface accessible β-1-3 glucan is low on resting conidia, it is much higher in swollen conidia that appear early during germination and infection [6]. Binding of β-glucan by Dectin-1 promotes macrophage killing of A. fumigatus conidia, and other macrophage receptors, such as the mannose receptor and toll-like receptors (TLR) -2 and -4, cooperate in this killing effect [7, 8]. Notably, however, the phagocytosis of A. fumigatus conidia by macrophages is incompletely blocked by inhibitors of Dectin-1, mannose receptor, and TLR-2/4 [9], which means that macrophages must employ additional mechanisms to phagocytose and kill A. fumigatus.
Many microorganisms use lectins as adhesins to interact with host glycoproteins. For example, Pseudomonas, Burkholderia, Ralstonia, and Chromobacterium bacteria all express adhesins that include galactophilic and fucophilic lectins that initiate bacterial adherence to glycan receptors on host cells and tissues [10–14]. In addition, fungi such as Aleuria aurantia, A. oryzae, and A. fumigatus all express fucophilic lectins [15–18] with multiple fucose binding sites [11, 19]. For example, A. fumigatus lectin (AFL, also known as FleA) exists as a dimer with 12 fucose-binding sites available for strong multivalent interactions with fucosylated structures [18]. As a result, FleA has unusually high binding affinity for fucosylated structures [18, 20], but its role in fungal biology is unknown. Fucosylated glycans are abundant in plants and animals, and FleA expression by fungi may help them bind plant or animal tissues, as has been found for fucose binding lectins in bacteria [11, 21]. Aspergillus fumigatus conidia enter the human host via inhalation, and fucosylated proteins in the lung that could bind conidial FleA include mucins in the airway mucus gel and multiple glycoproteins in the glycocalyx of macrophages [22–24]. Fucose in different linkages is a common carbohydrate structure in gel-forming mucins [22] such as MUC5AC and MUC5B [23], and recent studies in transgenic mice have revealed the essential role of gel-forming mucins in host defense against lung infection [25]. In addition, membrane-tethered mucins (such as MUC1 and MUC4) function as receptors in epithelial cells [26] and macrophages [24] and it is possible that they may function as adhesins for A. fumigatus. To determine the role of FleA in the pathogenesis of A. fumigatus pneumonia, we studied the behavior of recombinant FleA and fleA-deficient conidia in multiple complimentary functional assays, including mucin binding assays, macrophage binding assays, and a mouse model of A. fumigatus pneumonia.
We generated a phylogenetic tree to illustrate the fungal genera and species that contain a putative FleA ortholog. The tree shows that FleA is not widespread in the Kingdom Fungi and is found in several pathogenic species of fungi. FleA is present in the genomes of only a few Aspergillus species, including A. fumigatus, A. flavus, A. oryzae and A. calidoustus (S1 Fig). FleA is also present in other human pathogens including dermatophytes (Trichophyton, Microsporum and Arthroderma), entomopathogenic fungi (Metarhizium, Ophiocordyceps), a nematode pathogen (Arthrobotrys oligospora), several plant pathogens (Penicillium expansum, Marssonina, Magnaporthiopsis, Ceratocystis, Gaumannomyces, Rhizoctonia) and Trichoderma spp. (pathogenic on other fungi).
To determine if A. fumigatus conidia can bind to mucin glycans in a FleA-dependent manner, we coated microtiter plates with mucins that we purified from the induced sputum of healthy subjects, and we used labeled recombinant FleA to quantify FleA-mucin binding. We found that FleA binds strongly to airway mucin and that this binding is inhibited by fucose (Fig 1A). No inhibition was observed with galactose (Fig 1A). To examine the specificity of FleA for the different linkage forms of fucose found naturally on mucins, we synthesized disaccharides with fucose linked to glucose in α1,2, α1,3, α1,4, or α1,6 positions. These fucose-glucose compounds all strongly inhibited FleA binding to mucin in a dose dependent manner (Fig 1B).
To further characterize fleA in A. fumigatus and examine the functional role of FleA binding to mucin, we created reagents to allow us to make quantitative measurements of A. fumigatus binding to mucin and to dissect the specific binding role of FleA. To facilitate visualization of the conidia, we created strains of A. fumigatus expressing nuclear GFP by transformation of AF293.1 and AF293.6 strains with the plasmid pJMP51 resulting in histone 2A fused GFP prototrophic (TJMP131.5) and auxotrophic (TGJF5.3) strains, respectively. TGJF5.3 was further modified and transformed to prototrophy using a disruption cassette targeting disruption of the fleA locus (Fig 2A) and multiple ΔfleA deletion transformants were identified. Confirmation of fleA deletion was demonstrated by Southern and northern blotting (Fig 2B and 2C). Despite repeated attempts to complement the ΔfleA deletion strains, we encountered problems such as phenotypic abnormalities in spore size and shape, which precluded generation of reliable complement reagents. Hence the mucin binding studies described below were repeated with several ΔfleA deletions to confirm the role of FleA as an A. fumigatus fucose binding protein. The mucin binding studies were also repeated using a novel synthetic glycomimetic inhibitor of FleA as an alternative experimental approach. To assess localization of FleA in conidia, we made a C-terminal RFP tagged fleA mutant in A. fumigatus by modifying the native fleA locus using a disruption cassette similar to that used to delete the fleA ORF (Fig 2D and 2E).
Multiple ΔfleA deletion A. flavus strains were generated using the same methods described for A. fumigatus above and as illustrated in S2 Fig.
Protein levels of FleA on A. fumigatus conidia and hyphae have not been characterized. We found that RFP-tagged FleA is present on A. fumigatus conidia when fleA-rfp is expressed under its endogenous promoter (Fig 3A), and we used two different strains of FleA-RFP tagged conidia to quantify FleA levels over time by fluorescence microscopy (Fig 3B). FleA fluorescence was high on resting conidia (with some variation among conidia), low in swollen conidia, and largely absent in hyphae (Fig 3B and 3C). Time-lapse microscope images of germinating conidia show that FleA begins to decrease by 21 hours and is almost entirely absent by 27 hours (Fig 3C and S1 Movie). These results were confirmed in extracts from resting and swollen conidia and from hyphae that were analyzed for FleA protein by western blot (Fig 3D). We also investigated the possibility that FleA was secreted by conidia, even though FleA lacks a canonical secretion signal peptide according to the Signal P 4.0 server [27]. Analysis of proteins in concentrated culture supernatant of A. fumigatus at various developmental timepoints revealed that resting conidia shed significant amounts of FleA, whereas conditioned media from swollen conidia and hyphal culture supernatant did not have detectable FleA (Fig 3D).
To examine if binding of A. fumigatus conidia to airway mucins is FleA dependent, we first developed a binding assay to determine the specific binding role of FleA. For the conidia-binding assay, we coated mucins onto chamber slides and used confocal microscopy to image binding of conidia. As shown in Fig 4A and 4B, the binding of three independent ΔfleA deletion mutant strains (TGJF 6.7, 6.8 and 6.13) to mucin is much weaker than the binding of WT conidia. These data demonstrate that A. fumigatus conidia bind to mucin in a FleA-dependent manner. After confirming that all 3 ΔfleA mutant strains showed identical patterns in mucin binding assays, we chose the A. fumigatus strain TGJF6.7 for all further functional studies.
To determine if FleA is also required for binding of A. flavus conidia to mucin, we examined the binding of WT and ΔfleA A. flavus conidia in our chamber slide assay. As shown in Fig 4C, the binding of three independent ΔfleA deletion mutant strains of A. flavus (TFYL 62.1, 62.2 and 62.3) to mucin is much weaker than the binding of WT conidia.
Fucosylated glycoproteins are not restricted to mucin glycoproteins. Macrophages also express multiple fucosylated proteins on their cell surface, including membrane-tethered mucins [24]. We therefore explored whether fleA expression by A. fumigatus might mediate binding or phagocytosis by macrophages. We first explored binding of recombinant FleA to macrophages by flow cytometry and showed that FleA binds very strongly to the surface of RAW264.7 mouse macrophages and primary human alveolar macrophages in a fucose-dependent manner (Fig 5A and 5B). Next, we investigated phagocytosis of A. fumigatus conidia to RAW264.7 mouse macrophages. Using confocal microscopy, we found that A. fumigatus WT conidia bound to RAW264.7 cells and that many are internalized/phagocytosed (Fig 5C). Notably, the internalization of ΔfleA conidia was 50% less than that of WT (Fig 5C and 5D). We repeated these experiments in alveolar macrophages isolated from bronchoalveolar lavage from healthy human subjects. Similar to data from the RAW264.7 cells, we found that A. fumigatus conidia are internalized and phagocytosed by human macrophages and that ΔfleA conidia show a reduction in phagocytosis of 40% compared to WT (Fig 5E and 5F). Thus, fucosylated structures on lung macrophages act as receptors for A. fumigatus FleA, resulting in enhanced phagocytosis of conidia.
To rule out the possibility that deletion of FleA causes changes to β-glucan in the fungal cell wall that could be affecting binding and phagocytosis, we used FACS to examine the binding of a β-glucan ligand (Dectin-1) to ΔfleA and WT conidia. We found no significant differences in the affinity of recombinant biotinylated Dectin-1 for ΔfleA and WT conidia in these experiments (Fig 5G). We also coated fluorescent microspheres with FleA to create a simplified model of conidia and found that FleA causes a significant and fucose-dependent increase in binding and phagocytosis of microspheres by RAW264.7 cells (Fig 5H).
To provide additional evidence that binding of A. fumigatus conidia to mucins and alveolar macrophages is FleA-dependent, we synthesized a library of modified fucopyranoside structures using methods we previously described for inhibitors of the FimH lectin in E. coli [28] (Fig 6). To screen for the relative potency of different fucopyranosides, we used the FleA-mucin binding assay. In this way, we found that (2E)-hexenyl α-L-fucopyranoside (2EHex) (Fig 6A) inhibits FleA with marked (nanomolar) potency (Fig 6B). In contrast to its potent inhibition of FleA binding to mucin, 2EHex does not show potent inhibition of PA-IIL (a fucose-binding lectin from Pseudomonas aeruginosa) binding to mucin (Fig 6C). We next tested the effects of 2EHex and fucose on interactions between A. fumigatus conidia and mucins and macrophages. We found that both 2EHex and fucose decreased the mucin binding of WT A. fumigatus conidia to the levels observed with ΔfleA conidia (Fig 6D). 2EHex also decreased the phagocytosis of WT A. fumigatus conidia by RAW264.7 cells and primary human macrophages to the levels observed with the ΔfleA conidia (Fig 6E and 6F). These inhibition studies confirm that binding of A. fumigatus conidia to mucins and macrophages requires the fucose binding activity of FleA.
Based on our data that FleA is required for binding and phagocytosis of conidia by macrophages, we hypothesized that ΔfleA conidia might evade phagocyte killing leading to increased lung infection compared to WT. To test this hypothesis, we infected immunocompetent C57BL/6 mice intranasally with WT conidia or ΔfleA conidia. H&E staining of infected lung tissue showed that mice infected with WT conidia had well-contained pneumonia (Fig 7A) whereas ΔfleA conidia treated animals had a poorly contained pneumonia (Fig 7B). GMS staining showed limited numbers of conidia with little evidence of germination in sections from WT treated animals (Fig 7C and 7E). In contrast, a large number of conidia from ΔfleA treated animals showed hyphae generation (germlings) evident on the GMS stain, typical of invasive aspergillosis (Fig 7D and 7F). Both the total number and percentage of germinating conidia are significantly higher in mice infected with ΔfleA conidia than WT conidia (Fig 7G and 7H). Lung injury is more severe in ΔfleA-infected mice, as evidenced by a higher concentration of hemoglobin in bronchoalveolar lavage (BAL)(Fig 7I). Aspergillus 18S gene expression in lung homogenates from ΔfleA-infected mice is higher than in lung homogenates from WT infected mice, indicative of higher fungal burden (Fig 7J).
BAL cells from mice infected with WT and ΔfleA conidia were analyzed by flow cytometry to quantify multiple immune cells and by multiplex immunoassay to quantify multiple cytokines, chemokines and growth factors. Mouse BAL cells were labeled with a panel of antibodies including CD11c, F4/80, CD11b, MHCII, Ly6G, Ly6C, NK1.1, TCRβ, B220, CD4 and CD8. Compared to mice infected with WT conidia, we found that several cell types were decreased in BAL from mice infected with ΔfleA conidia, including alveolar macrophages, neutrophils, NK and NKT cells and CD4 and γΔ T cells (Fig 8). No significant differences were observed for eosinophils, dendritic cells, B cells, monocytes and inflammatory monocytes and CD8 T cells. Compared to mice infected with WT conidia, we found that the concentrations of multiple cytokines, chemokines and growth factors were similar in BAL from mice infected with ΔfleA conidia, including IL-6, IFNγ, KC and VEGF in BAL fluid (S1 Table).
We have discovered that FleA produced by A. fumigatus conidia allows airway mucins to bind conidia and macrophages to effectively phagocytose them. Notably, when we engineer conidia that lack FleA, the resultant ΔfleA conidia show increased virulence in a mouse model of A. fumigatus pneumonia. Together these data uncover a novel mechanism of host defense against A. fumigatus infection in which fucosylated receptors in the airway mucus gel and on the surface of macrophages bind FleA to hasten the elimination of A. fumigatus conidia.
To date, research on binding of A. fumigatus conidia to lung proteins has focused on binding to basement membrane proteins in the airway epithelium [29–32]. But mucins provide the first line of defense against inhaled pathogens in the bronchi and bronchioles [25], and studies of binding of A. fumigatus conidia to human airway mucins are highly relevant to mechanisms of infection and invasion. Mucin glycans include multiple fucosylated structures [22], and we report avid binding of FleA to human airway mucins that is inhibited by a wide range of fucose structures, including fucose in α1,2, α1,3, α1,4, or α1,6 linkages. We also report that deletion of FleA in Aspergillus fumigatus conidia markedly decreases mucin binding and that this role for FleA is conserved in the pathogenic A. flavus. We conclude that a range of sterically available fucosylated glycans in mucins can act as FleA ligands and the airway mucus gel is a powerful “sticky” barrier that can capture and remove conidia to guard against invasive infection.
Phagocytosis of inhaled conidia by alveolar macrophages represents an important innate immune defense against A. fumigatus infection, especially in the alveolar lung compartment where mucocilliary clearance does not operate. We explored the role of FleA in phagocytosis of A. fumigatus conidia by murine and human macrophages using two approaches. First, we compared the phagocytosis of wild type (WT) conidia and fleA-deficient (∆fleA) conidia and showed that macrophage phagocytosis of the ∆fleA conidia is decreased by 40–50%. Second, we screened a library of fucopyranosides to reveal that (2E)-hexenyl α-L-fucopyranoside (2EHex) inhibits FleA with nanomolar efficacy and also inhibits the binding and phagocytosis of WT A. fumigatus conidia by lung macrophages. Together, these two lines of evidence leads us to conclude that A. fumigatus conidia interact with macrophages in a mechanism that requires the fucose binding activity of FleA. Multiple glycoproteins on the surface of macrophages could act as the receptors for FleA, including membrane-tethered mucins [24]. Neutrophils are also important in cellular immunity against Aspergillus [33, 34], but we did not study whether neutrophils have defective uptake of fleA-deficient conidia, and it remains unknown whether fucosylated receptors on neutrophils have a role in host defense against Aspergillus. The identity of all of the cell types participating in FleA-dependent host defense in the lung has not been elucidated in this study. While we have described an alveolar macrophage dependent mechanism, it is also likely that other non-alveolar macrophage dependent cellular mechanisms, including neutrophils or monocytes, can contribute to FleA mediated host defense.
The FleA-dependent binding of A. fumigatus conidia to mucin and the poor uptake of ΔfleA conidia by lung macrophages indicate that FleA is not a virulence factor. Instead, we propose it is a fungal protein recognized by the host to promote defense against invasive aspergillosis. We provide evidence for this in our experiments in which we infected immunocompetent mice with A. fumigatus WT or ΔfleA conidia. The mice infected with the ΔfleA conidia developed invasive aspergillosis whereas those infected with WT conidia did not. Notably, the ΔfleA infected mice showed blunted recruitment of immune cells integral to the fungal immune response. Specifically, FACS analysis of BAL cells showed a significant decrease in the number of alveolar macrophages, neutrophils, NK and NKT cells and CD4 and γΔ T cells. Thus, it appears that ΔfleA conidia fail to elicit the same inflammatory response as WT conidia resulting in reduced recruitment of key effector cells in the fungal immune response in the lung. We conclude that the absence of FleA in A. fumigatus conidia results in a hypo-inflammatory response and promotes invasive infection.
Our experiments prove that FleA is a pathogen-associated molecular pattern that can be recognized by lung mucins and macrophages to protect the host from infection. The fact that we show that A. flavus also uses FleA to bind lung mucin and that phylogenetic analysis reveals FleA present in several pathogenic species may suggest a conserved role for FleA in fucose-mediated host pathogen interactions. It is unclear why A. fumigatus and other pathogenic fungi have evolved to express fleA. Since fungi commonly grow on carbon-rich carbohydrate substrates, it is possible that FleA is involved in helping these organisms to establish a niche on the surface of carbohydrate-rich substrates.
Our finding for a role for FleA-fucose mediated mechanism of host defense against A. fumigatus does not cast any doubt on the validity of the well established role for β-glucan/Dectin-1 interactions [35]. β-glucan is abundant on swollen conidia and hyphae, and Dectin-1 binds β-glucan on swollen conidia to trigger a robust host defense response [36]. But Dectin-1 does not bind to resting conidia, because these conidia have masked exposure of β-glucans due to the action of hydrophobin proteins [37, 38] and the mechanism of clearance of resting conidia from the lung is not completely understood. We propose a model whereby fucosylated receptors on mucins and macrophages interact with FleA on resting conidia to facilitate clearance of resting conidia via the mucociliary escalator and/or macrophage ingestion. Conidia that escape this initial response and mature (swell and germinate), possibly through the shedding of FleA, would be cleared by Dectin-1 mediated phagocytosis [9]. Our data thus reveal contrasting roles for distinct protein carbohydrate interactions in host immune responses to A. fumigatus. On one hand, a macrophage expressed lectin—Dectin-1—recognizes and binds a carbohydrate structure (β-glucan) on A. fumigatus conidia to promote phagocytosis and killing. On the other hand, we reveal here that fucosylated carbohydrates on macrophages engage a conidial lectin (FleA) to promote phagocytosis and killing. We conclude that protein- and carbohydrate-based defenses on macrophages provide complimentary mechanisms to prevent potentially fatal Aspergillus lung infections.
Induced sputum was collected from 5 healthy nonsmoking, non-allergic subjects aged between 24–55 years (4 male) as described [39]. Bronchoalveolar lavage (BAL) was collected from 3 healthy non-smoking, non-allergic subjects aged 30–45 years (2 male, 1 female) by instilling 4 aliquots of 50mLs of warmed (37°C) normal saline into a segmental bronchus in the right middle lobe or lingula. After the first two 50mL aliquots were instilled and aspirated from one segmental bronchus, the bronchoscope was moved to an adjacent bronchus in the same segment for collection of two additional 50 mL aliquots.
8M guanidine hydrochloride was added in 1:1 volume to sputum samples and the samples rotated at 4°C until homogenized. Mucins were then purified from the sputum as described previously [40, 41] with additional details in S1 Text.
Recombinant FleA (prepared as described previously [18]) was biotinylated with EZ-link sulfo NHS biotin (Pierce, Thermo Fisher, Rockford, IL). Purified human mucin was coated on a Nunc maxisorp plate at 20 μg/ml in carbonate bicarbonate buffer pH 9.6 overnight at 4°C, washed and blocked with TBS + 0.05% Tween-20, 10mM CaCl2, 3% BSA. Biotinylated recombinant FleA was incubated at 5 μg/ml in TBS + 0.05% Tween-20, 10 mM CaCl2, 1% BSA (binding buffer) in the presence or absence of 100mM L-fucose or 100mM L-galactose. For inhibition assays, recombinant FleA was incubated with a dilution series of synthesized carbohydrate compounds starting at 5mM. Plates were washed with binding buffer, incubated with ExtrAvidin-alkaline phosphatase (Sigma-Aldrich, St Louis, MO) and detected using phosphatase substrate (Sigma-Aldrich, St Louis, MO) in carbonate bicarbonate buffer pH 9.6 + 1 mM MgCl2 and read at 405nm (Biotek Synergy plate reader, Winooski, VT).
The disaccharides were synthesized by glycosylating appropriately protected glucose acceptors with a fucosyl bromide donor, 2,3,4-tri-O-benzyl-L-fucopyranosyl bromide, using Lemieux’s halide assisted conditions [42], followed by deprotection of the α-linked disaccharides using catalytic hydrogenolysis to give the target structures. Detailed synthesis methods are in S1 Text.
Unless noted, all A. fumigatus strains were propagated on solid glucose minimal media (GMM) at 37°C [43]. A. fumigatus asexual spore suspensions were fixed, where appropriate, in 4% formaldehyde in PBS.
A. fumigatus strains expressing GFP were constructed using pJMP51 to transform AF293.1 and AF293.6 which yielded TJMP131.5 (GFP::H2A) and TGJF5.3 (GFP::H2A, argB1), respectively. A fleA gene disruption cassette (Fig 2A) was used to transform TGJF5.3 to create TGJF6.7, 6.8 and 6.13 (GFP::H2A, ∆fleA). The strain was confirmed by Southern and northern analysis (Fig 2B and 2C). TGJF5.3 was also transformed with a FleA RFP (fleA:RFP) tagged cassette yielding the prototrophic fleA:RFP strain, TGJF7.11. FleA tagging was confirmed microscopically and by Southern and northern analysis Fig 2D and 2E). A. flavus ΔfleA deletion mutants were created by transforming the deletion construct into parental strain CA14∆ku70∆pyrG [44] to create strains TFYL62.1–62.3. Single integration of the deletion cassette was verified via Southern analysis (S2 Fig). More detailed methods are available in S1 Text.
A. fumigatus strains were cultured on GMM at 37°C for 3 days, spores were harvested, placed on a pre-cleaned glass slide and coverslipped. Images were taken of GFP and RFP fluorescence using a Nikon Ti inverted microscope equipped with a Nikon Plan Apo VC 60x/1.40 Oil DIC/∞/0.17 WD. Time-course microscopy was carried out over 27 hours at 37°C. The average fluorescent intensity at each developmental state (resting conidia, swollen conidia, and hyphae) of untagged FleA (TJMP131.5 or wild type) was subtracted from the mean fluorescent intensity value of two different transformants (TGJF7.11 and TGJF7.15) expressing RFP-tagged versions of FleA. The adjusted mean fluorescence was then standardized to area.
Resting and swollen A. fumigatus conidia and hyphae were isolated from WT and ΔfleA cultures as described in S1 Text and extracted in 50mM Tris/HCl pH 7.4, 50 mM EDTA, 2% SDS, and 40 mM β-Mercaptoethanol. Protein concentrations were determined by BCA assay and 15 μg of protein was loaded into each well of a 4–12% BOLT SDS PAGE gel (Life Technologies, Grand Island, NY) and electrophoresed. Gel was then blotted onto nitrocellulose, blocked with non fat milk and stained with an anti-FleA rabbit polyclonal antibody [18] and donkey anti-rabbit HRP (Jackson immunoresearch, West Grove, PA) prior to chemiluminescent detection. Culture supernatants were filtered through a 0.2 μM filter and concentrated 10x in a 0.5 ml Amicon Ultra (EMD Millipore, Billirica, MA) before being run as described above.
8 well glass chamber slides (Labtek, Scotts Valley,CA) were coated with 20 μg/ml purified human mucin in dH2O overnight at 37°C and then blocked in PBS + 1% BSA for 1 hour. Fixed A. fumigatus conidia suspensions were centrifuged at 6000 x g for 5 minutes to pellet and resuspended in PBS + 1% BSA in the presence or absence of 10mM (2E)-hexenyl α-L-fucopyranoside (2EHex) or 100mM fucose. 2x107 conidia were added per well and incubated for 4 hours at room temperature. Unbound conidia were removed by washing in PBS+ BSA, the slides were mounted in Prolong Gold anti-fade reagent (Life Technologies, Grand Island, NY) and allowed to cure for 24 hours prior to sealing. Images were acquired using an FV10i confocal microscope (Olympus, Center Valley, PA) using the multipoint Z-stack mode to acquire 9 fields per well with 3 wells imaged per condition per experiment. Each Z-stack image was compressed into a single plane of focus and conidia were counted using NIH Image J with the ITCN plugin. Each experiment was repeated at least 3 times. A. flavus-mucin interactions were investigated as described above with one exception. These conidia lack GFP so were stained with Calcofluor white for 5 minutes to allow imaging prior to adding to mucin-coated slides.
RAW 264.7 cells (UCSF cell culture facility) were maintained in DMEM + 10% fetal bovine serum + 1% penicillin/streptomycin until seeded and grown on 8 well chamber slides (Labtek, Scotts Valley, CA) overnight. Human alveolar macrophages from BAL were centrifuged at 450 x g for 10 minutes and washed with PBS prior to plating on poly-L-lysine coated 8 well chamber slides in RPMI 1640+ 10% fetal bovine serum + 1% penicillin/streptomycin + 0.5μg/ ml amphotericin B. Cells were washed after 2 hours of adherence and cultured overnight prior to experiments.
RAW264.7 or primary human lung macrophages were incubated with Alexa-488 tagged recombinant FleA in the presence or absence of 100mM fucose prior to analysis on a Becton Dickenson FACScalibur and Flow Jo software (Treestar, Ashland, OR).
RAW 264.7 cells were plated at 5x104/well on 8 well chamber slides (Labtek, Scotts Valley, CA) and allowed to grow overnight in culture media. 5x106 conidia from either the PFA-fixed WT or ΔfleA strain were added per well in the presence or absence of 10mM 2EHex or 500mM fucose and incubated at 37°C for 1 hour. Wells were washed and incubated for a further 2 hours at 37°C for complete uptake. Cells were stained with 7.5μg/ml CellMask Deep Red plasma membrane stain (Life Technologies, Grand Island, NY) and calcofluor white (Sigma, St Louis, MO), washed with PBS and mounted with Fluoromount-G (Southern Biotech, Birmingham, AL). Z-stack images were acquired using an FV10i confocal microscope (Olympus, Center Valley, PA). Each Z-stack image was compressed into a single plane of focus and both internalized conidia and cell number were counted using NIH Image J with the cell counter plugin. Phagocytic index was calculated as the number of conidia internalized per cell. Internalized conidia were counted as conidia within the boundary of the cell that were not stained with calcofluor white. Calcofluor white stained conidia were excluded from the count as they were not internalized. For human macrophages, cells were plated at 5x105 per well and grown overnight. 1.5x106 conidia from the PFA fixed WT or ΔfleA strains were added per well in the presence or absence of 10mM 2EHEX, incubated at 37°C for 30 minutes then washed, stained and mounted.
Recombinant Dectin-1 was purchased from R&D Systems (Minneapolis, MN) and biotinylated using the EZ-link sulfo NHS biotin kit (Thermo Fisher Waltham, MA) according to manufacturers recommendations. 1x107 WT or ΔfleA conidia were labeled with biotinylated Dectin-1 and Streptavidin-PE (Biolegend, San Diego, CA) and fixed in 4% paraformaldehyde prior to analysis on a Becton Dickenson FACSCalibur and FlowJo software (TreeStar, Ashland, OR).
Yellow-green 1μM sulfate treated FluoSpheres (Life Technologies, Grand Island, NY) were coated with recombinant FleA, added to RAW264.7 cells in the presence or absence of 500mM fucose and incubated with agitation for 2 hours at 37°C prior to extensive washing with DMEM to remove unattached FluoSpheres from the cell surface. Cells were fixed in 4% paraformaldehyde prior to analysis on a Becton Dickenson FACSCalibur and FlowJo software (TreeStar, Ashland, OR).
Male C57BL/6 mice from Charles River Laboratories (Wilmington, Massachusetts, MA) were housed in a pathogen free facility at UCSF. Animal experiments followed protocols approved by the UCSF Institutional Animal Care and Use Committee.
Eight to ten week old C57BL/6 mice were infected intranasally with 5x107 of WT or ΔfleA conidia. Mice were sacrificed 3 days post infection. Measures of lung inflammation, infection and lung injury were evaluated using methods described in S1 Text.
Data analyses were performed GraphPad Prism version 6 (GraphPad, San Diego, CA). ANOVA was used for three-group comparisons followed by pairwise analyses with the Tukey multiple comparisons test when appropriate. Two group comparisons were analyzed using the Students t-test or for non-parametric analyses, a ranked Mann-Whitney test.
Human samples were obtained from the UCSF Airway Tissue Bank (ATB). All participants signed two informed consent forms—one for the original study protocol and the other for the ATB protocol. All study and ATB procedures were reviewed and approved by the UCSF Committee on Human Research, protocol number 11–05176. All animal studies were carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institute of Health. All protocols involving animals were approved by the animal care and use committee at UCSF and are in compliance with Public Health Service Policy (PHS), (IACUC protocol number AN090458-03).
FleA, A. fumigatus, NC_007198.1, XP_753183.1
FleA, A. flavus, NW_002477244.1, XP_002380155.1
PAIIL, NC_002516.2 NP_252051.1
Dectin-1, NM_022570.4, NP_072092.2
MUC5AC, NM_001304359.1, NP_001291288.1
MUC5B, NM_002458.2, NP_002449.2
|
10.1371/journal.pgen.1004254 | Worldwide Patterns of Ancestry, Divergence, and Admixture in Domesticated Cattle | The domestication and development of cattle has considerably impacted human societies, but the histories of cattle breeds and populations have been poorly understood especially for African, Asian, and American breeds. Using genotypes from 43,043 autosomal single nucleotide polymorphism markers scored in 1,543 animals, we evaluate the population structure of 134 domesticated bovid breeds. Regardless of the analytical method or sample subset, the three major groups of Asian indicine, Eurasian taurine, and African taurine were consistently observed. Patterns of geographic dispersal resulting from co-migration with humans and exportation are recognizable in phylogenetic networks. All analytical methods reveal patterns of hybridization which occurred after divergence. Using 19 breeds, we map the cline of indicine introgression into Africa. We infer that African taurine possess a large portion of wild African auroch ancestry, causing their divergence from Eurasian taurine. We detect exportation patterns in Asia and identify a cline of Eurasian taurine/indicine hybridization in Asia. We also identify the influence of species other than Bos taurus taurus and B. t. indicus in the formation of Asian breeds. We detect the pronounced influence of Shorthorn cattle in the formation of European breeds. Iberian and Italian cattle possess introgression from African taurine. American Criollo cattle originate from Iberia, and not directly from Africa with African ancestry inherited via Iberian ancestors. Indicine introgression into American cattle occurred in the Americas, and not Europe. We argue that cattle migration, movement and trading followed by admixture have been important forces in shaping modern bovine genomic variation.
| The DNA of domesticated plants and animals contains information about how species were domesticated, exported, and bred by early farmers. Modern breeds were developed by lengthy and complex processes; however, our use of 134 breeds and new analytical models enabled us to reveal some of the processes that created modern cattle diversity. In Asia, Africa, North and South America, humpless (Bos t. taurus or taurine) and humped (Bos t. indicus or indicine) cattle were crossbred to produce hybrids adapted to the environment and local production systems. The history of Asian cattle involves the domestication and admixture of several species whereas African taurines arose through the introduction of domesticated Fertile Crescent taurines and their hybridization with wild African aurochs. African taurine genetic background is commonly observed among European Mediterranean breeds. The absence of indicine introgression within most European taurine breeds, but presence within three Italian breeds is consistent with at least two separate migration waves of cattle to Europe, one from the Middle East which captured taurines in which indicine introgression had already occurred and the second from western Africa into Spain with no indicine introgression. This second group seems to have radiated from Spain into the Mediterranean resulting in a cline of African taurine introgression into European taurines.
| High-throughput genotyping assays have allowed population geneticists to use genome-wide marker sets to analyze the histories of many species, including human [1], cattle [2]–[4], sheep [5], dog [6], horse [7], yeast [8], mouse [9], [10], rice [11], [12], maize [13]–[16], grape [17], and wheat [18]. We previously described the phylogeny of domesticated bovine populations using their genetic variation inferred from a sample of 40,843 single-nucleotide polymorphisms (SNPs) [3]. Although we had sampled 48 cattle breeds, we did not have samples from key geographic regions including China and Southeast Asia, Anatolia, the Baltic States, southern and eastern Africa, and the Iberian Peninsula. As a consequence of those gaps in geographic sampling, we were unable to address the origins of cattle in these regions and the extent to which these cattle influenced the population structure of regions such as the New World.
We have now assembled a genomic data set which represents the largest population sampling of any mammalian species. This allows for an extremely detailed description of the population structure of domesticated cattle worldwide. Using this data set, we accurately establish the patterns of exportation, divergence, and admixture for domesticated cattle.
We used principal component analysis (PCA) [19], ancestry graphs implemented in TreeMix [20], and ancestry models implemented in ADMIXTURE [21] to analyze the relationships between 134 breeds of domesticated bovids (Table S1). These breeds arose from three domesticated (sub)species: Bos javanicus, Bos taurus indicus and Bos taurus taurus (we use the terms breed and population interchangeably, due to the different definitions of breed worldwide). The principal source of SNP genotype variation was between B. t. taurus and B. t. indicus breeds (Figure 1). This split corresponds to the cattle which originated from the two separate major centers of domestication in the Fertile Crescent and Indus Valley [22]. Although Bos javanicus has a more distant common ancestor compared with Bos t. indicus and Bos t. taurus [3], the uneven sample sizes and ascertainment of SNPs common in Bos t. taurus in the design of the BovineSNP50 assay [23] caused the Bos t. indicus/Bos t. taurus split to be the main source of variation in these data. The second principal component split African taurine cattle from Eurasian taurine, indicine, and Bali cattle.
Early farmers were able to expand their habitat range because of the availability of a reliable supply of food and likely displaced indigenous hunter-gatherer populations by introducing new diseases [24]. The genomes of modern cattle reflect the history of animal movements by migratory farmers out of the ancient centers of cattle domestication. We first ran TreeMix with all 134 populations to identify patterns of divergence (Figure 2). We next ran TreeMix with 74 representative populations (Figure 3, residuals presented in Figure S1) and began to add migration edges to the phylogenetic model (Figure 4, residuals presented in Figure S2, see Methods for an explanation of TreeMix). The proportion of the variance in relatedness between populations explained by the model began to asymptote at 0.998 (a value also obtained by simulations [20]) when 17 migration edges were fit (Figure S3). The consistency of these migration edges was evaluated using 5 independent runs of TreeMix with 17 migration edges (Figure S4). In addition to the migratory routes previously described from the Fertile Crescent to Europe [3], we now find strong evidence of exportations from the Indian subcontinent to China and southeast Asia, India to Africa, Africa to the Iberian Peninsula and Mediterranean Europe, India to the Americas, and Europe to the Americas (Figures 4 and 5, discussed in detail in the following subsections). Subsequent to these initial exportations, there have been countless exportations and importations of cattle worldwide. When domesticated cattle were present and new germplasm was imported, the introduced cattle were frequently crossed with the local cattle resulting in an admixed population. Admixed populations were most readily identified when Bos t. indicus and Bos t. taurus animals were hybridized, which occurred in China, Africa, and the Americas (crosses in Figure 1).
In the late 18th and 19th centuries, European cattlemen began forming closed herds which they developed into breeds [25]. Because breeds are typically reproductively isolated with little or no interbreeding, we found that the cross-validation error estimates continued to decrease as we increased the number of ancestral populations K modeled in the admixture analysis (Table S2). This reflects the large differences in allele frequencies that exist between breeds resulting from separate domestication events, geographic dispersal and isolation, breed formation, and the use of artificial insemination. The method of Evanno et al. [26], which evaluates the second order rate of change of the likelihood function with respect to K (ΔK), identified K = 2 as the optimum level of K (Figure S5). This method was overwhelmed by the early divergence between indicine and taurine cattle, and was not sensitive to the hierarchical relationships of populations and breeds [27]. As we increased the value of K, we recapitulated the increasingly fine structure represented in the branches of the phylogeny (Figures 6, S6, S7, S8, S9, S10).
Anatolian breeds (AB, EAR, TG, ASY, and SAR) are admixed between blue Fertile Crescent, grey African-like, and green indicine-like cattle (Figures 5 and 6), and we infer that they do not represent the taurine populations originally domesticated in this region due to a history of admixture. Zavot (ZVT), a crossbred breed [25], has a different history with a large portion of ancestry similar to Holsteins (Figures 2 and S8, S9, S10). The placement of Anatolian breeds along principal components 1 and 2 in Figure 1 [23], the ancestry estimates in Figure 6, their extremely short branch lengths in Figures 2–4, and significant f3 statistics confirm that modern Anatolian breeds are admixed (see Methods for explanation of f-statistics). For example, the Anatolian Southern Yellow (ASY) has 3,003 significant f3 tests, the most extreme of which has Vosgienne (VOS, a taurine breed) and Achai (ACH, an indicine breed) as sister groups with a Z-score of −43.69. Our results support previous work using microsatellite loci [28] which inferred Anatolian cattle to possess indicine introgression. We further demonstrate that Anatolian breeds have introgression from African taurine. We calculated f4 statistics with East Anatolian Red, Anatolian Southern Yellow, and Anatolian Black as sister, and N'Dama, Somba, Lagune, Baole, Simmental, Holstein, Hereford, and Shorthorn as the opposing sister group. From Figure 2, we would expect these relationships to be tree-like. But 45 of the possible 84 f4 tests indicated significant levels of admixture. The most significant was f4(East Anatolian Red, Anatolian Southern Yellow; Somba, Shorthorn) = −0.0026±0.0003 (Z-score = −8.10, alternative trees have Z-scores of 9.88 and 5.20).
If African and Asian taurines were both exported from the Fertile Crescent in similar numbers at about the same time, we would expect them to be approximately equally diverged from European taurines. However, African taurines were consistently revealed to be more diverged from European and Asian taurines (Figures 1, 2, 3, and 5, Anatolian breeds are not considered in this comparison because of their admixed history). Two factors appear to influence this divergence. First, European cattle were exported into Asia and admixed with Asian taurines. In the admixture models in which K = 15 or 20 (Figures S9 and S10), there was evidence of European taurine admixture in the Mongolian (MG), Hanwoo (HANW), and Wagyu (WAGY) breeds. We ran TreeMix with 14 representative populations and estimated Wagyu to have 0.188±0.069 (p-value = 0.003) of their genome originating from northwestern European ancestry (Figure 7). We also see some runs of TreeMix placing a migration edge from Chianina cattle to Asian taurines (Figure S4). We ran f4 tests with Mongolian, Hanwoo, Wagyu, Tharparkar (THA), or Kankraj (KAN) as sister populations, and Piedmontese (PIED), Simmental (SIM), Brown Swiss (BSW), Braunvieh (BRVH), Devon (DEV), Angus (AN), Shorthorn (SH), or Holstein (HO) as the opposing pair of sister groups. From previous research [3] and Figures 2 and 3, these relationships should be tree-like if there were no admixture. For 53 of the possible 280 tests, the Z-score was more extreme than ±2.575829. The most extreme test statistics were f4(Wagyu, Mongolian; Simmental, Shorthorn) = −0.003 (Z-score = −5.21, other rearrangements of these groups had Z-scores of 7.32 and 16.55) and f4(Hanwoo, Wagyu; Piedmontese, Shorthorn) = 0.002 (Z-score = 4.90, other rearrangements of these groups had Z-scores of 21.79 and 27.77). When K = 20, Hanwoo appear to have a Mediterranean influence, whereas Wagyu have a northwestern European, including British, influence (Figure S10). We conclude that there were two waves of European introgression into Far East Asian cattle, first with Mediterranean cattle (which carried African taurine and indicine alleles) brought along the Silk Road [29] and later from 1868 to 1918 when Japanese cattle were crossed with British and Northwest European cattle [25].
The second factor that we believe underlies the divergence of African taurine is a high level of wild African auroch [30], [31] introgression. Principal component (Figure 1), phylogenetic trees (Figures 2 and 3), and admixture (Figure 6) analyses all reveal the African taurines as being the most diverged of the taurine populations. Because of this divergence, it has been hypothesized that there was a third domestication of cattle in Africa [32]–[36]. If there was a third domestication, African taurine would be sister to the European and Asian clade. When no migration events were fit in the TreeMix analyses, African cattle were the most diverged of the taurine populations (Figures 2 and 3), but when admixture was modeled to include 17 migrations, all African cattle, except for East African Shorthorn Zebu and Zebu from Madagascar which have high indicine ancestry, were sister to European cattle and were less diverged than Asian or Anatolian cattle (Figure 4), thus ruling out a separate domestication. Our phylogenetic network (Figure 4) shows that there was not a third domestication process, rather there was a single origin of domesticated taurine (Asian, African, and European all share a recent common ancestor denoted by an asterisk in Figure 4, with Asian cattle sister to the rest of the taurine lineage), followed by admixture with an ancestral population in Africa (migration edge a in Figure 4, which is consistent across 6 separate TreeMix runs, Figure S4). This ancestral population (origin of migration edge a in Figure 4) was approximately halfway between the common ancestor of indicine and the common ancestor of taurine. We conclude that African taurines received as much as 26% (estimated as 0.263 in the network, p-value<2.2e-308) of their ancestry from admixture with wild African auroch, with the rest being Fertile Crescent domesticate in origin. Although three other migration edges originate from the branch between indicine and taurine (such as edge b), all of the receiving populations show indicine ancestry in the ADMIXTURE models. But African auroch are extinct and samples were not available for the ADMIXTURE model, thus the admixed auroch ancestry of African taurines cannot specifically be discovered by this model [27], [37] and African taurine, especially Lagune, are depicted as having a single ancestry without indicine influence (Figures 5 and 6, see f3 and f4 statistics reported later). Unlike ADMIXTURE, TreeMix can model admixture from an unsampled population by placing a migration edge more basal along a branch of the phylogeny, in this case African auroch.
Others have observed distinct patterns of linkage disequilibrium in African taurines, resulting in larger estimates of ancestral effective population size than for either Bos t. taurus or Bos t. indicus breeds [2] consistent with greater levels of admixture from wild aurochs. Just as Near Eastern domesticated pig mitochondrial lineages were replaced by mitochondria from indigenous wild populations [38], we infer that the divergent T1d African mitochondrial subgroup [39] previously observed originated either from Fertile Crescent domesticates or admixture with wild African auroch. Similar patterns of admixture from wild forebears have been observed in other species [38], such as pig [40]–[42], chicken [43], and corn [14], and this conclusion represents the most parsimonious explanation of our results. We hypothesize that the auroch introgression in Africa may have been driven by trypanosomiasis resistance in African auroch which may be the source of resistance in modern African taurine populations [44]. Admixture with distant relatives has had an important impact on the immune system of other species, such as human [45] and possibly chicken [46]. More sophisticated demographic models and unbiased whole-genome sequence data will be needed to further test these hypotheses.
African cattle also demonstrate a geographical gradient of indicine ancestry [47]. Taurine cattle in western Africa possess from 0% to 19.9% indicine ancestry (Figures 5 and 6, LAG, ND1, ND2, NDAM, BAO, OUL, SOM), with an average of 3.3%. Moving from west to east and from south to central Africa, the percent of indicine ancestry increases from 22.7% to 74.1% (Figures 5 and 6, ZFU, ZBO, ZMA, BORG, TULI, BOR, SHK, ZEB, ANKW, LAMB, an AFR), with an average of 56.9%. As we increased values of K to 10, 15, and 20 (Figures S8, S9, S10), we revealed two clusters of indicine ancestry possibly resulting from the previously suggested two waves of indicine importation into Africa, the first occurring in the second millennium BC and the second during and after the Islamic conquests [25], [34], [48]. The presence of two separate clades of African cattle in Figure 4 also supports the idea of two waves of indicine introgression.
Asian cattle breeds were derived from cattle domesticated in the Indian subcontinent or imported from the Fertile Crescent and Europe. Cattle in the north and northeast are primarily of Bos t. taurus ancestry (Figures 5 and 6; HANW, WAGY, and MG), but Hanwoo and Mongolian also have Bos t. indicus ancestry (Figures 5, 6, S9, and S10). Cattle in Pakistan, India, southern China and Indonesia are predominantly Bos t. indicus (Figures 5 and 6; ONG, MAD, BRE, HN, ACE, PES, ACH, HAR, BAG, GUZ, SAHW, GBI, CHO, GIR, KAN, THA, RSIN, HIS, LOH, ROJ, DHA, and DAJ). Cattle located between these two geographical regions are Bos t. taurus×Bos t. indicus hybrids (Figures 1, 4, 5, and 6; QC and LX). Our results suggest an additional source for increased indicine diversity—admixture with domesticated cattle from other species. In addition to cattle domesticated from aurochs (Bos primigenius), bovids were also domesticated from water buffalo (Bubalus bubalis), yak (Bos grunniens), gaur (Bos gaurus), and banteng (Bos javanicus), represented in our sample by the Bali breed [25], [49]. We find that the Indonesian Brebes (BRE) and Madura (MAD) breeds have significant Bos javanicus (BALI) ancestry demonstrated by the short branch lengths in Figures 2–4, shared ancestry with Bali in ADMIXTURE analyses (light green in Figures S8, S9, S10), and significant f3 statistics (Table S3). The Indonesian Pesisir and Aceh and the Chinese Hainan and Luxi breeds also have Bali ancestry (migration edge c in Figure 4, migration edges in Figure S4, and light green in Figures S8 and S9).
Cattle were imported into Europe from the southeast to the northwest. The descendants of Durham Shorthorns (the ancestral Shorthorn breed [25]) were the most distinct group of European cattle as they clustered at the extremes of principal component 2 (lower left hand corner of Figure 1), and they formed a distinct cluster in the ADMIXTURE analyses whenever K was greater than 4 (Figures S6, S7, S8, S9, S10). As shown in Figures S6 through S10, f3 statistics in Table S4, and from their breed histories [25], many breeds share ancestry with Shorthorn cattle, including Milking Shorthorn, Beef Shorthorn, Lincoln Red, Maine-Anjou, Belgian Blue, Santa Gertrudis, and Beefmaster.
From the previous placement of the American Criollo breeds including Romosinuano, Texas Longhorn, and Corriente, it has been posited that Iberian cattle became admixed as a result of an introgression of cattle from Africa into the local European cattle [3], [50], [51]. Our genotyping of individuals from 11 Spanish breeds supported, but clarified, this hypothesis. On average, Spanish cattle had 19.3% of African ancestry when K = 3, with a minimum of 8.8% and a maximum of 23.4%, which supports previous analyses of mitochondrial DNA [52], [53]. Migration edge d in the phylogenetic network (Figure 4, and consistently seen in Figure S4) estimates that Iberian cattle, Texas Longhorn, and Romosinuano derive 7.5% of their ancestry from African taurine introgression, similar to the ancestry estimates from the models with larger K values (Figures S8, S9, S10). The Oulmès Zaer (OUL) breed from Morocco also shows that cattle were transported from Iberia and France to Africa (tan and red in Figure S10, and short branch length in Figure 4). However, the 11 Spanish breeds had no more indicine ancestry than all other European taurine breeds (essentially none for the majority of breeds, see Figures 5 and 6). Maraichine (MAR), Gascon (GAS), Limousin (LIM), and other breeds from France, and Piedmontese cattle (PIED) from northwest Italy have a similar ancestry. These data indicate that the reason that the American Criollo breeds were found to be sister to European cattle in our previous work [3] was because of their higher proportion of indicine ancestry. The 5 sampled American Criollo breeds had, on average, 14.7% African ancestry (minimum of 6.2% and maximum of 20.4%) and 8.0% indicine ancestry (minimum of 0.6% and maximum of 20.3%).
Other Italian breeds (MCHI, CHIA, and RMG) share ancestry with both African taurine and indicine cattle (Figures 6, S6, S7, S8). This introgression may have come from Anatolian or East African cattle that carried both African taurine and indicine ancestry, which is modeled as migration edge b in Figure 4. The placement of Italian breeds is not consistent across independent TreeMix runs (Figure S4), likely due to their complicated history of admixture.
We also used f-statistics to explore the evidence for African taurine introgression into Spain and Italy. We did not see any significant f3 statistics, but this test may be underpowered because of the low-level of introgression. With Italian and Spanish breeds as a sister group and African breeds, including Oulmès Zaer, as the other sister group, we see 321 significant tests out of 1,911 possible tests. Of these 321 significant tests, 218 contained Oulmès Zaer. We also calculated f4 statistics with the Spanish breeds as sister and the African taurine breeds as sister (excluding Oulmès Zaer). With this setup, out of the possible 675 tests we saw only 1 significant test, f4(Berrenda en Negro, Pirenaica;Lagune, N'Dama (ND2)) = 0.0007, Z-score = 3.064. With Italian cattle as sister and African taurine as sister (excluding Oulmès Zaer), we saw 17 significant tests out of the 90 possible. Patterson et al. [54] defined the f4-ratio as f4(A, O; X, C)/f4(A, O; B, C), where A and B are a sister group, C is sister to (A,B), X is a mixture of B and C, and O is the outgroup. This ratio estimates the ancestry from B, denoted as , and the ancestry from C, as . We calculated this ratio using Shorthorn as A, Montbeliard as B, Lagune as C, Morucha as X, and Hariana as O. We choose Shorthorn, Montbeliard, Lagune, and Hariana as they appeared the least admixed in the ADMIXTURE analyses. We choose Morucha because it appears as solid red with African ancestry in Figure S10. This statistic estimated that Morucha is 91.23% European ( = 0.0180993/0.0198386) and 8.77% African, which is similar to the proportion estimated by TreeMix. The multiple f4 statistics with Italian breeds as sister and African breeds as the opposing sister support African admixture into Italy. The f4-ratio test with Morucha also supports our conclusion of African admixture into Spain.
It has recently been concluded that indicine ancestry is a common feature of European cattle genomes [55]. However, our data refute this conclusion. McTavish et al. relied on the Evanno test to arrive at an optimal number of ancestral populations of K = 2, which masks the fact that there are cattle breeds in Africa with 100% African taurine ancestry (Figure 6). Although our K = 2 ADMIXTURE results suggested that most African breeds had at least 20% indicine ancestry (Figure S5), when we increased K to 3, Lagune (LAG) revealed no indicine ancestry, and Baoule (BAO) and N'Dama (NDAM) possess very little indicine ancestry. If the K = 2 model was correct, we would expect to see numerous significant f3 and f4 tests with Eurasian taurine and indicine as sister groups. Whereas, if the K = 3 model more accurately reflected the heritage of European and African taurines, we would not observe any significant f3 or f4 tests showing admixture of taurine and indicine in the ancestry of African taurine. For the Lagune, Baoule and N'Dama (NDAM and ND2) breeds we found no significant f3 statistics. Among the 225 f4 statistics calculated with NDAM, LAG, BAO, ND2, SH, and MONT as sisters and BALI, GIR, HAR, SAHW, PES, and ACE as the opposing sister group, only 36 were significantly different from 0 (Table 1). When ND2 was excluded from the results, only 4 tests were significant (Table 1), and we have no evidence that the Lagune breed harbors indicine alleles. Thus, we conclude that contrary to the assumptions and conclusions of [55] cattle with pure taurine ancestry do exist in Africa. Further, we conclude that indicine ancestry in European taurine cattle is extremely rare, and that some breeds, especially those prevalent near the Mediterranean, possess African taurine introgression—but with the exception of the Charolais, Marchigiana, Chianina and Romagnola breeds—not African hybrid or African indicine introgression. We concur that Texas Longhorn and other American Criollo breeds possess indicine ancestry, but infer that this introgression occurred after the arrival of Spanish cattle in the New World and likely originated from Brahman cattle (migration edges e and f in Figure 4). In TreeMix replicates, Texas Longhorn and Romosinuano are either sister to admixed Anatolian breeds or they receive a migration edge that originates near Brahman (Figure S4). To reiterate, Iberian cattle do not have indicine ancestry, American Criollo breeds originated from exportations from Iberia, Brahman cattle were developed in the United States in the 1880's [25], American Criollo breeds carry indicine ancestry, and the introgression likely occurred from Brahman cattle.
Domestication, exportation, admixture, and breed formation have had tremendous impacts on the variation present within and between cattle breeds. In Asia, Africa, North and South America, cattle breeders have crossbred Bos t. taurus and Bos t. indicus cattle to produce hybrids which were well suited to the environment and endemic production systems. In this study, we clarify the relationships between breeds of cattle worldwide, and present the most accurate cattle “Tree of Life” to date in Figure 4. We elucidate the complicated history of Asian cattle involving the domestication and subsequent admixture of several bovid species. We provide evidence for admixture between domesticated Fertile Crescent taurine and wild African auroch in Africa to form the extant African taurine breeds. We also observe African taurine content within the genomes of European Mediterranean taurine breeds. The absence of indicine content within the majority of European taurine breeds, but the presence of indicine within three Italian breeds is consistent with two separate introductions, one from the Middle East potentially by the Romans which captured African taurines in which indicine introgression had already occurred and the second from western Africa into Spain which included African taurines with no indicine introgression. It was this second group of cattle which likely radiated from Spain into Southern France and the Alps. The prevalence of admixture further convolutes the cryptic history of cattle domestication.
We used 1,543 samples in total, including 234 samples from [3] and 425 samples from [4], see Table S1. We selected samples that had fewer than 10% missing genotypes, and for breeds with fewer than 20 genotyped samples, we used all available samples which passed the missing genotype data threshold. When pedigree data were absent for a breed, the 20 samples with the highest genotype call rates were selected. For breeds which had pedigree information, we filtered any animals whose sire or dam was also genotyped. For identified half-siblings, we sampled only the sibling with the highest genotype call rate. After removing genotyped animals known to be closely related, we selected the 20 animals with the highest genotype call rate to represent the breed. All DNA samples were collected in an ethical manner under University of Missouri ACUC approved protocol 7505.
Samples were genotyped with the Illumina BovineSNP50 BeadChip [56]. Autosomal SNPs and a single pseudo-autosomal SNP were analyzed, because the data set from Gautier et al. [4] excluded SNPs located exclusively on the X chromosome. We also filtered all SNPs which mapped to “chromosome unknown” of the UMD3.1 assembly [57]. In PLINK [58], [59], we removed SNPs with greater than 10% missing genotypes and with minor allele frequencies less than 0.0005 (1/[2*Number of Samples] = 0.000324, thus the minor allele had to be observed at least once in our data set). The average total genotype call rate in the remaining individuals was 0.993. Genotype data were deposited at DRYAD (doi:10.5061/dryad.th092) [60].
The sample genotype covariance matrix was decomposed using SMARTPCA, part of EIGENSOFT 4.2 [19]. To limit the effects of linkage disequilibrium on the estimation of principal components, for each SNP the residual of a regression on the previous two SNPs was input to the principal component analysis (see EIGENSOFT POPGEN README).
TreeMix [20] models the genetic drift at genome-wide polymorphisms to infer relationships between populations. It first estimates a dendrogram of the relationships between sampled populations. Next it compares the covariance structure modeled by this dendrogram to the observed covariance between populations. When populations are more closely related than modeled by a bifurcating tree it suggests that there has been admixture in the history of those populations. TreeMix then adds an edge to the phylogeny, now making it a phylogenetic network. The position and direction of these edges are informative; if an edge originates more basally in the phylogenetic network it indicates that this admixture occurred earlier in time or from a more diverged population.
TreeMix was used to create a maximum likelihood phylogeny of the 134 breeds. Because TreeMix was slow to add migration events (modeled as “edges”) to the complete data set of 134 breeds, we also analyzed subsets of the data containing considerably fewer breeds. For these subsets, breeds with fewer than 4 samples were removed. To speed up the analysis, we iteratively used the previous graph with m-1 migrations as the starting graph and added one migration edge for a total of m migrations. We rooted the graphs with Bali cattle, used blocks of 1000 SNPs, and used the -se option to calculate standard errors of migration proportions. Migration edges were added until 99.8% of the variance in ancestry between populations was explained by the model. We also ensured that the incorporated migration edges were statistically significant. To further evaluate the consistency of migration edges, we ran TreeMix five separate times with -m set to 17.
ADMIXTURE 1.21 was used to evaluate ancestry proportions for K ancestral populations [21]. We ran ADMIXTURE with cross-validation for values of K from 1 through 20 to examine patterns of ancestry and admixture in our data set. Map figure was generated in R using rworldmap (http://cran.r-project.org/web/packages/rworldmap/index.html).
The f3 and f4 statistics are used to detect correlations in allele frequencies that are not compatible with population evolution following a bifurcating tree; these statistics provide support for admixture in the history of the tested populations [54], [61]. The THREEPOP program from TreeMix was used to calculate f3 statistics [54] for all possible triplets from the 134 breeds. The FOURPOP program of TreeMix was used to calculate f4 statistics for subsets of the breeds.
|
10.1371/journal.ppat.1000863 | In Vitro Reconstitution of SARS-Coronavirus mRNA Cap Methylation | SARS-coronavirus (SARS-CoV) genome expression depends on the synthesis of a set of mRNAs, which presumably are capped at their 5′ end and direct the synthesis of all viral proteins in the infected cell. Sixteen viral non-structural proteins (nsp1 to nsp16) constitute an unusually large replicase complex, which includes two methyltransferases putatively involved in viral mRNA cap formation. The S-adenosyl-L-methionine (AdoMet)-dependent (guanine-N7)-methyltransferase (N7-MTase) activity was recently attributed to nsp14, whereas nsp16 has been predicted to be the AdoMet-dependent (nucleoside-2′O)-methyltransferase. Here, we have reconstituted complete SARS-CoV mRNA cap methylation in vitro. We show that mRNA cap methylation requires a third viral protein, nsp10, which acts as an essential trigger to complete RNA cap-1 formation. The obligate sequence of methylation events is initiated by nsp14, which first methylates capped RNA transcripts to generate cap-0 7MeGpppA-RNAs. The latter are then selectively 2′O-methylated by the 2′O-MTase nsp16 in complex with its activator nsp10 to give rise to cap-1 7MeGpppA2′OMe-RNAs. Furthermore, sensitive in vitro inhibition assays of both activities show that aurintricarboxylic acid, active in SARS-CoV infected cells, targets both MTases with IC50 values in the micromolar range, providing a validated basis for anti-coronavirus drug design.
| In 2003, an emerging coronavirus (CoV) was identified as the etiological agent of severe acute respiratory syndrome (SARS). SARS-CoV replicates and transcribes its large RNA genome using a membrane-bound enzyme complex containing a variety of viral nonstructural proteins. A critical step during RNA synthesis is the addition of a cap structure to the newly produced viral mRNAs, ensuring their efficient translation by host cell ribosomes. Viruses generally acquire their cap structure either from cellular mRNAs (e.g., “cap snatching” of influenza virus) or employ their own capping machinery, as is supposed to be the case for coronaviruses. mRNA caps synthesized by viruses are structurally and functionally undistinguishable from cellular mRNAs caps. In coronaviruses, methylation of mRNA caps seems to be essential, since mutations in viral methyltransferases nsp14 or nsp16 render non-viable virus. We have discovered an unexpected key role for SARS-CoV nsp10, a protein of previously unknown function, within mRNA cap methylation. Nsp10 induces selective 2′O-methylation of guanine-N7 methylated capped RNAs through direct activation of the otherwise inactive nsp16. This finding allows the full reconstitution of the SARS-CoV mRNA cap methylation sequence in vitro and opens the way to exploit the mRNA cap methyltransferases as targets for anti-coronavirus drug design.
| In 2003, the severe acute respiratory syndrome coronavirus (SARS-CoV), which was likely transmitted from bats, was responsible for a worldwide SARS-outbreak [1]. Coronaviruses belong to the order Nidovirales and are characterized by the largest positive-strand RNA ((+) RNA) genomes (around 30,000 nt) known in the virus world. The enzymology of their RNA synthesis is therefore thought to be significantly more complex than that of other RNA virus groups [2], [3], [4]. The 5′-proximal two-thirds of the CoV genome (open reading frames 1a and 1b) are translated into the viral replicase polyproteins pp1a and pp1ab (Figure 1), which give rise to 16 nonstructural proteins (nsps) by co- and post-translational autoproteolytic processing. The 3′-proximal third encodes the viral structural proteins and several so-called accessory proteins, which are expressed from a set of four to nine subgenomic (sg) mRNAs. The latter are transcribed from subgenome-length minus-strand templates, whose production involves a unique mechanism of discontinuous RNA synthesis (reviewed by [5], [6]). To organize their complex RNA synthesis and genome expression, the CoV proteome includes several enzyme activities that are rare or lacking in other (+) RNA virus families (reviewed in [2]). In the years following the 2003 SARS outbreak, bioinformatics, structural biology, (reverse) genetics and biochemical studies have contributed to the in-depth characterization of CoV nsps in general and those of SARS-CoV in particular [7]. Currently documented enzyme activities include two proteinases (in nsp3 and nsp5; [8], [9]), a putative RNA primase (nsp8; [10]), an RNA-dependent RNA polymerase (nsp12; [11], [12]), a helicase/RNA triphosphatase (nsp13; [13], [14]), an exo- and an endoribonuclease (nsp14 and nsp15; [15], [16], and an S-adenosyl-L-methionine (AdoMet)-dependent (guanine-N7)-methyltransferase (N7-MTase), which were proposed to play a role in the formation of CoV mRNA caps (nsp14; [17]). Based on comparative sequence analysis, nsp16 presumably encodes an AdoMet-dependent mRNA cap (nucleoside-2′O)-methyltransferase (2′O-MTase) [3], [18], [19]. For SARS-CoV nsp16, however, this enzyme activity has remained elusive thus far, and experimental evidence for its existence has only been obtained for the related feline coronavirus (FCoV) nsp16 [18]. CoV nsps form the viral replication/transcription complex (RTC), which is thought to localize to a network of endoplasmic reticulum-derived, modified membranes in the infected cell [20], [21]. Protein-protein interactions were proposed to be essential for the assembly of the RTC and may therefore also regulate the activities of enzymes involved in viral RNA synthesis.
Although the 5′ ends of SARS-CoV mRNAs have not been characterized yet, they are assumed to carry a cap structure. This assumption is based on the characterisation of genomic and subgenomic mRNAs of the coronavirus murine hepatitis virus (MHV) [22], [23] and the related equine torovirus (EToV or Berne virus), which also belong to the Coronaviridae family [23], [24]. The mRNAs of both viruses were concluded to carry a 5′-terminal cap structure. Moreover, in the coronavirus and torovirus genome three enzymes putatively involved in mRNA capping have been identified, although they remain poorly characterised [13], [14], [17], [18], [19]. Cap structures promote initiation of translation and protect mRNAs against exoribonuclease activities [25], [26], [27]. The synthesis of the cap structure in eukaryotes involves three sequential enzymatic activities: (i) an RNA triphosphatase (RTPase) that removes the 5′ γ-phosphate group of the mRNA; (ii) a guanylyltransferase (GTase) which catalyzes the transfer of GMP to the remaining 5′-diphosphate terminus; and (iii) an N7-MTase that methylates the cap guanine at the N7-position, thus producing the so-called “cap-0 structure”, 7MeGpppN. Whereas lower eukaryotes, including yeast, employ a cap-0 structure, higher eukaryotes convert cap-0 into cap-1 or cap-2 structures [25], [26], [28] by means of 2′O-MTases, which methylate the ribose 2′O-position of the first and the second nucleotide of the mRNA, respectively. RNA cap methylation is essential since it prevents the pyrophosphorolytic reversal of the guanylyltransfer reaction, and ensures efficient binding to the ribosome [25], [26]. In the case of (+) RNA viruses such as alphaviruses and flaviviruses, mutations in RNA cap methylation genes were shown to be lethal or detrimental to virus replication [29], [30], [31], [32], [33]. For coronaviruses, a functional and genetic analysis performed on MHV temperature sensitive mutants mapping to the N7-MTase domain of CoV nsp14 and in the 2′O-MTase nsp16 indicated that both are involved in positive-strand RNA synthesis by previously formed replicase-transcriptase complexes [11]. The importance of nsp14 and nsp16 for viral RNA synthesis is further supported by data obtained by mutagenesis of MTase catalytic residues in SARS-CoV RNA replicon systems [17], [30].
In the case of coronaviruses, the machinery putatively involved in equipping both genome and subgenomic mRNAs with a cap-1 structure is thought to consist of (i) the multifunctional nsp13, which may contribute the RTPase activity of the helicase domain [13], [34], (ii) a still unknown GTase, (iii) the C-terminal domain of nsp14, which was recently identified as the N7-MTase [17] and (iv) nsp16, the putative 2′O-MTase [3], [17], [18].
Using mammalian and yeast two-hybrid systems as well as pull-down assays, it was shown that SARS-CoV nsp14 and nsp16 specifically interact with nsp10 [35], [36] suggesting that nsp10 may play a role in the viral capping pathway. The crystal structure of nsp10, a small RNA-binding protein that contains two zinc fingers, was recently solved [37], [38], but its role and mode of action in the viral replicative cycle remains elusive. In view of the phenotype of some mouse hepatitis virus (MHV) mutants, a role in viral RNA synthesis was postulated [11], [39], but other studies implicated nsp10 in replicase polyprotein processing [40]. SARS-CoV nsp10 was also shown to bind single- and double-strand RNA and DNA with low affinity and without obvious sequence specificity [37].
In this study, we report the discovery of a function for SARS-CoV nsp10 as an essential factor to trigger full nsp16 2′O-MTase activity. We deciphered the RNA cap methylation sequence where the guanine-N7-methylation by nsp14 necessarily precedes the 2′O-methylation by the nsp10/nsp16 pair. The SARS-CoV nsp10/nsp14/nsp16 trio constitutes an attractive target package for antiviral drug discovery and design; and indeed nsp14 and nsp16 seem to play an important role in viral replication [11], [17], [30]. Accordingly, we set up sensitive inhibition tests for both activities, validated by low IC50 values of known AdoMet-dependent MTase inhibitors. Moreover, we show that aurintricarboxylic acid (ATA), which was shown to inhibit SARS-CoV replication [41], targets both MTases indeed.
Unlike flaviviruses, which use a single active site in the NS5 protein for both N7- and 2′O-MTase activities [32], [42], coronaviruses presumably encode two separate MTases that catalyze the last two steps in the formation of a methylated RNA-cap structure. SARS-CoV nsp14 has been shown to be an RNA-cap N7-MTase [17]. Sequence motifs that are canonical in 2′O-MTases were identified in nsp16 [3], [19], but the experimental verification of the MTase activity has not been reported, in contrast to FCoV nsp16, for which a rather low activity could be demonstrated [18]. SARS-CoV nsp10 was previously shown to interact with both nsp14 and nsp16 [35], [36], suggesting its involvement in RNA capping and/or methylation. Consequently, we cloned and expressed both nsp10 and nsp16 in E. coli and purified both recombinant proteins, using their N-terminal His6-tag, by metal affinity chromatography. Nsp14 was expressed as a fusion protein with an intein tag at its C-terminus. The nsp14-intein product was bound to a chitin affinity column and the untagged protein was eluted after removal of the tag by DTT treatment. All three proteins were further purified by size exclusion chromatography. Upon SDS-PAGE, the purified proteins migrated as single bands corresponding to their expected molecular masses (nsp14: 57 kDa; His6-nsp16: 35 kDa, and His6-nsp10: 15 kDa) (Figure 2A). The identity of the recombinant proteins was confirmed by trypsin digestion and mass spectrometry ((MALDI-TOF), data not shown).
Using the purified recombinant proteins, we first conducted in vitro MTase assays on short capped RNA substrates methylated or not at the N7-position of the guanine cap (7MeGpppAC5 and GpppAC5). We used all possible combinations of the three proteins (nsp10, nsp14, and nsp16) and incubated them with the substrate in the presence of the tritiated methyl donor [3H]-AdoMet. The extent of [3H]-CH3 transfer was quantified after reaction times of 5, 30, and 240 min by using a DEAE filter-binding assay (see Materials and Methods). Figure 2B shows that nsp14 methylated GpppAC5 in a time-dependent manner whereas neither nsp16 nor nsp10 alone did. Apparently, the activity of nsp14 was barely influenced by the presence of nsp10 or nsp16. In addition, we observed that nsp14 did not methylate 7MeGpppAC5 (Figure 2C) suggesting that nsp14 methylates only the N7-position of the cap structure. In contrast to nsp14, nsp16 catalyzed methyltransfer to neither GpppAC5 nor to 7MeGpppAC5 under these reaction conditions. Surprisingly, when nsp16 activity assays were supplemented with nsp10, robust methylation of 7MeGpppAC5 was observed (Figure 2C), but not of GpppAC5 (Figure 2B). In control reactions, containing either nsp10 alone or nsp10 supplemented with nsp14 no 7MeGpppAC5-specific MTase activity was detected (Figure 2C). When the GpppAC5 substrate was incubated with a combination of nsp10, nsp14, and nsp16 (Figure 2B), the level of substrate methylation was enhanced compared to reactions performed with nsp14 alone. After overnight incubation of GpppAC5 with the three proteins, the methyl incorporation reached a plateau and the incorporation level was twice higher than after a reaction in the presence of nsp14 alone (not shown). In contrast, no significant difference was observed between the methylation level reached after incubation of the 7MeGpppAC5 substrate with either all three proteins or the nsp16-nsp10 pair only (Figure 2C). Taken together, these results suggest that, (i) SARS-CoV nsp14 methylates GpppAC5 at the N7-position of the cap guanine and indeed acts as an N7-MTase on these substrates, (ii) nsp16 acts as an nsp10-dependent 2′O-MTase on 7MeGpppAC5, (iii) the 2′O-MTase activity of nsp16-nsp10 requires the presence of a cap structure already methylated at its N7-position and (iv) nsp14 and the nsp16-nsp10 pair can perform sequential double methylation of GpppAC5, presumably at the N7- and 2′O-positions.
To determine how nsp10 stimulated nsp16 MTase activity, we co-expressed in E. coli an N-terminally Strep-tagged nsp10 and a His6-tagged nsp16. The bacterial cell lysate containing these proteins was incubated with Strep-Tactin beads (see Materials and Methods), in order to bind the tagged nsp10. After extensive washing, the proteins bound to the beads were analysed using SDS-PAGE. Figure 2D indicates that nsp16 remained associated with nsp10, whereas nsp16 alone was unable to bind to the beads. These data suggest that nsp10 can stimulate the MTase activity of nsp16 by direct association resulting in the formation of a nsp10/nsp16 complex. When the intensities of the bands corresponding to nsp10 and nsp16 were quantified, a ratio of nsp10 to nsp16 of 1.1 was obtained. Correcting for the respective molecular masses, and assuming that they bind Coomassie blue dye with the same affinity, this yields a nsp10 to nsp16 ratio of about 2.3. This suggests that the complex does not contain a large molar excess of nsp10, as one might have expected due to the fact that nsp10 seems to form dodecamers under certain conditions [38].
In order to test MTase activities of nsp14 and nsp10/nsp16 on virus-specific capped RNA substrates, we synthesized a 5′-triphosphate-carrying RNA corresponding to the first 264 nucleotides of the SARS-CoV genome using the T7 RNA polymerase. Since the canonical T7 promoter inefficiently directs transcription of RNA beginning with an A, as is required to make transcripts resembling the 5′ end of coronavirus RNAs, we used the T7 class II φ2.5 promoter [43]. Additionally, we introduced a U→G substitution in the 2nd position of the RNA to increase the in vitro transcription efficiency (data not shown). The RNA was capped with [α-32P]-GTP using the vaccinia virus (VV) capping enzyme (containing RTPase, GTase and N7-MTase activities, see Materials and Methods) in the presence or absence of the methyl donor AdoMet. The substrates GpppAG-SARS-264 and 7MeGpppAG-SARS-264 were then incubated with various combinations of nsp14, nsp16, and nsp10. Reaction products were digested by nuclease P1 in order to release the RNA cap structure. Radiolabeled cap molecules were subsequently separated on TLC plates and visualized using autoradiography. The comparison with commercially available and in-house synthesized cap analogs allowed the identification of the methylation position of the cap structure. Figure 3A shows that the cap structure released after nuclease P1 digestion of substrates GpppAG-SARS-264 and 7MeGpppAG-SARS-264 RNA co-migrated, as expected, with GpppA and 7MeGpppA cap analogs, respectively. In the presence of nsp14, or the VV:N7-MTase positive control, the GpppA cap structure present at the 5′ end of the RNA was converted into 7MeGpppA (left panel of Figure 3A). We also observed that the methylation of the N7-position induced by nsp14 was weakly stimulated in the presence of nsp10, but was not influenced by the presence of nsp16. Indeed, nsp14 converts 83% of the substrate into the 7MeGpppA product, whereas in the presence of nsp10 97% of the substrate was converted, as judged by autoradiography analysis. Nsp10 or nsp16 alone did not show any MTase activity. When all three proteins are present, the substrate is fully methylated at the N7- and 2′O-positions of the cap, as judged by the comparison with products generated by the bifunctional N7- and 2′O-MTase domain of dengue virus protein NS5 (DV:NS5MTase), which was used as a positive control [32], [42]. The right panel of Figure 3A shows that incubation of 7MeGpppAG-SARS-264 RNA, with nsp14, nsp16 or nsp10 alone did not result in 2′O-methylation of the 7MeGpppA structure. The same was true when nsp14/nsp10 or nsp14/nsp16 combinations were tested. In contrast, 2′O-methylation of the cap structure of 7MeGpppAG-SARS-264 occurred upon incubation with nsp10/nsp16, and also when all three proteins were used together. We therefore conclude that capped RNA corresponding to the first 264 nucleotides of the SARS-CoV genome represents a bona fide substrate to follow the RNA cap MTase activities of SARS-CoV nsp14 and nsp10/nsp16. Moreover, the TLC analysis allowed us to demonstrate that nsp14 indeed specifically methylates RNA cap structures at the N7-position and that nsp10/nsp16 methylates capped RNA at the 2′O-position of the first nucleotide after the N7-methylated cap. As also observed when using short substrates, nsp10/nsp16 could only methylate 7MeGpppAG-SARS-264 and not GpppAG-SARS-264, suggesting that N7-methylation by nsp14 must precede 2′O-methylation by nsp10/nsp16. We conclude that nsp14 exhibits N7-MTase activity in the absence of nsp10, whereas the latter is an absolute requirement for nsp16-mediated 2′O-methylation of the cap structure. Nsp10, which was previously shown to interact with both nsp14 and nsp16 [35], [36], modestly stimulates the nsp14-mediated cap N7-MTase activity (Figure 3A and S1B; 10 to 15% increase of activity at a broad optimum around a 4-fold molar excess).
In order to directly monitor the order of SARS-CoV RNA-cap methylation, we performed a time-course experiment using the GpppAG-SARS-264 substrate in conjunction with nsp10, nsp14, and nsp16. The results, shown in Figure 3B, indicate that methylation of the substrate indeed starts at the N7-position. Subsequently, the 7MeGpppA cap-0 structure is converted to an 7MeGpppA2'OMe cap-1 structure. A GpppA2'OMe structure was never observed in this assay, not even when using larger amounts of nsp10/nsp16 or nsp16 (data not shown), in agreement with the data presented in Figures 2B and 3A that show that GpppAC5 and GpppAG-SARS-264 substrates are not methylated by nsp10/nsp16. Thus, the N7-methylation of the SARS-CoV cap structure by nsp14 is a pre-requisite for its recognition by the nsp10/nsp16 pair, which then converts the cap-0 into a cap-1 structure by 2′O-methylation.
The recent identification of the C-terminal domain of nsp14 as an N7-MTase [17] revealed that this replicase subunit is a multifunctional protein, since it also carries an exoribonuclease activity embedded in its N-terminal domain [16]. The interplay between these two functionalities was analyzed using mutagenesis experiments. We mutated conserved residues in both the MTase and the exoribonuclease domain to evaluate the possible interplay or long-range regulation of both activities. The conserved residue D331, which is presumably involved in AdoMet-binding, was mutated to alanine. In the exoribonuclease domain, we replaced conserved residues from exonuclease motifs I (D90XE92), II (D243) and III (D273 and H268) of the DE(A/D)D nuclease superfamily. All the His-tagged nsp14 mutant proteins could be expressed, except the D243A mutant, which was barely soluble. Figure 4A shows that they migrated at a molecular mass similar to that of wt nsp14 upon SDS-PAGE. We next analyzed their N7-MTase activity on GpppAC5 using [3H]-AdoMet as methyl donor. The results show that the D331A point mutation completely abolished nsp14 N7-MTase activity. This is in agreement with the hypothesis [17] that the MTase domain is located in the C-terminal half of nsp14 protein and that the conserved residue D331 is important for N7-MTase function. In contrast, the mutations in the exonuclease domain did not significantly interfere with nsp14 MTase activity, excepted in the case of the motif I-double mutant (D90XE92) which displayed attenuated N7-MTase activity (∼2-fold). This observation is in agreement with the fact that a N-terminal truncation of 90 amino acids of the nsp14 exoribonuclease domain abolished the N7-MTase activity in a yeast trans-complementation assay [17]. Thus an altered N-terminus of the exoribonuclease domain may still interfere with the MTase activity to a certain extent.
In order to ascertain that nsp16 supports the 2′O-MTase activity and not the nsp10 protein, we engineered and characterized a set of nsp16 point mutations. We mutated the conserved residues K46-D130-K170-E203, which form the canonical catalytic tetrad of mRNA cap 2′O-MTases [42], [44]. The putative catalytic residues of SARS-CoV nsp16 were identified using sequence alignment with the homologous FCoV nsp16 2′O-MTase and other family members [18]. Three of the four alanine point mutants could be expressed as efficiently as wt nsp16, allowing their purification to homogeneity using a single-step of affinity chromatography (see Materials and Methods). Still, smaller amounts of the fourth mutant (K46A) could also be produced and purified. We obtained sufficient soluble protein to perform MTase assays, although protein yield and purity were lower than for the other mutants (Figure 4B). For all mutant proteins, the 2′O-MTase activity was tested on 7MeGpppAC5 and compared to that of the wt nsp16/nsp10 control pair. The 2′O-MTase activity was indeed completely abolished by any single mutation of the putative K46-D130-K170-E203 catalytic tetrad residues of nsp16. This result demonstrated that although nsp10 stimulates the 2′O-MTase activity by a yet unknown mechanism, the catalytic activity itself resides in nsp16.
Viral MTases exhibit many original features relative to their host cell MTase counterparts, and are increasingly explored as putative targets for the development of antivirals [45]. In order to set up sensitive N7- and 2′O-MTase inhibition tests, we determined more precisely the conditions to measure optimum MTase activity for nsp14 and nsp16/nsp10 using their respective substrates GpppAC5 and 7MeGpppAC5 (see Text S1). We thus defined the following standard assay conditions: the N7-MTase activity of nsp14 was measured in presence of 40 mM Tris-HCl, pH 8.0 and 5 mM DTT. For the 2′O-MTase assays, the same buffer was used together with 1 mM of MgCl2. As Figure S1B illustrates, nsp10 stimulates nsp16 2′O-MTase activity in a dose-dependent manner. We used a 6-fold molar excess of nsp10 over nsp16 corresponding to ≈75% of the maximal stimulation that could be achieved.
Inhibition was tested for two AdoMet analogs with documented mRNA cap MTase inhibition properties: AdoHcy (S-adenosyl-l-homocysteine), the co-product of methyl transfer, and sinefungin [18], [46], [47], [48], [49]. We also used compounds known to target other AdoMet-dependent MTases, such as SIBA, 3-deaza-adenosine [50] and MTA [51]. Based on their adenosine or guanosine-containing structures, adenosine- and AdoMet-analogs 2′,3′,5′-tri-O-acetyl-adenosine and S-5′-adenosyl-L-cysteine were tested as well as GTP, 7MeGTP, GTP- or cap-analogs (ribavirin and its triphosphate as well as EICAR-triphosphate, GpppA and 7MeGpppA). Finally, we included two inhibitors of flavivirus mRNA cap MTase activites: aurintricarboxylic acid (ATA), which is expected to bind to the MTase active site [52], and a substituted adamantane compound supposedly binding to the AdoMet-binding site [53]. Interestingly, ATA has been shown recently to inhibit SARS-CoV replication by an unknown mechanism of action [41].
Nsp14 and nsp16/nsp10 were first incubated with 100 µM of each candidate inhibitor in the presence of [3H]-AdoMet. N7- and 2′O-MTase activities were determined by quantification of methyl transfer to the GpppAC5 and 7MeGpppAC5 RNA substrates, respectively. As shown in Figure 5A, 10 out of 16 tested molecules barely inhibited the SARS-CoV MTases. Cap analogs (GpppA and 7MeGpppA) showed limited (50%) inhibition capacity on both SARS-CoV MTases. In contrast, we observed that AdoHcy, sinefungin and ATA efficiently inhibited both enzymes at 100 µM. The IC50 values of AdoHcy were 16 and 12 µM for the N7- and 2′O-MTase activities, respectively (Figure 5B) ten-fold higher than Ki values reported for VV:N7- and 2′O-MTases (1 and 0.5 µM, respectively, [47]). Sinefungin, a potent inhibitor of VV:N7- and 2′O-MTases with reported IC50 values of 12.0 and 39.5 nM, respectively [46], showed the most potent inhibition profile on nsp14 and nsp10/nsp16 with IC50 values of 496 nM and 736 nM, respectively (Figure 5C). These values are similar to the IC50 value reported for the inhibition of the 2′O-MTase activity of DV:NS5MTase (420 nM [49] and 630 nM [48]). The obtained IC50 values of ATA for SARS-CoV nsp14 and nsp10/nsp16 were 6.4 µM and 2.1 µM, respectively (Figure 5D). These results demonstrate that sensitive assays are now available to discover and characterize inhibitors of the SARS-CoV N7- and 2′O-MTases rendering low IC50 values of known AdoMet-dependent MTase inhibitors like AdoHcy and sinefungin. Moreover we have shown that nsp14 and nsp16 MTases are two putative targets of ATA, that was shown to inhibit SARS-CoV replication in infected cells [41].
Enzymatic activities postulated to be involved in the SARS-CoV RNA capping pathway were previously documented for the ORF1b-encoded replicase subunits nsp13 (RNA 5′- triphosphatase/helicase [13]) and nsp14 (N7-MTase [17]). Thus far, the predicted 2′O-MTase activity of nsp16 [3], [19] could only be verified for FCoV nsp16 [18]. Surprisingly, SARS-CoV nsp16 2′O-MTase failed to exhibit activity under a wide range of experimental conditions, including those used for FCoV nsp16 (not shown). In this study, we have characterized the MTase activities of both SARS-CoV nsp14 and nsp16, and in particular established that the in vitro activity of SARS-CoV nsp16 critically depends on the presence of nsp10. The latter had no known role or function, but was previously shown to interact with both MTase proteins nsp14 and nsp16 [35], [36].
Here, we show that the nsp14 AdoMet-dependent MTase activity can methylate GpppAC5 RNA, but not a 7MeGpppAC5 substrate, indicating that nsp14 specifically targets the N7-position of the guanine residue in the cap structure. This was verified using a substrate mimicking the capped 5′ end of the SARS-CoV genome. Nuclease P1 enzymatic digestion and TLC analysis confirmed the position of methylation by nsp14; and mutagenesis of a predicted AdoMet binding site residue abolished N7-MTase activity. We therefore conclude that nsp14 alone can act as an AdoMet-dependent MTase that specifically targets the N7-position of the cap structure, thus converting GpppRNA into 7MeGpppRNA. These results confirm and extend the recently described observations on the cap N7-MTase activity of SARS-CoV nsp14 in vitro and in a yeast-based complementation system [17].
In contrast to nsp14, bacterially expressed SARS-CoV nsp16 is less stable, reluctant to crystallization (not shown), and inactive on 7MeGpppRNA and GpppRNA in our in vitro assays. We report here that SARS-CoV nsp16 forms a complex with nsp10 that is endowed with robust and long-lived MTase activity. In contrast, FCoV nsp16 by itself was shown to possess 2′O-MTase activity under similar reaction conditions, but at much higher enzyme concentration (SARS-CoV: 200 nM; FCoV: 3 µM [18]). This suggests that FCoV nsp16 might also need FCoV nsp10 for its proper activation. As in the case of FCoV nsp16, SARS-CoV nsp16 in the presence of nsp10, specifically methylates capped RNAs carrying a methyl group at the N7-guanine position, allowing the conversion of cap-0 into cap-1 structures. Using 7MeGpppAG-RNA corresponding to the 5′ end of the SARS-CoV genome, we have confirmed that nsp10/nsp16 catalyzes the transfer of a methyl group from the AdoMet donor to the 2′O-position of the first nucleotide after the cap guanosine. Finally, the intrinsic nsp16 2′O-MTase activity was corroborated by mutagenesis of its predicted, canonical catalytic tetrad K46-D130-K170-E203 [3], [18], [19].
The nsp10 protein was previously proposed to play a role in viral RNA synthesis [11], [39], [40] and replicase polyprotein processing [40] on the basis of the analysis of MHV nsp10 mutants. In this work, we propose a new function for nsp10 as a regulator of an enzyme involved in the methylation of cap structures. Our observation seems not directly related to the phenotype previously described for nsp10 mutants [11], [39], [40]. Nevertheless, RNA cap methylation defects should limit RNA stability and may therefore contribute to a decrease in viral RNA synthesis observed in MHV mutants [11], [39], [40]. Here, we show that nsp10 itself is catalytically inert in methylation reactions (Figures 2, and 3) and that it forms a complex with nsp16 (Figure 2C). Interestingly, whereas at least a 10-fold molar excess of nsp10 is required for maximal stimulation of nsp16 (Figure S1B), quantification of the protein bands of nsp10 and nsp16 in the purified complex indicates a maximum ratio of 2.3 (Figure 2D). We assume that, under conditions of maximal stimulation, nearly all nsp16 molecules are associated with one or two nsp10 molecules. Considering that the reaction mixture at 50% stimulation contained 200 nM of nsp16 and around 400 nM of nsp10 (Figure S1B), the dissociation constant of the nsp10/nsp16 complex can roughly be estimated to be in the order of 400 nM for a 1∶1 complex or 200 nM for a 2∶1 complex. This is in agreement with a Kd of 250 nM determined by plasmon surface resonance analysis (Lecine P. personal communication).
What could be the mechanism of nsp16 activation by nsp10? Nsp10 may increase the stability of nsp16, stabilize the nsp16 RNA binding groove, contribute to RNA substrate binding and/or allosterically regulate its substrate affinity and activity. Similar activation of an MTase involved in the capping pathway was previously reported for the VV capping enzyme [54]. The catalytic efficiency of the N7-MTase domain of the VV:D1 protein was 370-fold stimulated by addition of an equimolar concentration of the small VV:D12 protein, which does not contain any catalytic residues [55]. Activation is achieved through increase of substrate and co-substrate affinity as well as of the turn over number. At the same time, VV:D12 exerts a stabilizing effect on VV:D1 [55]. The determination of the crystal structure of the protein complex VV:D1/D12 [56] revealed that the VV:D12 protein is structurally homologous to the cap 2′O-MTase of reovirus, with a truncation of the AdoMet binding domain. The SARS-CoV nsp10 crystal structure did not reveal any similarity to an MTase fold nor to any protein in the Protein Data Bank (PDB) [37], [38], but the activating effect of nsp10 on nsp16 might also be exerted on different levels via allosteric activation by increasing substrate affinity and/or turn over number, and/or by stabilization of nsp16.
The presence of two Zn fingers is a major structural feature of nsp10 that is likely related to its biological functions, since Zn fingers typically function as interaction modules binding to proteins, nucleic acid and small molecules [57]. Interestingly, several MTases have previously been shown to be regulated through specific interactions with Zn finger domains [58], [59]. Since the first Zn finger of nsp10 lies in a positively charged surface patch, it might be involved in the low affinity interaction of nsp10 with single- and double-strand RNA and DNA [37]. The affinity of nsp10 for single–stranded RNA appears too weak to explain a direct role in RNA recruitment for nsp16 [36]. We expect that SARS-CoV nsp16 contains a specific binding site for a cap-0 structure followed by a small stretch of 3 to 4 nucleotides as predicted for FCoV nsp16 from enzymatic assays [18]. The formation of the complex nsp10/nsp16 might provide a longer substrate-binding site and in that way enhance affinity of nsp16 for its RNA substrate. Nevertheless, given the fact that the full activation effect by nsp10 is also seen when short substrates containing the cap and 5 nucleotides are used, we surmise that extension of the RNA binding site is of minor importance. Further work is needed in order to understand the molecular basis of nsp16 activation by nsp10.
There are indications that CoV genomic RNAs and subgenomic RNAs carry a 5′-terminal cap-1 structure (see Introduction) and three of four putative cap-forming enzyme functions required to produce this structure have now been identified for SARS-CoV (nsp13, nsp14, and nsp16) [3], [17], [18]. The CoV cap structure methylation seems to follow the “classic” sequence of N7-methylation preceding the 2′O-methylation. The modular structure of two separate single-domain enzymes corresponds to the scenario in metazoan and plants [25], [26]. It contrasts to dsRNA reoviruses where one multi-domain protein contains two MTase domains [60] and to flaviviruses and negative-strand RNA ((-) RNA) vesiculoviruses where both MTase activites reside in a single domain of larger proteins and use a single active site [32], [61]. A characteristic feature of CoV MTases is that nsp14 recognizes non-methylated RNA cap exclusively, and nsp10/nsp16 recognizes N7-methylated RNA cap exclusively. In contrast, bi-functional MTases recognize both non-methylated and methylated cap structures with equal affinity [48], [62], [63]. Interestingly, the flaviviral N7-MTase activity is regulated by specific 5′-proximal viral RNA secondary structures and both N7- and 2′O-MTase activities seem to require in particular the terminal dinucleotide AG [32], [64]. Since the SARS-CoV nsp14 N7-MTase activity can complement N7-MTase defects in yeast, [17], it suggests that specific sequences and/or RNA structures are not required for this activity. This was confirmed in our in vitro assays, where both N7- and 2′O-methylation was observed using small GpppAC5 RNA substrates that do not correspond to the natural sequence present at the 5′ end of CoV mRNAs. Nevertheless, the CoV capping machinery is likely to act specifically on viral mRNA substrates, which present a common 5′-terminal leader sequence (72 nucleotides long in the case of SARS-CoV [4]). The mechanism to achieve this selectivity may depend on the GTase reaction or on the fine regulation of capping enzymes by protein-protein interactions within the replication and transcription complex. The regulation of the 2′O-MTase activity of nsp16 by the small nsp10 protein is clearly an original feature of CoV mRNA cap methylation.
Virally encoded RNA cap N7- and 2′O-MTase activities have been identified in various virus families, such as dsDNA poxviruses [65], [66], dsRNA reoviruses [67], [68], (-) RNA viruses such as vesicular stomatitis virus [61], and (+) RNA viruses like flaviviruses [33], [42], [69]. For many of them, including coronaviruses [17], [30], it has been shown that mutations abolishing the N7-MTase activity have a clear detrimental effect on replication [32], whereas 2′O-MTase knockouts exerted more moderate effects [32], [33]. These observations suggest that compounds specifically inhibiting cap MTases could be potent antiviral agents. Although some viral MTase inhibitors have also been reported to inhibit mammalian MTases [70], [71], sinefungin or other AdoMet analogs might have higher specificity towards viral MTases. Accordingly, it has been shown that sinefungin inhibits fungal mRNA cap N7-MTases with 5 to 10 times more potency than the human isoform [71].
Here, we report assays using GpppAC5 and 7MeGpppAC5 substrates that constitute sensitive screening tests for the identification and characterization of inhibitors of the N7- and 2′O-MTase activities, respectively, of SARS-CoV. We confirmed this by obtaining low IC50 values of known AdoMet-dependent MTase inhibitors AdoHcy and sinefungin. Furthermore, we found that ATA, a compound previously reported as a putative blocker of the catalytic site of NS5MTase of flaviviruses [52], and of SARS-CoV replication in infected cells [41], inhibited both SARS-CoV MTase activities with IC50 values of 2.1 and 6.4 µM, respectively. Thus, we propose that nsp14 and nsp16/nsp10 are two of the SARS-CoV targets of ATA leading, or at least contributing, to the inhibition of SARS-CoV replication.
In conclusion, our results identify and characterize the main viral protein players of SARS-CoV mRNA cap methylation. Its specificity and mechanistic originality remain unparalleled thus far and should open new avenues to investigate viral RNA capping, a field that is increasingly permeable to drug design projects.
AdoMet and cap analogs GpppA and 7MeGpppA were purchased from New England BioLabs. The compounds tested as MTase inhibitors were purchased from the following providers: Sigma-Aldrich: AdoHcy (adenosine-homocysteine), GTP, 7MeGTP, 3-deaza-adenosine, SIBA (5′-S-isobutylthio-5′-deoxyadenosine), sinefungin (adenosyl-ornithine), ribavirin (1-β-D-ribofuranosyl-1,2,4-triazole-3-carboxamide), MTA (5′-deoxy-5′-methylthio-adenosine), 2′,3′,5′-tri-O-acetyladenosine, S-5′-adenosyl-L-cysteine, 1,2-(((3-(4-methylpenyl)adamantine-1-yl)cabomoyl) and ATA (aurintricarboxylic acid); TriLink Biotechnologies: Ribavirin-triphosphate and ribavirin. EICAR-(5-Ethynyl-1-β-D-ribofuranosylimidazole-4-carboxamide)-triphosphate was a kind gift from P. Herdewijn (Leuven, Belgium). They were dissolved in H2O or DMSO as previously described [18], [52], [53] and ATA was dissolved in 0.1 M NaOH as described in [52]. Concentrations were set to 10 mM and compounds stored at -20°C. All radioactive reagents were purchased from Perkin Elmer.
The SARS-CoV nsp10-, nsp14-, and nsp16-coding sequences were amplified by RT-PCR from the genome of SARS-CoV Frankfurt-1 (accession number AY291315) as previously described [72]. The nsp10, nsp14, and nsp16 genes (encoding residues 4231–4369, 5903–6429, and 6776–7073 of replicase pp1ab) were cloned using Gateway technology (Invitrogen) into expression vector pDest14 (pDest14/6His-nsp10, pDest14/6His-nsp14 and pDest14/6His-nsp16) to produce recombinant proteins carrying an N-terminal His6-tag. The nsp14 gene was also cloned into the pTXB1 vector from the Impact kit (New England Biolabs) to generate the pTXB1-nsp14 plasmid that allows the expression of the nsp14 protein fused to the intein-chitin binding domain.
SARS-CoV nsp10/nsp16 complex was produced in E. coli in a bi-promotor expression plasmid kindly provided by Bruno Coutard (AFMB France). In this backbone, SARS CoV nsp10 can be expressed under a tet promoter and encodes a protein in fusion with a N-terminal strep tag, whereas nsp16 is expressed under a T7 promoter and encodes a protein in fusion with a N-terminal hexa-histidine tag. The single point mutants of pDest14/6His-nsp14 (the mutant numbering starts at the beginning of the nsp14 sequence; D90A & E92A, H268A, H268L, D273A and D331A) and the mutants of pDest14/6His-nsp16 (the mutant numbering starts at the beginning of the nsp16 sequence; K46A, D130A, K170A, E203A) were generated by PCR using the Quickchange site–directed mutagenesis kit (Stratagene), according to the manufacturer's instructions.
E. coli C41 (DE3) cells (Avidis SA, France), containing the pLysS plasmid (Novagen), were transformed with the various expression vectors and grown in 2YT medium containing ampicillin and chloramphenicol. Protein expression was induced by addition of IPTG to a final concentration of 500 µM (nsp10) or 50 µM (nsp14 and nsp16), when the OD600 nm value of the culture reached 0.5. Nsp10 expression was performed during 4 h at 37°C, whereas nsp14- and nsp16-expressing bacteria were incubated during 16 h at 17°C. Bacterial cell pellets were frozen and resuspended in lysis buffer (50 mM HEPES, pH 7.5, 300 mM NaCl, 5 mM MgSO4, 5 mM β-mercaptoethanol (only for nsp10) supplemented with 1 mM PMSF, 40 mM imidazole, 10 µg/ml DNase I, and 0.5% Triton X-100. After sonication and clarification, proteins were purified by two steps of chromatography except the nsp14 mutants, which were purified by one-step of IMAC (HisPurTM Cobalt Resin; Thermo Scientific) and concentrated on 50-kDa centrifugal filter units (Millipore).
Two-step purification of the His6-tagged proteins started with IMAC (HisPurTM Cobalt Resin; Thermo Scientific) eluting with lysis buffer supplemented with 250 mM imidazole. Protein fractions were then loaded on a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare), and eluted with 10 mM HEPES, pH 7.5, 150 mM NaCl. The protein fractions were concentrated to around 2 mg/ml and stored at −20°C in the presence of 50% glycerol.
The nsp14 protein expressed in fusion with the intein-chitin binding domain was purified on a chitin column using the IMPACT kit (New England Biolabs). The bacterial lysate was loaded onto the column, washed with 50 mM HEPES pH 7.5, 1 M NaCl and 0.5% Triton X-100. The column was then incubated in 50 mM HEPES pH 7.5, 500 mM NaCl, 50 mM DTT at 4°C for 48 hours in order to induce the intein cleavage. Next, the protein was eluted in 50 mM HEPES pH 7.5, 1 M NaCl buffer and subsequently purified on a HiLoad 16/60 Superdex 200 gel filtration column (GE Healthcare) as describe above. The identity of each of the purified proteins was confirmed by MALDI-TOF after trypsin digestion.
SARS-CoV nsp10/nsp16 co-expression was performed in E. coli strain C41 (DE3) (Avidis SA, France) transformed with the pLysS plasmid (Novagen). Cultures were grown at 37°C until the OD600nm reached 0.6. Expression was induced by adding 50 µM IPTG and 200 µg/L of anhydrotetracycline; then cells were incubated for 16 h at 24°C. Bacterial pellets were treated as given above and the soluble protein fraction incubated with Strep-Tactin sepharose (IBA Biotagnology). After 3 washes, bound proteins were eluted with 2.5 mM D-desthiobiotin in binding buffer. After analysing the purified protein complex by SDS-PAGE, the intensities of Coomassie-stained bands were quantified using ImageJ.
Short capped RNAs (7MeGpppAC5, GpppAC5, were synthesized in vitro using bacteriophage T7 DNA primase and were purified by high-performance liquid chromatography (HPLC) as previously described [69].
RNA substrate corresponding to the 5′-terminal 264 nucleotides of the SARS-CoV genome (5′ SARS-264) was prepared as follows. The 5′ UTR of the SARS-CoV genome Frankfurt-1 was amplified by PCR using the primers BamH1-T7phi2.5-5′SARS (s) (CGGGATCCCAGTAATACGACTCACTATTATATTAGGTTTTTACCTACCC) and EcoRI-SARS-264 (as) (GGAATTCCTTACCTTTCGGTCACAC) and cloned in the pUC18 (Fermentas) plasmid after BamHI/EcoRI restriction-ligation procedure. The T7 class II Φ2.5 promoter [43] was used (underlined in the primer) and the second nucleotide of the genome (U) was substituted by a G. The transcription matrix, was amplified by PCR (primers BamH1-T7phi2.5-5′SARS-AG (s) (CGGGATCCCAGTAATACGACTCACTATTAGATTAGGTTTTTACCTACCC) and SARS-264 (as) (CTTACCTTTCGGTCACAC)) and purified on agarose gel using the QIAquick gel extraction kit (Qiagen). The AG-SARS-264 RNA substrate was synthesized by in vitro transcription using the MEGAshortscript T7 RNA polymerase kit (Ambion). After DNase treatment (Ambion), and purification by RNeasy mini kit (Qiagen), the AG-SARS-264 RNA was incubated for 1 h at 37°C with the VV capping enzyme (ScriptCap m7G Capping kit, Epicentre Biotechologies) in a reaction volume of 20 µl, either in the absence or in the presence of AdoMet, according to the instructions of the manufacturer. 10 µCi [α-32P]-GTP (PerkinElmer, Boston, MA) and 0.05 units of inorganic pyrophosphatase (Sigma–Aldrich) were used. Radiolabeled capped RNAs GpppAG-SARS-264 and 7MeGpppAG-SARS-264 were then purified with the RNeasy mini kit (Qiagen).
MTase activity assays were performed in 40 mM Tris-HCl, pH 8.0, 5 mM DTT, 1 mM MgCl2 (only for nsp16/nsp10), 2 µM 7MeGpppAC5 or GpppAC5, 10 µM AdoMet, cand 0.03 µCi/µl [3H]AdoMet (GE Healthcare). In the standard assay, nsp10, nsp14, and nsp16 were added at final concentrations of 1.2 µM, 50 nM, and 200 nM, respectively. The final concentrations of nsp14 and nsp16 used in the assays were chosen so as to stay in the linear phase of product formation after a 1 h incubation when using GpppAC5 or 7MeGpppAC5 as substrates. A 6-fold molar excess of nsp10 over nsp16 was chosen to achieve about 75% of the maximal stimulation of 2′O-MTase activity. Under these conditions, the nsp14 and nsp10/nsp16 methylation reactions converted similar amounts of substrate after 1 h of reaction. No sign of protein inactivation was found up to the apparent end of the linear phase. Reaction mixtures were incubated at 30°C and stopped after the indicated times by a 10-fold dilution of the reaction mixture in 100 µM ice-cold AdoHcy. Samples were kept on ice and then transferred to glass-fiber filtermats (DEAE filtermat; Wallac) by a Filtermat Harvester (Packard Instruments). Filtermats were washed twice with 0.01 M ammonium formate, pH 8.0, twice with water, and once with ethanol, dried, and transferred into sample bags. Betaplate Scint (Wallac) scintillation fluid was added, and the methylation of RNA substrates was measured in counts per minute (cpm) by using a Wallac 1450 MicroBeta TriLux liquid scintillation counter. For inhibition assays, we set up the reactions as described above with 7MeGpppAC5 for nsp16 and GpppAC5 for nsp14 in the presence of 100 µM inhibitor candidate. Enzymes and RNA substrates were mixed with the inhibitor before the addition of AdoMet to start the reaction. The final concentration of DMSO in the reaction mixtures was below 5%, and control reactions were performed in presence of DMSO, which does not alter MTase activity. Reaction mixtures were incubated at 30°C for 4 h and analyzed by filter binding assay as described above. The IC50 (inhibitor concentration at 50% activity) value of AdoHcy, sinefungin and ATA were determined using Kaleidagraph. Data were adjusted to a logistic dose-response function, % activity = 100/(1+[I]/IC50)b, where b corresponds to the slope factor and [I] corresponds to the inhibitor concentration [73].
MTase activity assays were performed in 40 mM Tris-HCl, pH 8.0, 5 mM DTT, 1 mM MgCl2 (only for nsp16/nsp10), 50 µM AdoMet and 0.75 µM of capped AG-SARS-264 RNA at 30°C. The reaction was stopped after different reaction times by incubating samples for 5 min at 70°C. Samples were treated overnight with proteinase K (0.1 µg/µl, Invitrogen). Proteinase K was inactivated by addition of 5 mM PMSF; and the RNAs were subsequently digested for 4 h with nuclease P1 (0.05 U/µl, USBiological). Radiolabeled cap analog standards were produced by direct digestion of the substrates leading to GpppA and 7MeGpppA or digestion after methylation of the 2′O-position using VV 2′O-MTase (ScriptCap 2′O-methyltransferase kit, Epicentre Biotechnologies) leading to GpppA2'OMe and 7MeGpppA2'OMe. Digestion products were separated on polyethyleneimine cellulose thin-layer chromatography (TLC) plates (Macherey Nagel) using 0.45 M (NH4)2SO4 as mobile phase. After drying TLC plates, the caps released by nuclease P1 were visualized using a phosphorimager (Fluorescent Image Analyzer FLA3000 (Fuji)).
|
10.1371/journal.pcbi.1007015 | Evolution of major histocompatibility complex gene copy number | MHC genes, which code for proteins responsible for presenting pathogen-derived antigens to the host immune system, show remarkable copy-number variation both between and within species. However, the evolutionary forces driving this variation are poorly understood. Here, we use computer simulations to investigate whether evolution of the number of MHC variants in the genome can be shaped by the number of pathogen species the host population encounters (pathogen richness). Our model assumed that while increasing a range of pathogens recognised, expressing additional MHC variants also incurs costs such as an increased risk of autoimmunity. We found that pathogen richness selected for high MHC copy number only when the costs were low. Furthermore, the shape of the association was modified by the rate of pathogen evolution, with faster pathogen mutation rates selecting for increased host MHC copy number, but only when pathogen richness was low to moderate. Thus, taking into account factors other than pathogen richness may help explain wide variation between vertebrate species in the number of MHC genes. Within population, variation in the number of unique MHC variants carried by individuals (INV) was observed under most parameter combinations, except at low pathogen richness. This variance gave rise to positive correlations between INV and host immunocompetence (proportion of pathogens recognised). However, within-population variation in host immunocompetence declined with pathogen richness. Thus, counterintuitively, pathogens can contribute more to genetic variance for host fitness in species exposed to fewer pathogen species, with consequences to predictions from “Hamilton-Zuk” theory of sexual selection.
| Highly polymorphic genes of the Major Histocompatibility Complex (MHC) code for proteins responsible for presenting antigens to lymphocytes, thus initiating adaptive immune response. The polymorphism is driven by coevolution with parasites which are selected to evade recognition by MHC proteins. Expressing many MHC molecules could ensure that an individual could present antigens of most pathogen species encountered, but this comes at a cost, such as enhanced negative selection on lymphocytes leading to holes in T-cell receptor repertoire. Our simulations showed that evolution of the number of MHC genes in the genome is driven by a complex interaction between three factors we explored: pathogen richness, the intrinsic cost of expressing additional MHC variants, and pathogen mutation rate. In contrast to verbal arguments, our results indicate that pathogen richness does not always selects for MHC gene family expansion. Taking into account factors other than pathogen richness, in particular costs of expressing additional MHC variants which are still poorly understood, may help explain striking interspecific variation in the number of MHC genes. Counterintuitively, our results also demonstrated that opportunity for selection on immunocompetence should decrease with MHC gene family expansion.
| Major histocompatibility complex (MHC) genes code for proteins that present pathogen-derived oligopeptides (antigens) to T-cells, thus initiating an adaptive immune response. MHC genes are highly polymorphic, with dozens to hundreds of variants typically segregating in natural populations (reviewed in [1–3]). This extreme polymorphism is thought to result from balancing selection imposed by pathogenic organisms [4, 5], and broadly-reported associations between MHC variants and susceptibility to infection are consistent with the role of pathogens in driving MHC evolution (reviewed in [3]). Correlative and comparative analyses reported positive associations between parasite community richness and the number of MHC alleles within a population and strength of positive selection on MHC [6–9], providing further support for the role of parasites in driving MHC diversity. However, a meta-analysis based on 112 mammalian species showed that the signs, let alone the strength, of such associations may vary between taxa [10]. Interpretation of these differences is hindered by the scarcity of theoretical work exploring the impact of parasite richness on MHC diversity.
The majority of MHC research has focused on amino acid sequence polymorphism. However, an aspect of MHC diversity that has received less attention is the number of MHC variants carried by individuals (in this article, we use the term “variants” to describe individual MHC diversity, which is the number of distinct MHC molecules carried by an individual; we prefer this to the term “alleles” often used in MHC literature, as the variants are not alleles in a strict sense, being often distributed over several, functionally equivalent MHC loci). This number of variants carried by individuals is typically much lower than the number found in the population. For example, in humans, there are 6–7 classical MHC loci, allowing for up to 12–14 different variants in a fully heterozygous individual, while the number of currently identified MHC alleles summed across those loci in the human population exceeds 17 000 (IPD-IMGT/HLA Database (8), Release 3.30.0). Given that most alleles segregating in a population are thought to be maintained by selection from pathogens [3], such discrepancy suggests that any individual’s MHC diversity is unlikely to be sufficient to efficiently respond to the whole spectrum of pathogens a host may encounter. This implies there is some intrinsic cost of expressing too high MHC diversity. One possible mechanism constraining evolution of individual MHC diversity is the deletion of self-reacting T-cells, during negative selection in the thymus. This deletion is likely to intensify with an increased number of expressed MHC variants, leading to a sub-optimal T-cell repertoire[11, 12, but see 13 for criticism]. Recently, this mechanism has been supported by the study of Migalska et al. [14], who reported a negative correlation between the number of expressed MHC class I variants and T-cell receptor repertoire in the bank vole. However, alternative mechanisms [reviewed in 13], such as increased risk of autoimmunity or the necessity to reach a critical concentration of MHC–peptide ligands at the surface of antigen-presenting cells, can also play a role.
However, there are huge differences among species in the number of MHC loci, ranging from a very few e.g. in chicken [15] or humans [16] to dozens in some rodents [17] or passerine birds [18]. This raises the question: why should stabilizing selection on individual MHC diversity lead to such different numbers of MHC loci in different species? Answering this question may have broad implications beyond immunogenetics and host-parasite coevolution. For example, it has been suggested that the exceptional evolutionary success of passerines, a family comprising ca. 70% of all bird species, has been facilitated by their supreme immunity due to extremely high numbers of MHC genes they harbour [19]. Furthermore, evolution of individual MHC diversity may have implications for biological conservation [20] or speciation [21].
Similarly to population-level polymorphism, interspecific differences in MHC copy number could be due to differences in the richness of parasites the species is exposed to, although studies which have examined this association are rare. O’Connor et al [22] found that among passerines, the number of unique MHC variants carried by an individual (which should correlate with the number expressed MHC loci) is lower in the Palearctic compared to Africa, which they ascribed to higher parasite species richness in the latter region. Similarly, Minias et al. [23] showed that passerine MHC expansion is related to migratory behaviour, likely in response to larger diversity of pathogens encountered by migratory species. In a more direct approach, Eizaguirre et al. [24] compared two lakes and two river populations of three-spined sticklebacks Gasterosteus aculeatus and found that lake populations, which systematically harboured more parasite species, had more MHC variants per individual. Similarly, Radwan et al. [25] found a positive relationship between a proxy for parasite load and individual number of MHC variants in ornate dragon lizard Ctenophorus ornatus populations inhabiting isolated patches of natural habitat. Interestingly, the authors did not find a significant association of parasite load with population-level allelic MHC richness and speculated that evolution of high copy number may weaken the balancing selection that might otherwise maintain high polymorphism. Similarly, Dearborn et al. [26] argued that high individual MHC diversity which arose in Leach’s storm-petrels, Oceanodroma leucorhoa by duplication followed by diversification of MHC class II genes should weaken advantage of heterozygosity at MHC. However, there is a lack of theoretical work on how parasite richness simultaneously affects MHC allelic richness and the number of MHC loci.
Here, we aim to fill this gap using computer simulations based on a framework that has previously been shown to be effective in recovering some of the most important features of MHC evolution, such as high polymorphism, frequency-dependent selection, heterozygote advantage and positive selection [27, 28]. The model simulates interactions of MHC molecules and antigens produced by pathogens by matching strings of bits, which can mutate both in hosts and in parasites [27, 28]. Here, we introduce a new feature to the framework to allow duplication and deletion of MHC genes. We then investigate how the number of pathogen species infecting a host affects the evolution of MHC allelic richness and the number of MHC loci.
Pathogen richness affected the number of unique MHC variants per individual (individual number of variants, INV henceforth) in a complex way, shaped by significant interactions with pathogen evolution rate and with the intrinsic cost of expressing more MHC variants (described by cost parameter α) (Table 1). Parasite species richness clearly increased INV at lower α, but at higher α there was little change in the INV across the levels of pathogen richness (Fig 1). There was also a significant pathogen richness × pathogen mutation rate interaction (Table 1), with the positive effect of higher pathogen mutation rate observed at low pathogen richness, but declining to zero as pathogen richness increased (Fig 1).
The selection acting on host MHC genotypes, as measured by coefficient of variation (CV) in host fitness (which in our simulation was determined solely by host immunocompetence, i.e. the proportion of pathogens recognized), was shaped by the significant interaction between pathogen richness and mutation rate (Table 2). CV in host fitness was much higher at higher pathogen mutation rate when pathogen species number was low (Fig 2). However, the differences between mutation rates declined, as did CV itself, with an increase in pathogen richness.
We observed considerable within-population variation in the INV under most scenarios, except when the number of pathogens was very low (S4 Fig). The slopes of the relationship between the number of pathogens presented to the immune system and INV increased with pathogen richness, but slopes were generally low at higher pathogen mutation rate (Fig 3).
The number of MHC variants segregating in a host population (PNV henceforth) was driven by the significant three-way interaction between α, pathogen richness and mutation rate (Table 3). PNV generally increased with pathogen richness (Fig 4, Table 3), but the increase was lower at α = 0.08. High pathogen mutation rate increased PNV only at the combination of low α and high parasite richness (Fig 4). Interestingly, at low α, PNV largely reflected INV, whereas at high α PNV increased (Fig 4) despite that INV did not (Fig 1; see S5 Fig for correlation between INV and PNV).
Our model showed that under the Red Queen-like dynamics of MHC evolution, evolution of INV is shaped by a complex interaction of several factors, including pathogen richness, pathogen mutation rate, and the intrinsic cost of expressing many MHC molecules. Verbal arguments [e.g. 8, 22, 23] assumed that INV should generally increase with the number of pathogen species. In our simulations, this was the case only under some parameter combinations, and the form of the relationship depended both on the intrinsic costs of expressing additional MHC variants and on pathogen mutation rate. INV consistently increased across the investigated range of parasite species when the intrinsic cost of large MHC repertoire was small. However, with higher values for the cost factor (α), we did not observe such an increase. This shows that high pathogen richness will not necessarily lead to the evolutionary expansion of MHC gene family. Little is known about the nature of the intrinsic costs of MHC expansion, and even less on how taxa differ in this respect, and therefore we have not modelled any particular mechanism underlying these costs in our simulations. The prevalent explanation is that high MHC diversity increases negative selection of self-reactive T-cell receptors [11, 12], impairing efficiency of immune response. This scenario has recently been supported in bank voles, where TCR repertoire has been demonstrated to decrease with the number of MHC class II variants [14]. Under such a scenario, intermediate numbers of MHC variants should result in the most efficient clearing of infections, as has been observed in some empirical studies, including bank voles [29–31]. However, several studies utilising extensive variation in INV present in passerine birds have observed either no such a relationship, or negative associations between INV and infection [e.g. 32, 33–35]. This suggests that the nature of the intrinsic costs of expressing many MHC variants may differ between passerines and mammals. One possibility is that expressing too many MHC variants does not compromise passerine TCR repertoire in a way similar to that observed in bank voles [14], allowing rapid expansion of MHC gene family (compare Fig 1). The study of TCR repertoires in birds, and the way they are shaped by MHC, emerges as an attractive target for future studies.
More generally, understanding inter-specific difference in INV will require extensive study of intrinsic costs of expressing additional MHC variants across vertebrate taxa. Our model indicates that higher pathogen richness is unlikely to explain a spectacular expansions of MHC gene family, such as those observed among passerines. Ancestrally, birds have been characterised by a small number of MHC genes, which is still observed in non-passerines [23]. Our results suggest that expansion to dozens of MHC loci observed among some passerine superfamilies (Sylvioidea, Passeroidea and Muscicapoidea [23]) would require the number of pathogen in these lineages to be manifold higher compared to basal groups (compare Fig 1), which does not appear biologically feasible.
Another factor which influenced the evolution of INV in our simulations was pathogen mutation rate, the effect of which was most pronounced at low pathogen species numbers (Fig 1). This pattern was mirrored by variance in host fitness (measured as CV), which was the highest for high pathogen mutation rate combined with low pathogen richness (Fig 2). Host haplotypes with more MHC variants should be more likely to carry a variant conferring resistance to a parasite, but efficient evasion of MHC-recognition by fast-evolving pathogens could weaken association between INV and pathogen recognition, consistent with our results (Fig 3). Still, efficient parasite evasion should favour novel MHC variants [28], and such variants are more likely to arise when the number of copies in the genome is high. When average number of MHC variants is already high, however, possessing an extra MCH copy provides relatively smaller advantage in terms of potential for beneficial mutation. This may explain why the effect of pathogen evolution rate on INV was observed only at low pathogen richness (where INV was relatively low).
Similarly, high CV in host fitness at low numbers of pathogen species likely results from the fact that a haplotype that is resistant to a prevalent pathogen genotype (of any species) will gain considerable advantage, whereas with many pathogen species resistance to any given pathogen contributes relatively less to fitness. This may explain why CV in host fitness declined with pathogen richness, which may have interesting implications for predictions stemming from Hamilton and Zuk’s (1982) theory of sexual selection. This theory poses that costly epigamic traits, such as long feathers or bright colouration, are subject to mating preferences because they reflect the genetic aspect of resistance to pathogens. At the interspecific level, the Hamilton-Zuk hypothesis predicts that higher risk of parasite infection should enhance sexual selection for extreme values of such epigamic traits, because of increasing contribution of pathogens to genetic variance in fitness (Hamilton and Zuk 1982, Berlanger and Zuk 2014). Paradoxically, our results indicate that while host genetic diversity for resistance (measured by the number of MHC variants segregating in populations) increased, the variance in host fitness decreased. Our results thus indicate that if the number of pathogen species attacking the host is used as a measure of selective pressure from pathogens, the predicted relationship with an elaboration of epigamic traits might be counter-intuitive.
INV was positively correlated with pathogen recognition ability (Fig 3), as assumed by models of copy-number evolution [11, 12]. Nevertheless, our simulations suggest no such association should be expected when the number of MHC variants in the species is low (Fig 3). Indeed, in root voles Microtus oeconomus and guppy fish Poecilia reticulata, both characterised by a low to a moderate number of MHC loci (1–3), possessing particular variants has been shown to be more important than the number of expressed MHC loci [36, 37].
More interestingly, INV was not a good predictor of pathogen recognition efficiency when parasites evolved fast (Fig 3). As discussed above, fast-evolving parasites are more effective in evading recognition by MHC haplotypes prevalent in a population than slow-evolving ones. In consequence, when parasites evolve fast, possessing a rare-but-resistant MHC variant should have more of an effect on resistance than possessing many variants.
Our simulations revealed tight associations between PNV and INV, but the slope of the associations depends on the intrinsic cost of expressing additional variants (S5 Fig). At high α, where increase in pathogen richness does not result in a consistent increase in INV, PNV nevertheless increases, resulting in slope >1. At low α, at which INV is more free to evolve, PNV largely reflects INV, which implies that when selection from many parasites favours gene duplication, per-locus polymorphism may change very little. Our results may explain the findings of comparative analyses showing that high pathogen richness is sometimes not found to be associated with MHC allelic richness (a per-locus measure of variation), despite its effect on the rate of molecular evolution at MHC antigen binding sites [8, 9]. Two recent comparative studies [22, 23] demonstrated that among passerines, individual number of MHC variants decreases with such likely correlates of pathogen richness as latitude or migratory behaviour (although we know of no work directly linking INV to parasite richness). It would be interesting to see if INV could explain PNV in this system, as predicted by our model.
Concluding, our study showed that in general, pathogen richness selects for expansion of MHC gene family, but is unlikely to explain striking inter-specific differences in the number of MHC genes. The latter can be can be modulated both by the rate at which parasites evolve and, probably more critically, on the strength of mechanisms selecting against the high number of copies in the genome. These mechanisms are not well understood, but warrant investigation as potential causal factors underlying differences in MHC genes family sizes between species. In species which evolved high INV under selective pressure from many pathogen species, within population variation in INV can nevertheless be maintained. Despite high variation in INV, host variance in immunocompetence should, according to our model results, be lower in species experiencing selection pressure from higher diversity of parasites.
The model is based on an approach first used by Borghans et al. [27], which simulated interactions between the peptide-binding grooves of MHC molecules and antigens derived from pathogens by aligning two strings of zeros and ones (bitstrings). In our model, each MHC molecule was represented as a 16-bit-long string, which can be thought of as a representation of the amino acids that form pockets implicated in the specificity of antigen binding (there are 12–23 polymorphic sites contacting antigens in human MHC molecules [38]). A pathogen was represented by a single 6000-bit long antigenic molecule, which was tested for a match with host MHC at all possible 16-bit epitopes which could be produced from the antigenic molecule. Antigen binding occurred when there was a match in all position of the bit strings representing the peptide bindig groove of MHC molecules and an epitope (S1 Fig). Utilising 16 bits, we could simulate 65,536 (216) MHC epitopes. The probability of finding a random 16-bit sub-string (epitope) in a random 6000 bit antigen was approximately 0.084, a number corresponding to the empirical estimates of an MHC molecule binding a random epitope produced by viral pathogens [39, 40]. The way we simulated antigens differed from that in Borghans et al. [27] and earlier adaptations of their approach [e.g. 28, 41] in which a single parasite was represented by a set of 20 independent, 16-bit-long antigens, and 7 matched bits were used as a threshold for pathogen recognition. The rationale for simulating a long antigenic molecule and a higher threshold number of matching bits was that it reduced the number of recognition motifs shared between pathogen species, and, additionally, it facilitated further diversification of species-specific motifs by conserving some of them in a species-specific manner (see below). Nevertheless, the probability of binding a random antigen produced by a given pathogen remained broadly consistent with those earlier studies [27, 28, 41].
Hosts co-evolved with a variable number (2–64) of haploid pathogen species, which, to simplify simulations, had population sizes equal to that of their hosts [as in previous studies, e.g. 28]. Instead of simulating larger pathogen populations (as would have been observed in nature), higher probability of a mutation in large populations was emulated by a higher pathogen mutation rate. There were 10 pathogen generations per one host generation to reflect the fact that pathogens typically have faster generation times than hosts.
The fitness of pathogen haplotypes was proportional to the number of hosts a pathogen successfully infected, and host fitness was proportional to the number of pathogens recognized (see below for details). The next generation of both hosts and pathogens was drawn in proportion to their fitness. The algorithm described above effectively simulates a host-parasite co-evolution system with Red Queen dynamics [see 28 for more details].
MHC genes (i.e. 16 bit-long strings) were located on one diploid pair of host chromosomes. The size of the host population was fixed at 1000 individuals. These individuals were exposed to one, randomly chosen individual of each pathogen species. If the infection was successful (i.e. the pathogen was not recognized by any of the host’s MHC genes), the parasite clone could evolve in the host for 10 generations, ecologically excluding infections by other clones of the same species. If the infection was unsuccessful, a new, randomly selected individual attempted an infection in the next pathogen generation; if successful, this pathogen would be allowed to reproduce until 10 pathogen generations were completed. After 10 pathogen generation passed, host fitness was determined. The fitness was proportional to the number of pathogens presented by the host, but we additionally introduced a cost of having additional MHC variants (see below). The cost was introduced to reflect various mechanisms thought to counteract unconstrained expansion of MHC region [11, 13]. Our preliminary analyses indicated that the number of MHC loci rapidly increased and did not stabilise even after thousands of generations if no cost was introduced. The host fitness function was calculated according to the equation:
fhost=P⋅e−(αN)2
(1)
where P is the number of pathogen species a host recognized (thus avoiding infection), N is the number of unique MHC variants in the host’s genome and α is the cost factor. The cost factor α was selected to achieve a realistic number of unique MHC types in an individual (i.e. from a few to few dozens).
After interactions with pathogens (across 10 pathogen generation cycles), hosts reproduced with probability proportional to their fitness. We have not modelled separate sexes (i.e. our hosts were equivalent to out-crossing hermaphrodites). During reproduction, each of the diploid mates provided one chromosome (selected randomly) to the resulting progeny. Each mating resulted in one offspring, but individuals could be selected for mating more than once (which was more likely for high fitness hosts), and random mating pair selection was repeated until the size of the host population NH was restored.
Host chromosomes could undergo two types of mutations: micromutations within the 16-bit string, and copy number mutations. Micromutations were represented by a flip of a single bit with a given probability. This can be thought of as a non-synonymous substitution in an antigen binding site of MHC molecule, which could occur as a non-synonymous mutation, or micro-recombination (the latter may be the predominant mode of mutation at human MHC [42]). For the sake of consistency with previous simulation studies, in which mutation rates were given as the probability of change in MHC molecule as a whole (replacement of old MHC with a new one), we report the mutation rate per MHC molecule (i.e. 16 bit string), which translates into per bit rate according to the equation:
μbit=1−(1−μMHC)1/16
(2)
where μMHC is the mutation rate per MHC peptide, μbit is the mutation rate per single bit in the MHC PBR (see also S1 Appendix in [28]. We used a host mutation rate of 10−4 per MHC molecule (or 6.25 × 10−6 per PBR), which appears realistic based on published literature [42]. We also simulated “macromutations” in MHC, which could be thought of as recombination or gene conversion of large fragments of an exon coding for peptide binding groove. Following earlier work [27], we simulated macromutations by producing random strings of bits. However, mutation mode have not qualitatively affect our results (S6 Fig), therefore in the main text we only present results for micromuations.
Copy number of MHC genes could change via duplication or deletion. Duplication was modelled by adding a new copy of the original sequence on the same chromosome, and during deletion, a gene disappeared from the chromosome. However, the algorithm did not allow the number of MHC loci to go below 1 per chromosome. Each gene could be duplicated with probability 10−3 and deleted with probability 10−3, which is higher than direct empirical estimates for large structural variant indels in human genomes [43], but was the minimum necessary for the number of copies to stabilise within realistic computing time. Neo- or sub- functionalization of duplicated loci could occur by mutations described above.
We simulated a variable number of haploid pathogen species, with the population size of each species equal to that of the host. A species was initiated as a single antigen, and thus individuals were sharing the evolutionary origin and history within species, but separate species were initialized independently. Because the possible number of distinct 6000 bit antigens is very large (~1.5 × 101806), pathogen species showed little overlap in their antigenic profiles (S2 Fig; the probability that a random 16-bit-long sub-string will be present in both of two random and independent 6000-bit-long strings equals to ~0.0842). We trialled a variant of the simulations in which each pathogen species had a randomly-assigned, species-specific 33% of bits conserved, but this did not result in a different interpretation, and we do not report results from this version. Pathogen haplotypes were selected for reproduction with probability proportional to the number of hosts they had infected. During each of 10 pathogen generations, every host was matched with a randomly selected individual of each pathogen species and the outcome of the infection was evaluated according to the bit-matching rules described above. A pathogen species could infect an individual host only once per host generation. The successful pathogens reproduced parthenogenetically by producing 'clonal' progeny. The progeny could mutate by changing a single bit to the opposite before advancing to the next round of infections. To examine the role of pathogen evolution rate on our results, we simulated two pathogen mutation rates: 10−5 and 5 × 10−5. These values resulted in the host-parasite coevolution we sought to produce in our simulations. Exploratory analysis showed that at lower pathogen mutation rates than reported above, pathogens were unable to adapt to host genotypes fast enough, whereas at higher mutation rate fitness differences between host genotypes were small, precluding effective co-evolution [see 44]. For comparison, the influenza virus NS gene mutates at a rate of 1.5 × 10−5 [45].
The model’s program was written in C++14 language, which generates a number of text files of simulation results that were then analysed and plotted using Python scripts. The general scheme of the algorithm is shown in S3 Fig. The source code and its documentation can be obtained from https://github.com/pbentkowski/MHC_Evolution. Summaries of the model’s parameters and their values are given in Table 4. Each combination of parameters was run 20 times, except for the most computationally demanding simulations with 64 pathogens, which were run 10 times.
For evaluation purposes, we considered the last 1250 host generation when the dynamics of the host-parasite co-evolution stabilised in term of the numbers of MHC variants in both populations and individuals. For that period and for each run, we calculated mean PNV and mean CV in host fitness (pathogen presentation ability) by averaging it over 1250 latest host generations. To calculate mean INV, we first averaged across individuals at a given time step, and then took the averaged simulated values across 1250 latest host generations. Coefficients of regression of INV on pathogen presentation ability (Fig 3) was based on a population 'snapshot' at host generation #5000 (the last one), when we recorded detailed information on each host (what genes they had, what pathogen species they presented). These data are available in Supplementary File 1. Results were analysed with linear models, with an average INV, PNV or CV in host fitness as a response variable, and α, pathogen mutation rate and pathogen richness, and their interactions, as fixed factors. Statistical analyses were done in R 3.4.2.[46].
|
10.1371/journal.pgen.1007395 | Distinctive types of postzygotic single-nucleotide mosaicisms in healthy individuals revealed by genome-wide profiling of multiple organs | Postzygotic single-nucleotide mosaicisms (pSNMs) have been extensively studied in tumors and are known to play critical roles in tumorigenesis. However, the patterns and origin of pSNMs in normal organs of healthy humans remain largely unknown. Using whole-genome sequencing and ultra-deep amplicon re-sequencing, we identified and validated 164 pSNMs from 27 postmortem organ samples obtained from five healthy donors. The mutant allele fractions ranged from 1.0% to 29.7%. Inter- and intra-organ comparison revealed two distinctive types of pSNMs, with about half originating during early embryogenesis (embryonic pSNMs) and the remaining more likely to result from clonal expansion events that had occurred more recently (clonal expansion pSNMs). Compared to clonal expansion pSNMs, embryonic pSNMs had higher proportion of C>T mutations with elevated mutation rate at CpG sites. We observed differences in replication timing between these two types of pSNMs, with embryonic and clonal expansion pSNMs enriched in early- and late-replicating regions, respectively. An increased number of embryonic pSNMs were located in open chromatin states and topologically associating domains that transcribed embryonically. Our findings provide new insights into the origin and spatial distribution of postzygotic mosaicism during normal human development.
| Genomic mosaicism led by postzygotic mutation is the major cause of cancers and many non-cancer developmental disorders. Theoretically, postzygotic mutations should be accumulated during the developmental process of healthy individuals, but the genome-wide characterization of postzygotic mosaicisms across many organ types of the same individual remained limited. In this study, we identified and validated two types of postzygotic mosaicism from the whole-genomes of 27 organs obtained from five healthy donors. We further found that the postzygotic mosaicisms arising during early embryogenesis and later clonal expansion events show distinct genomic patterns in mutation spectrum, replication timing, and chromatin status.
| Postzygotic mutations refer to DNA changes arising after the formation of the zygote that lead to genomic mosaicisms in a single individual [1, 2]. Unlike de novo or inherited germline mutations, postzygotic mutations only affect a fraction of cells in multicellular organisms, and individuals carrying a functional mosaic mutation typically exhibit a milder phenotype [3–5]. The roles of postzygotic single-nucleotide mosaicisms (pSNMs) have been demonstrated in numerous cancers [6, 7] and various types of developmental disorders, including malformations [8, 9] and autism [10, 11]. We and another research group have reported the first genome-wide identification and characterization of pSNMs from the peripheral blood samples of healthy individuals [12, 13]. More recently, the accumulation of postzygotic mutations during aging process has been reported in blood or brain samples [14–17]. Yadav et al. studied pSNMs in apparently benign tissue samples obtained from cancer patients [18], but the contribution of pre-cancerous mutations could not be completely ruled out and the study was restricted to exonic regions. As such, the occurrence and genomic pattern of pSNMs in normal tissues of healthy individuals remains under-investigated.
It has been reported that cancer genomes have distinct mutational signatures resulting predominantly from exposure to mutagenic agents and dysfunction of the DNA repair machinery [19]. Additional genomic factors, such as replication timing and chromatin status, could also impact the distribution of pSNMs in cancer genomes [20–22]. Whether and how these genomic factors might contribute to the genomic distribution of pSNMs in organs of healthy individuals remains largely unexplored [23]. Tumorigenesis has been considered as an evolutionary process in which tumor cells with increased fitness will proliferate faster than normal cells and lead to the clonal expansion of tumor cell population in a specific organ [24, 25]. Although such events of clonal expansion have been previously reported in apparently normal skin and blood samples [17, 26], it remains unclear whether clonal expansion plays a role in other non-cancer tissue types. Understanding the origin and spatial distribution of pSNMs in normal tissues of healthy individuals could provide an important baseline for interpreting their contributions to disease states [27].
The next-generation sequencing technologies (NGS) have greatly advanced the study of pSNMs [28]. Sequencing the genomes of single cells after whole-genome amplification or in vivo clonal proliferation have been applied to the study of pSNM profiles of normal human cells, including germ cells [29], adult stem cells [30], and neurons [31]. Typically, tens or hundreds of cells from each sample need to be sequenced to identify and quantify pSNMs, which tends to increase the cost [32]. The inaccurate process of whole-genome amplification in single-cell sequencing makes it difficult to distinguish real pSNMs from technical artifacts, and the challenge of rigorously validating the pSNMs in a cell that have been already amplified aggravates the uncertainties [33, 34]. Bulk sequencing is potentially a reliable and cost-effective alternative that, importantly, allows for rigorous validations of pSNMs [23]. Utilizing bulk whole-genome sequencing (WGS) and ultra-deep amplicon re-sequencing, this current study identified and validated pSNMs from 27 different organ samples obtained from five healthy donors and investigated the origin and spatial distribution of pSNMs in the developmental process of these healthy individuals.
Postmortem organ samples derived from five healthy Asian donors (age 20–45 yr) were obtained from BioServe, including a total of 27 organ samples from brain, liver, colon, skin, artery, breast, ovary, and prostate (Table 1). The donors died from motor vehicle accidents and were not known to be affected by any types of cancer or other overgrowth disorders. The samples were sequenced using an Illumina HiSeq X Ten sequencing platform with an average depth of 114-150X (Table 1).
It is expected that many postzygotic mutations occurring at an early stage of embryogenesis may be shared between two or more organs from one individual [13, 27]. Thus, conventional mutation callers, which require matched negative control samples for comparison, would likely miss these mutations. We had previously developed MosaicHunter [35], a bioinformatics pipeline that can detect pSNMs without the need of control sample obtained from the same individual. MosaicHunter incorporated a Bayesian genotyper to distinguish pSNMs from germline variants and base-calling errors and a series of stringent filters to remove systematic errors. Using the Bayesian genotyper, we calculated the posterior probability of mosaic genotype versus three germline genotypes across all the genomic sites with at least 5% mutant allele fraction (the fraction of reads supporting the mutant allele) and 3 or more reads supporting the mutant allele. As a result, we identified a total of 251 candidate pSNMs in the 27 samples from the five donors; among them 41 pSNMs were found in more than one sample from the same donor (Table 1).
Next, we validated the pSNMs and quantified their minor allele fractions in all of the organ samples (Methods). We used an amplicon-based ultra-deep resequencing method, PASM (PGM Amplicon Sequencing of Mosaicism), which we had previously developed and benchmarked [5]. Of the 251 candidate pSNM sites, 27 were excluded due to failure to design amplicon primers or to get enough sequencing depth in the negative controls. For the remaining 224 sites, the average sequencing depth of the amplicons was greater than 4000X per sample (S1 Fig). The peripheral blood samples of two unrelated healthy Asians (ACC1 and ACC4) served as negative controls. A pSNM was considered validated only if the mutant allele was detected in an organ sample in a mosaic state but undetectable in both the negative controls. Three sites with abnormal copy numbers estimated from the WGS data were further excluded (S1 Table and Methods). In summary, we successfully validated 164 pSNMs in these five donors, with an overall validation rate of 73.2% (Table 1). The full list of the 164 validated pSNMs was described in S2 Table, which was used in the following analyses.
The validated pSNMs were located in 21 autosomes and the X chromosome (Fig 1A). We calculated the genomic distance between nearby pSNMs and found no significant difference between the observed and expected distances if pSNMs were uniformly distributed along the human genome (Kolmogorov-Smirnov test, P-value > 0.05), indicating that there was no observable clustering of postzygotic mutations in healthy individuals. The minor allele fractions estimated by PASM ranged from 1.0% to 29.7%, significantly correlating with the fractions estimated by WGS (Fig 1B; Pearson’s r = 0.89 and P-value < 2.2×10−16). The allele fraction of each pSNM varied across the different organs from the same donor (S2–S6 Figs).
Based on the presence or absence of the validated pSNMs in the organ samples from an individual donor, we grouped the pSNMs into two categories: 60 were present in two or more organ samples of the same donor (27 of which were globally present in all the sequenced organs of the donor) and 104 were uniquely present in a single organ. Given the low postzygotic mutation rate in healthy individuals [36], it was unlikely that multiple postzygotic mutation events involving the same nucleotide alteration occurred independently within one individual. It was more likely that the pSNMs shared by more than one organ resulted from mutation events that had occurred at early developmental stages, and the mutant alleles were passed on to cell lineages of more than one organ type. Comparison of the minor allele fractions in the two categories of pSNMs supported this hypothesis. As shown in Fig 2A, the allele fraction of pSNMs shared by more than one organ was significantly higher than that of pSNMs unique to only one organ (Wilcoxon rank-sum test, P-value = 1.2×10−3). In particular, 40% (24 out of 60) of the pSNMs shared by more than one organ had allele fractions greater than 1/16, suggesting that they might have originated during the first few cell divisions of embryogenesis [37]. We refer to these pSNMs shared by more than one organ as “embryonic pSNMs” in the following analyses. On average, we identified 4.6~14.5 embryonic pSNMs from each organ of the five individuals, and the occurrence rate was similar across different organs (Fig 2B). We further compared the allele fractions across multiple organs of the same individual, and found that more than 95% of the embryonic pSNMs showed <5% standard deviation of allele fraction (S2 Table), indicating no dramatic allele fraction change for embryonic pSNMs.
Close inspection of the pSNMs unique to only one organ revealed a distinctive type of pSNMs. Two organs had a dramatic excess of organ-unique pSNMs compared to other organs (Fig 2B). Specifically, the liver sample of BBLD1005 and the breast sample of BBL11121 carried 42 and 32 organ-unique pSNMs, respectively, compared to an average of 1.1 organ-unique pSNMs for the other organ samples. This suggested that the majority of pSNMs in these two organs might originate organ-specifically after embryogenesis [30].
To further investigate these excessive organ-unique pSNMs in these two organs, we sampled three additional adjacent samples from each organ with varying physical distances to the original samples used for WGS (S7 Fig) and applied PASM to profile the allele fractions of validated pSNMs (S8 and S9 Figs). While 8 of the 9 (88.9%) embryonic pSNMs could be detected in all three intra-organ samples (Fig 2C and 2D), consistent with our prediction that these mutations occurred early in embryogenesis, the organ-unique pSNMs manifested with a distinct intra-organ pattern. In BBLD1005, 22 out of 42 (52%) liver-unique pSNMs identified in the original sample (liver #9) were also detected in the physically closest sample (liver #8), whereas only one liver-unique pSNM was detected in the two samples further away (liver #2 and #5) (Fig 2C). Given that the physical distance between liver #8 and #9 was about 0.5 cm and the distance between liver #2/#5 and liver #9 was approximately 3.5 and 2 cm, respectively, these results suggested that the majority of liver-unique pSNMs were locally restricted to a small volume of liver cells. A similar observation was made in the breast samples of BBL11121 that breast #7 shared more pSNMs to breast #9 than breast #2 and #5 (Fig 2D). We reconstructed the inter-sample similarity using the minor allele fractions of pSNMs, and indeed the originally-sequenced liver or breast samples shared the largest similarity to their physically nearest samples (Fig 2E and 2F).
Analysis of minor allele fractions of the liver- and breast-unique pSNMs revealed a single narrow peak for each organ sample (S10 Fig), with an average of 3.1% and 4.2%, respectively. Considering that such pSNMs were restricted to a small region within the organ, the narrow peaks likely resulted from clonal expansion events during the process of organ self-renewal that generated a sub-population of cells carrying postzygotic mutations large enough to be detected in bulk sequencing [23]. We refer to these pSNMs as “clonal expansion pSNMs” in the following analyses. Our results demonstrated the presence of clonal expansions in various types of non-cancer organs and highlighted clonal expansion as one of the major sources of pSNMs in clinically unremarkable individuals.
If the embryonic pSNMs arose from early mutations during embryogenesis and the clonal expansion pSNMs arose from more recent mutations during organ self-renewal, they may present different mutational characteristics. To explore this possibility, we compared these mutations in terms of mutation spectrum, replication timing, and chromatin status.
We first studied the mutation spectrum of the two types of pSNMs identified. For embryonic pSNMs, C>T mutations were the most predominant type (65.0%), with a significant elevated mutation rate at CpG sites vs non-CpG sites (Proportion Z-test, P-value < 2.2×10−16, Fig 3A). The enrichment of C>T mutation at CpG sites could be explained by the spontaneous deamination of 5-methylcytosines (5mC) [20], which has also been reported as one of the most common signatures in cancers [38]. The predominant C>T mutation at CpG sites for embryonic pSNMs were consistent with previous studies of early pSNMs in human [14, 39] and mouse [40]. On the contrary, we observed predominant C>A (39.5%) and T>C (42.4%) mutations for the clonal expansion pSNMs identified in BBLD1005’s liver and BBL11121’s breast samples, respectively (Fig 3B and 3C). Oxidative DNA damage was one of the major cause for C>A mutations [22], and the higher proportion of C>A mutation in the liver sample could be explained by the accumulated oxidative stress of hepatocytes.
Previous studies had reported elevated rates of germline and cancer-related somatic mutations in late-replicating regions [41, 42]. Using data from the replication timing profile of lymphoblastoid cell-lines [43], we observed significantly different distributions of replication timing between embryonic and clonal expansion pSNMs (Wilcoxon rank-sum test, P-value = 9.7×10−3; Fig 3D). Clonal expansion pSNMs were significantly enriched in late-replication regions (Permutation test, P-value = 0.006), similar to previous reports of germline and cancer-related somatic mutations, while embryonic pSNMs were significantly enriched in genomic regions that replicated earlier (Permutation test, P-value = 0.026). Embryonic pSNMs with a wide range of allele fractions contributed to the early-replication enrichment (S11 Fig), suggesting that the enrichment was not caused by a small number of outliers. This bimodal distribution was confirmed using the replication timing profiles from five other cell-lines: GM12878, K562, HeLa-S1, HepG2, and HUVEC (Wilcoxon rank-sum test, P-value < 0.05). We further confirmed our finding by using the pSNMs identified from the single-clone sequencing of neuronal progenitor cells [14], where the mutations which were shared by other brain regions and non-brain tissues were significantly enriched in early-replicating regions than those specifically present in the clone of neuronal progenitor cells (Wilcoxon rank-sum test, P-value = 0.028). The distinct pattern of replication timing between the two types of pSNMs might reflect different mutational effects of replication timing during different stages of human development.
Last but not the least, we investigated whether chromatin status contributed to the mutation rate of pSNMs. For embryonic pSNMs, the genomic distance between a pSNM and its closest DNase sensitive zone in embryonic stem cells was significantly smaller than the expectation under uniform distribution (Permutation test, P-value = 0.013). In contrast, clonal expansion pSNMs did not showed the enrichment of DNase sensitive zone (Permutation test, P-value = 0.54). We further found that embryonic pSNMs were significantly enriched in the topologically associating domains (TADs) containing embryonically-transcribed genes (Fisher’s exact test, P-value = 0.046), and this pattern was robust with different thresholds for embryonically-transcribed genes (S12 Fig). Moreover, we observed a significantly larger proportion of embryonic pSNMs compared to clonal expansion pSNMs within transcribed chromatin regions using epigenetic data from three cell-lines of different origins (Fisher’s exact test, P-value < 0.05; Fig 3E–3G). Analyses of tissue-shared pSNMs versus clone-specific pSNMs previously identified in neuronal progenitor cells [14] further confirmed our finding (Fisher’s exact test, P-value < 0.01; S13 Fig). In summary, we reported an elevated rate of postzygotic mutations in open and transcribed chromatin regions during embryogenesis, which might result from the exposure of external or internal mutagens within these regions [44].
Recent researches have significantly expanded what is known about the functional roles of postzygotic mutations, which now include not only cancers and overgrowth disorders [45], but also other complex disorders [11, 46]. With the help of next-sequencing technologies, postzygotic mutations have been identified and validated in healthy individuals [13, 39], confirming the theoretical predictions that postzygotic mutations are prevalent and every person is a mosaic [23]. However, the number of rigorously validated postzygotic mutations in healthy individuals has been small, which has hindered statistical analyses of their genomic patterns. In particular, little is known about the genomic patterns of postzygotic mutations in the normal development process of healthy human organs.
In this study, we discovered two distinct types of pSNMs, one occurring during early embryogenesis and the other likely to occur during later tissue-specific clonal expansion. Surprisingly, these mutations manifested many distinct features in regard to mutation spectrum, replication timing, and chromatin status, implying dynamic mutational effects across different developmental stages. Unsurprising in hind sight, clonal expansion pSNMs shared many mutational features with previously reported cancer mutations [47], as tumorigenesis is a specialized process involving clonal expansion of cancer cells [28]. Previous studies reported high proportion of C>A and T>C mutations as well as enrichment of late-replicating regions for clonal expansion pSNMs that were identified from skin fibroblasts [48–50], which was concordant with our findings of clonal expansion pSNMs in the liver and breast samples. In contrast, our embryonic pSNMs demonstrated a range of unique features, including an elevated C>T mutation rate in CpG sites, an enrichment in early-replicating regions, and a stronger effect of transcribed chromatin status (Fig 3). Similar patterns in mutation spectrum (S14 Fig), replication timing (S15 Fig), and chromatin status (S3 Table) could be observed between the embryonic pSNMs that were globally present in all the sequenced organs of the donor and those only present in some but not all the sequenced organs. To further cross-validate our findings, we further analyzed an independent pSNM list that had been identified from human neuronal progenitor cells [14], and confirmed the varied genomic patterns between pSNMs which originated at different developmental stages (Results). In addition to WGS, the elevated C>T mutation rate in CpG sites was also reported in high-fraction pSNMs identified from whole-exome sequencing data [11, 46].
Two of the reasons why the study of postzygotic mutations in healthy organs lags behind that of tumors include the lack of matched control samples in healthy individuals and the significantly lower abundance of postzygotic mutations. Our results showed that 38% of the validated pSNMs were shared by more than one organ, proving the importance of using a control-free pSNM-caller such as MosaicHunter. Furthermore, the high specificity of MosaicHunter compared to other callers enabled us to generate a candidate list that was specific enough to be validated. Single-cell sequencing has been demonstrated to be an alternative approach to study postzygotic mutations [14, 15]. However, compared to single-cell sequencing, bulk sequencing is able to not only provide the genomic location of the postzygotic mutations but also their allele fractions (Fig 1B), which are informative for assessing the proportion of cells that carry the mutation as well as reconstructing the lineage similarity across multiple samples within an individual (Fig 2E and 2F).
The list of pSNMs that had arisen locally during clonal expansion events in the liver and breast samples identified in our study deserve further discussion here. A cell clone with fitness advantage can predominantly proliferate faster and drive all private mutations that were originally carried by that clone to higher allele fractions, allowing them to be detected by bulk sequencing [26]. Because early embryonic pSNMs might affect only a fraction of cells in a certain organ, clonal expansion events could, in theory, make some pSNMs become undetectable from bulk sequencing if the carrier clone was out-competed. Indeed, we observed the breast #7 and breast #9 samples of BBL11121 had lost nine pSNMs that were detected in other breast samples and other organs of the same individual (Fig 2D). These results demonstrated the dynamics of allele fraction for pSNMs driven by clonal expansion events in healthy individuals. We further screened 1407 cancer-related genes from BBLD1005’s liver samples using panel sequencing, and identified four more pSNMs with allele fraction around 1% (S4 Table). However, none of the clonal expansion pSNMs had been previously reported in cancer studies and more functional experiments might be required to examine their relationship with clonal expansion.
The current ~100X WGS bulk sequencing data in our study might not provide enough sensitivity to detect the whole spectrum of pSNMs, especially for those with allele fractions less than 1%. The genomic pattern we reported here were based on the analysis of eight organ types from five individuals. With reduced cost of NGS technology, we can expect a better-characterized spectrum of pSNMs in more and more organ samples and individuals in the future. A combination of deeper bulk sequencing and single-cell sequencing on the same organ sample could provide additional insights for pSNMs with lower allele fractions or even those present in only one or a few cells. This will enable a better characterization of postzygotic mutations in the human population and shed new light on distinguishing clinically-relevant postzygotic mutations from the genomic background.
Twenty-seven postmortem organ samples from five donors were obtained from BioServe Biotechnologies (Beltsville, MD, USA), with an approved protocol from the Institutional Review Board of Medical Informatics Multi-Media Systems, Inc. (5-2-07) and written informed consent obtained from all participants or their legal guardians (Table 1). The clinical histories of all five donors showed no diagnosis of cancer or other known overgrowth disorders. Each organ sample was dissected into nine pieces (roughly 0.5×0.5×0.5 cm each) perpendicular to its long axis and labeled from #1 to #9 (S7 Fig). The peripheral blood samples of two unrelated clinically unremarkable individuals of Asian descent (ACC1 and ACC4) were collected with written informed consent and approval by the Institutional Review Board at Peking University (IRB00001052-13025). Genomic DNA was extracted using an AllPrep DNA/RNA Mini Kit (Qiagen, Hilden, Germany) after homogenization.
Genomic DNA extracted from one piece (labeled as #9) of each of the 27 organ samples was used for WGS and subsequent validation. To reduce potential bias introduced by library preparation, three sequencing libraries were constructed independently for each sample using a KAPA LTP Library Preparation Kit for Illumina platforms (Kapa Biosystems, Wilmington, MA, USA). Size selection was performed for each library with a target insert size of 350–450 bp using a Pippin Prep system (Sage Science, Beverly, MA, USA). Libraries were purified using Agencourt AMPure XP beads (1.0× volume; Beckman Coulter, Brea, CA, USA) and underwent subsequent quality control using a 2100 Bioanalyzer (Santa Clara, CA, USA). Each library was sequenced in one lane on an Illumina Hiseq X Ten platform (Illumina, San Diego, CA, USA) using 150-bp paired-end reads.
Raw sequencing reads were aligned to the GRCh37 human reference genome using the paired-end mode of BWA [51]. The aligned reads were processed using Picard and GATK [52] for the removal of duplicated and ambiguous reads (mismatch >4), indel realignment, and base-quality recalibration. The average depths of processed reads in each organ sample ranged from 83X to 113X (Table 1). CNVs and indels were called using CNVnator [53] and GATK [52], respectively, and all the involved regions as well as annotated repetitive regions were masked, because these regions were more vulnerable to false positives due to mis-alignment or abnormal copy numbers. To maximize the detection sensitivity of pSNMs, the single-sample and paired-sample modes of MosaicHunter [35] were applied to each organ sample with default parameters of genotyper and filters. For the paired-sample mode, the WGS data of the other organs obtained from the same donor served as paired controls. Candidates with at least 5% mutant allele fractions and 3 reads supporting the mutant allele were considered in our subsequent analyses. We randomly chose 16 candidates that are present in the latest versions of dbSNP [54] or the 1000 Genomes Project [55] to validate. As a result, all of them were genotyped as heterozygous rather than pSNMs, suggesting that they were more likely to be caused due to the large variation of allele fraction in WGS. Therefore, we only considered the candidates absent in both databases for thorough validation below.
All the candidate pSNMs were analyzed using the standard workflow of PASM, which had been previously benchmarked by pyrosequencing and micro-droplet digital PCR [5, 56]. PCR primers for PASM were successfully designed for 241 out of the 251 candidate pSNMs, with their amplicon lengths ranging between 380 and 420 bps; for a few candidates located in highly homologous genomic regions, two rounds of nested PCR were carried out to achieve higher amplification specificity (S5 Table). For one-step amplicons, 35 cycles of PCR amplification was carried out using 2X Ex-Taq Premix (Takara Bio, Dalian, China). For nested amplicons, additional 15 cycles of first-round amplification and 25 cycles of second-round amplification was carried out to capture the target region specifically.
Amplicons from different organ samples and individuals were barcoded during library preparation and then pooled and sequenced using either the Ion Torrent PGM or Ion S5 XL sequencer (ThermoFisher, Guilford, CT, USA), following the manufacturer’s protocols. Sequencing of the same PASM library on the PGM and S5 XL platforms showed a similar distribution of depth-of-coverage (S16 Fig). In each sample, we calculated the 95% credible intervals of minor allele fractions for each candidate site with at least 30X depth [5]. Candidates with 95% credible intervals between 0.5% and 40% were considered a mosaic genotype, whereas those sites with a 95% credible intervals’ lower bound below 0.5% or an upper bound above 40% were considered a homozygous or heterozygous genotype, respectively. Five candidate sites without sufficient depth in the negative control samples were removed. As shown in S17 Fig, the inter-organ and intra-organ variation of the minor allele fractions for pSNMs could not be explained by the technical variance induced by DNA extraction or PASM.
The differences in pSNM profiles across multiple samples from one individual were assessed using the Euclidean distance (D) of the square root of the allele fraction for all the validated sites. The relative similarity (RS) between sample i and j was defined as
RSij=1−Dij−min(D)max(D)−min(D)
CNVkit [57] was applied to estimate copy number from the WGS data, and the 100 bp windows centered by each pSNM were considered. The individual-specific copy number was then normalized using the mean across all the five individuals. Three of the 167 PASM-validated pSNMs were found to demonstrate abnormal copy number (< 1.7 or > 2.3 for autosomes and females’ X chromosome and < 0.9 or > 1.1 for males’ X chromosome) in the corresponding carriers (S1 Table). The estimated copy numbers were 2.79, 2.80, and 2.59 for these three pSNMs (Q7, P35, and N26), respectively. Their mutant alleles were globally present in all the sequenced organs of the carrier, and the average allele fractions were close to 1/3 (26.2% to 30.9%). Rather than postzygotic mutations, these three sites were very likely to be explained by the germ-line events of copy number gain which made the allele fraction of involved heterozygous mutations deviate from 50%. Therefore, we excluded these three sites from all the following analyses.
The genome-wide annotation of DNA replication timing was extracted from two independent studies [43, 58]. For Hansen et al., we downloaded the wavelet-smoothed signal datasets of five different cell-lines, including GM12878, K562, HeLa-S1, HepG2, and HUVEC [58]. For Koren et al., the genome-wide profile was averaged from six lymphoblastoid cell-lines [43]. In both studies, a higher value represented an earlier DNA replication timing. For the permutation test, because our MosaicHunter pipeline only considered candidate pSNMs in non-repetitive regions, we compared the observed median of replication timing for each type of pSNMs against the null distribution estimated by using 1000 times of genome-wide random shuffling among non-repetitive regions.
The DNase-seq data of the embryonic stem cell-line H1-hESC was downloaded from the ENCODE project [59], and the DNase I sensitive zones were identified using the HotSpot algorithm [60]. For the permutation test, we randomly permutated the genomic positions of pSNMs among non-repetitive regions and assessed the median distance between each permutated pSNM and its closest DNase I sensitive zone. Each permutation was replicated 1000 times to estimate the distribution under the null hypothesis.
The annotation of chromatin states in HepG2, HMEC, and K562 cell-lines was downloaded from the UCSC Genome Browser [61] to represent cell types derived from the three primary germ layers in early embryo. Chromatin states were inferred from ChIP-seq data of ten epigenetic factors using a Hidden Markov Model [62]. The inferred chromatin states of active promoter, enhancer, and transcription were defined as “transcribed” chromatin status, whereas the other states involving repressed elements and heterochromatin were defined as “repressed” chromatin status.
The annotation of TADs for the H1-hESC cell-line was downloaded from the ENCODE project [59], which was generated based on the Hi-C data of Dixon et al. [63]. The expression profile of the same cell-line was also downloaded from the ENCODE project under GEO accession number GSM958733. A gene was defined as embryonically transcribed if its FPKM was larger than 20 in H1-hESC. We also used other FPKM thresholds to confirm the robustness of our finding.
To screen for potential driver mutations related to the clonal expansion events in BBLD1005’s liver samples, we captured 1407 cancer-related genes from the liver #2 and #8 samples of BBLD1005 using a Roche SeqCap panel (Pleasanton, CA, USA) designed by Genecast Biotechnology (Beijing, China). The captured libraries were sequenced by Illumina Novaseq6000 (Illumina, San Diego, CA, USA) using 150-bp paired-end reads, with an average depth of ~2000X. Sequencing reads were aligned to the GRCh37 human reference genome by BWA. The pSNMs were identified using MosaicHunter, and the 95% credible intervals of mutant allele fraction were estimated by the Bayesian model implemented in PASM.
|
10.1371/journal.pntd.0004724 | Genomic African and Native American Ancestry and Chagas Disease: The Bambui (Brazil) Epigen Cohort Study of Aging | The influence of genetic ancestry on Trypanosoma cruzi infection and Chagas disease outcomes is unknown.
We used 370,539 Single Nucleotide Polymorphisms (SNPs) to examine the association between individual proportions of African, European and Native American genomic ancestry with T. cruzi infection and related outcomes in 1,341 participants (aged ≥ 60 years) of the Bambui (Brazil) population-based cohort study of aging. Potential confounding variables included sociodemographic characteristics and an array of health measures. The prevalence of T. cruzi infection was 37.5% and 56.3% of those infected had a major ECG abnormality. Baseline T. cruzi infection was correlated with higher levels of African and Native American ancestry, which in turn were strongly associated with poor socioeconomic circumstances. Cardiomyopathy in infected persons was not significantly associated with African or Native American ancestry levels. Infected persons with a major ECG abnormality were at increased risk of 15-year mortality relative to their counterparts with no such abnormalities (adjusted hazard ratio = 1.80; 95% 1.41, 2.32). African and Native American ancestry levels had no significant effect modifying this association.
Our findings indicate that African and Native American ancestry have no influence on the presence of major ECG abnormalities and had no influence on the ability of an ECG abnormality to predict mortality in older people infected with T. cruzi. In contrast, our results revealed a strong and independent association between prevalent T. cruzi infection and higher levels of African and Native American ancestry. Whether this association is a consequence of genetic background or differential exposure to infection remains to be determined.
| Chagas disease (ChD), which is caused by the protozoan Trypanosoma cruzi, affects approximately 8 million people worldwide. ChD is known as a neglected tropical disease. The disease is endemic in South and Central American countries, and is an emerging issue in North America and Europe. This study examined, for the first time, the association between genomic ancestry and T. cruzi infection, Chagasic cardiomyopathy and its ability to predict long term mortality. Our results show that persons with higher levels of African and Native American ancestries (and the reverse for European ancestry) are more likely to be infected with T. cruzi. However, genomic ancestry had no effect on either Chagasic cardiomyopathy or on its ability to predict mortality. Whether the association between T. cruzi infection and genomic ancestry is a consequence of genetic susceptibility or differential exposure to infection due to poor socioeconomic circumstances over the life course, remains to be determined.
| Chagas disease (ChD), which is caused by the protozoan Trypanosoma cruzi, affects approximately 5.7 million people in 21 Latin American countries [1]. ChD is known as a neglected tropical disease and is an emerging issue in North America and Europe [2–5]. ChD is autochthonous in South and Central America but T. cruzi infection has spread to other regions of the world primarily due to immigration of infected persons [2], although there has been evidence of some locally-occurring infections in the United States [3]. Currently, at least 300,000 persons with T. cruzi infection live in the US [4] and at least 80,000 in Europe [5]. The disease is costly to individuals and society with estimates of over USD 100 million spent on treatments and over USD 800 million in lost productivity each year [6]. Up to one third of those infected with ChD may develop chronic heart abnormalities and other complications of which Chagas cardiomyopathy is the most severe and life-threatening form [7]. The presence of major electrocardiogram (ECG) abnormalities (a diagnostic measure of Chagas cardiomyopathy) doubles the risk for mortality in T. cruzi-infected elderly populations [8].
The influence of African and/or Native American ancestry on T. cruzi infection and/or ChD outcomes is unknown. The existence of an association is plausible for at least two reasons: first, familial aggregation of T. cruzi seropositivity and ECG abnormalities have been found in highly endemic areas, suggesting that genetic variation may play a role in susceptibility to infection as well as disease progression [9,10]; second, an earlier publication, using ethnoracial self-classification, reported greater prevalence of ECG abnormalities among Black middle-aged adults relative to their White counterparts [11].
Latin America is one of the most ethnoracially heterogeneous regions of the world [12], and Brazil is the largest and the most populous ChD endemic country in the region. The current Brazilian population’s genetic makeup is the product of admixture between Amerindians, Europeans colonizers or immigrants, and African slaves [13]. Brazil received nearly 4 million slaves from Africa, about seven times more than the United States [14]. Thus, the Brazilian population provides an opportunity to assess the relationship between T. cruzi infection and its complications with genetic ancestry in admixed populations.
The Bambui-Epigen Cohort Study of Aging is conducted in a well-defined population of older Brazilian adults living in a formerly ChD endemic area [15]. We examined for the first time the association between genome-wide proportions of genomic ancestry with T. cruzi infection and cardiomyopathy, taking into account an array of socioeconomic and health indicators that could confound such an association. Additionally, we examined whether genomic ancestry affects the prognostic value of major ECG abnormalities for 15-year mortality in T. cruzi-infected individuals.
The Bambui cohort study of aging is ongoing in Bambuí, a city of approximately 15,000 inhabitants in the state of Minas Gerais in Southeast Brazil, which is one of the oldest known endemic areas for ChD [16–18]. Detailed information on this cohort can be found elsewhere [15]. Briefly, the population eligible for the cohort consisted of all residents aged 60 years and over on 1 January 1997 (92% of the 1,742 inhabitants in this age group participated). Most participants had some degree of admixture between African, European and Native American genomic ancestry [19,20].
T. cruzi infection status was assessed by means of three different assays performed concurrently: a hemagglutination assay (Biolab Merieux SA, Rio de Janeiro, Brazil) and two enzyme-linked immunosorbent assays (Abbott Laboratories, Inc., North Chicago, Illinois; and Wiener Laboratories, Rosario, Argentina). Infection with T. cruzi was defined by seropositivity in all of the three examinations; seventeen persons had discordant results among the assays and were excluded from the analysis. As far as we could determine, none of the cohort participants had a history of use of antitrypanosomal medications, and none of the seropositive subjects reported such treatment over the ensuing decade during annual follow-up visits. Thus, the use of antitrypanosomal therapy was not considered in the present analysis. In addition, no cohort participant had received a cardiac transplant.
At the baseline examination, a digitally recorded 12-lead ECG (Hewlett Packard MI700A) reading was obtained at rest. ECGs were analyzed at the ECG Reading Center (EPICARE, Wake Forest University) and classified using the Minnesota Code (MC) criteria [21,22]. Major ECG abnormalities were defined by the presence of at least one of the following: old (MC 1.1.x or 1.2.x) or possible myocardial infarction (1.3.x and 4.1.x, 4.2, 5.1, or 5.2), complete intraventricular blocks (MC 7.1, 7.2, 7.4, or 7.8), frequent supraventricular or ventricular premature beats (MC 8.1.x, except 8.1.4), major isolated ST segment or T-wave abnormalities (MC 4.1.x, 4.2, 5.1 or 5.2), atrial fibrillation or flutter or supraventricular tachycardia (MC 8.3.x.or 8.4.2), other major arrhythmias (MC 8.2.x, except 8.2.1), major atrioventricular conduction abnormalities or pacemaker use (MC 6.1, 6.2.x, 6.4, 6.8, 8.6.1 or 8.6.2), major QTi prolongation (>115%) and left ventricular hypertrophy (LVH) (MC 3.1 together with [4.1.x, 4.2, 5.1, or 5.2]). Further details can be seen elsewhere [8].
Cohort participants were genotyped with the Omni 2.5M array (Illumina, San Diego, California) [13]. We performed ancestry inferences using the model-based method [23], implemented in the Admixture software. First, we used 370,539 SNPs to estimate for each individual African, European and Native American tri-hybrid ancestry proportions, using 266 African, 262 European and 93 Native American individuals from public datasets as parental populations [13]. Further, we inferred a kinship coefficient for each pair of individuals, using the software Reap [24], conditioning on tri-hybrid individual admixture proportions. We used complex networks to identify families from the matrix of pair-wise kinship coefficients [13]. In this approach, pairs of individuals (i.e. families) are related if they have a kinship coefficient >0.1 (first and second-degree relatives). Given that Brazilians with African ancestry generally have a high proportion of East African genetic markers (as opposed to markers of West African origin), relative to African Americans and those from the Caribbean [13,25,26], we used 331,790 SNPs and the reference dataset “U” [13] to further divide total African ancestry into its two components: a Western-African/non Bantu and an Eastern African/Bantu, hereafter called Western African and Eastern African, respectively. The fact that many Bambuí residents are related could affect high-resolution inferences of biogeographic ancestry (such as West- and East-African) with the Admixture software. To overcome this limitation, we performed separate Admixture runs to infer West- and East- African ancestry components, avoiding the presence of related individuals in the same run. Further details on how genetic and ancestry analyses of the Bambui cohort population were performed can be found elsewhere [13,27].
Deaths that occurred between study enrollment in 1997 and December 31, 2011, were included in the present analysis. Deaths were reported by next of kin during the annual follow-up interview and verified through the Brazilian mortality information system. Death certificates were obtained for 95.7% of the participants who died. Deaths from any cause were considered in this analysis.
Potential confounding variables included baseline sociodemographic characteristics (age, sex, schooling, household income and father’s occupation) and health measures (current smoking, hypertension, diabetes, coronary heart disease, C-reactive protein and non-HDL cholesterol level). We categorized schooling into incomplete primary school (<4 years) and complete primary and higher (4 years and more). We categorized monthly household income per capita into equal or superior to the median value (median = 1.5 Brazilian minimum wages or USD 180 in 1997). Occupation of the study participant’s father (as informed by cohort members) was categorized into urban workers, landowners, manual rural workers and unknown. Current smokers were persons who had smoked at least 100 cigarettes during their lifetime and who still smoke. Body mass index (BMI) was defined as weight (in kg) divided by height (in meters) squared. Hypertension was defined by mean (two out of three measures) systolic blood pressure of ≥140 mmHg and/or diastolic pressure of ≥90 mmHg and/or treatment [28]. Diabetes mellitus was defined by fasting blood glucose ≥126mg/dL and/or treatment [29]. Coronary heart disease was defined by prior medical diagnosis of myocardial infarction and/or symptoms of angina pectoris [30]. High sensitivity C-Reactive Protein was measured by the CRP immunonephelometric method (BNII, Dade Behring, Marburg, Germany). Blood fasting glucose and cholesterol were determined by using standard enzymatic methods (Merck, Darmstadt, Germany). Non-HDL cholesterol was defined by total cholesterol level minus HDL cholesterol.
Unadjusted analyses were based on Pearson´s chi square, oneway ANOVA and Kruskall Wallis tests to examine differences across frequencies, means and medians, respectively. Individual proportions of genomic ancestries were expressed as medians or divided into quintiles.
Prevalence ratios (PR) estimated by multivariable Poisson regression [31] were computed to examine associations between (i) genomic ancestry in quintiles and T. cruzi infection and (ii) genome ancestry in quintiles and major ECG abnormality among persons infected with T. cruzi. Further, we used Cox proportional hazard models to implement an analysis restricted to persons infected with T. cruzi to assess the influence of each category of genomic ancestry on the risk of major ECG abnormalities and subsequent mortality.
The above-mentioned statistical analyses were based on two models. First, prevalence and hazard ratios were adjusted for age (continuous), sex, smoking, hypertension, diabetes, coronary heart disease (all dichotomous variables) plus body mass index, log-transformed C-reactive protein and non-HDL cholesterol (as continuous measures). We then added schooling, monthly household income per capita, and father’s occupation to the previous models. Because 913 participants were first- or second-degree relatives, and excluding them would lead to loss of power and possible selection bias, we kept all related individuals in our analyses and used robust variance estimators in multivariate models to correct results for clustering by family structure. Finally, we examined separately the significance of the effect of multiplicative interactions between sex and genomic ancestry on each outcome by means of cross-product terms in Poisson and Cox proportional hazards regression models, respectively. Since there was no evidence of interaction with sex, the analyses were carried out for both men and women with sex included as a covariate.
Separate analyses were performed for African, Native American and European genomic ancestries and further for Western African sub continental ancestry. Statistical analyses were conducted using STATA 13.0 statistical software (Stata Corporation, College Station).
The Bambui cohort study of aging was approved by the Institutional Review Board of the Oswaldo Cruz Foundation, Rio de Janeiro, Brazil. Genotyping was approved by Brazil’s national research ethics committee, as part of the Epigen-Brazil protocol (CONEP, resolution 15895). Written informed consent was obtained from all participants at baseline and at all follow-up interviews.
Of the 1,606 baseline cohort participants, 1,343 had complete information for all study variables and were included in the current analysis. As shown in Table 1, the prevalence of T. cruzi infection was 37.6% (n = 505). At baseline, the mean age of participants was 68.8 years, 61.2% were women, and low schooling level (<4 years) largely predominated (64.1%). The median proportions of African, Native American and European genomic ancestries were 9.6%, 5.4% and 83.8%, respectively. The median proportion of Western African sub-continental ancestry relative to total African ancestry was 63.9% (complementarily, the corresponding value for Eastern African ancestry was 36.1%). T. cruzi infected participants had significantly higher median individual proportions of African and Native American ancestries and significantly lower median European genomic ancestry. Other baseline characteristics of the study participants, by T. cruzi infection status, are presented in Table 1.
Table 2 presents median individual proportions of African, Native American and European genomic ancestries by baseline characteristics. Median African and Native American genomic ancestries were significantly higher (and European ancestry was significantly lower) among those with lower schooling and income levels, those whose fathers were manual workers or had an unknown occupation, as well as those with any major ECG abnormality or previous coronary heart disease. Median African ancestry was lower in those aged 69 years and over and in those with BMI under 25 kg/m2. No significant associations with genomic ancestry were found for other study variables.
Associations between the different genomic ancestries and T. cruzi infection are shown in Table 3. There was a graded positive univariate association between T. cruzi infection with the proportion of African and Native American ancestry, and a graded negative relationship with a greater proportion of European ancestry (p<0.001 for all). After adjustments for age, sex and health measures, persons at the intermediate and highest quintiles of African and Native American ancestry were significantly more likely to be infected with T. cruzi relative to their counterparts in the lowest quintiles. After further adjustments for socioeconomic indicators (schooling, income and father’s occupation) these associations were attenuated, but remained largely significant (PR = 1.38; 95% CI 1.07, 1.79 and PR = 1.74; 95% CI 1.37, 2.35 for those at the intermediate and highest quintiles of African ancestry, respectively, and PR = 1.54; 95% CI 1.19, 1.99 for those at the highest quintile of Native American ancestry). The opposite trend was found for European ancestry (PR = 0.73; 95% CI 0.63, 0.85 and PR = 0.54; 95%CI 0.41, 0.70, respectively).
Among those infected, 56.4% had at least one major ECG abnormality (31.7% among the non-infected). As shown in Table 4, in the bivariate analysis, a major ECG abnormality among infected persons was not found to be significantly (p>0.05) associated with African, Native American or European ancestry levels. This absence of association remained in analyses adjusted for age, sex and health measures, as well as in analyses further adjusted for socioeconomic indicators.
Over a 15 year follow-up period, 683 participants died and 109 (8.1%) were lost to follow-up, leading to 14,680 person-years (pyrs) of observations (5,251 pyrs among the infected). The death rate was 46.4 per 1,000 pyrs (56.2 and 40.9 per 1,000 pyrs among T. cruzi infected and non-infected, respectively). As shown in Table 5, persons infected with T. cruzi with any major ECG abnormality were at significantly increased risk of death, compared to their counterparts with no such abnormalities, independent of age, sex and other health measures (HR = 1.83; 95% CI 1.44, 2.34). Further adjustments for socioeconomic indicators had little impact on this association (HR = 1.78; 95% CI 1.39, 2.28). The association was consistent across different levels of African, Native American and European genomic ancestries. We found no evidence of statistically significant multiplicative interactions between African, Native American and European genome ancestry levels and major ECG abnormalities on mortality (p>0.05 for all).
As shown in Table 6, a statistically significant association between Western African proportion and T. cruzi infection was found in bivariate analysis, but the association lost significance after adjustments for socio demographic characteristics and health measures. Furthermore, we did not find any evidence of an association between the above mentioned ancestry levels and the presence of major ECG abnormalities among people infected with T. cruzi in either univariate or multivariate analyses (p>0.05 for both). Finally, as previous observed for global African ancestry, levels of Western African ancestry did not modify the association between a major ECG abnormality and subsequent mortality among infected subjects (p value for interactions >0.05).
The key findings of the current study are: first, T. cruzi infection was strongly correlated with both African and Native American ancestry—and conversely showed a negative correlation with European ancestry—and this association had a graded effect; second, cardiomyopathy in infected persons was not associated with either African or Native American or European ancestry levels; third, genomic ancestry had no significant effect modification on the prognostic value of major ECG abnormalities for mortality in T. cruzi infected older adults; fourth, Western African sub continental origin was not associated with either T. cruzi infection or related outcomes. The above-mentioned findings were independent of an array of sociodemographic and biological confounders.
The association between T. cruzi infection and higher levels of African and Native American ancestry may result from genetic influence on susceptibility and/or greater exposure to infection in these groups during the life course. Our study population was born before 1940, and this cohort has experienced dramatic political and social changes during their lifetimes. Brazil has transitioned from a low-income, primarily rural country in the mid-1950s, to one of the largest economies in the world, with 84% of the population living in urban areas by 2010 [32,33]. Chagas disease is related to poor socio-economic circumstances, mostly in early life. In endemic areas, the main source of infection is a bloodsucking triatomine insect that colonizes poor households. Most individuals in these areas acquire the infection before they reach 20 years of age [34]. Further, ethnoracial disparities in Brazil are remarkable. Persons of African origin are more likely to have lower income and education, to experience race-based discrimination, and to report worse health outcomes [14,35]. Native Americans experience sustained marginalization [36]. Our results are in agreement with these observations, revealing higher levels of African and Native American ancestry in those with lower schooling and family income levels, as well as those whose fathers were rural workers or had an unknown occupation (which suggests a less prestigious occupational category). T. cruzi infection followed this trend, with higher prevalence associated with worse current (measured by income) and worse early socioeconomic circumstances (educational attainment and father’s occupation). However, the association between higher levels of African and Native American ancestry with T. cruzi infection was attenuated, but still remained largely significant after adjustments for socioeconomic indicators, suggesting a possible independent effect of genomic ancestry. Despite this finding, it is important to emphasize that although we control for several important measures of current and early socioeconomic circumstances, they cannot completely account for the complexity of unfavorable trajectories of persons with higher levels of African and Native American ancestry in Brazilian society [14]. Thus, we cannot exclude the possibility that residual confounding may still account for the association between higher levels of African and Native American ancestry and prevalent T. cruzi infection in our analysis. The fact that analyses of subsequent complications (cardiomyopathy) showed no association with genomic ancestry further tempers any inference regarding a causal relationship between genetic ancestry and increased vulnerability to T. cruzi.
Chronic Chagas cardiomyopathy is the most clinically relevant manifestation of the disease. It manifests as heart failure, arrhythmia, heart block, thromboembolism, stroke and sudden death [7,16]. The pathogenesis of chronic chagasic cardiomyopathy is not completely understood [37], but inflammation caused by persistent parasitism of the heart tissue appears to play an important role [38,39]. Additionally, a recent genome-wide study (GWAS) identified suggestive single nucleotide polymorphisms (SNPs) that may impact the risk of progression to cardiomyopathy in seropositive persons [37]. Electrocardiography has been considered an important tool in the management of ChD patients [7]. Information on ECG findings among the elderly infected with ChD is scant, and very few studies in middle-aged or older adults have used core-lab readings using classifications developed by the internationally accepted Minnesota Code [8]. A previous study in the Bambui cohort showed that any major ECG abnormality (classified by the Minnesota Code) was strongly and independently associated with increased risk for 10-year mortality among T. cruzi infected older adults [8]. The results of the current analysis, based on an extended 15 year-follow-up, are in agreement with these findings. Additionally, we found no evidence of an association between African and Native American ancestries and major ECG abnormalities among T. cruzi infected persons. The absence of an association was consistent in bivariate analyses as well as those adjusted for an array of potential confounding factors. Furthermore, African, Native American and European ancestry showed no significant interactions affecting the ability of major ECG abnormalities to predict subsequent mortality.
Strengths of this study include the large population-based cohort followed for an extended period, and minimal loss of participants to follow-up. Another major strength is the use of genome-wide measures of ancestry. Genomic ancestry does not change over time, while ethnoracial self-classification is prone to misclassification—particularly in admixed populations [14,19]. Another strength is the inclusion of several biological and non-biological risk factors in our analysis. However, one cannot exclude the possibility that there may be additional unmeasured factors, including unknown genetic factors that confound our results.
The current study is, to our knowledge, the first investigation on the influence of African, Native American and European genomic ancestry on T. cruzi infection and related outcomes. Our findings indicate that African and Native American ancestry have no influence on the presence of major ECG abnormalities and had no influence on the ability of an ECG abnormality in predicting mortality in older people infected with T. cruzi. In contrast, our results revealed a strong positive association between prevalent T. cruzi infection with higher levels of African and Native American ancestry. Whether this association is a consequence of genetic background, differential exposure to infection, or a combination of both factors, remains to be determined.
|
10.1371/journal.pgen.1004526 | Asymmetric Division and Differential Gene Expression during a Bacterial Developmental Program Requires DivIVA | Sporulation in the bacterium Bacillus subtilis is a developmental program in which a progenitor cell differentiates into two different cell types, the smaller of which eventually becomes a dormant cell called a spore. The process begins with an asymmetric cell division event, followed by the activation of a transcription factor, σF, specifically in the smaller cell. Here, we show that the structural protein DivIVA localizes to the polar septum during sporulation and is required for asymmetric division and the compartment-specific activation of σF. Both events are known to require a protein called SpoIIE, which also localizes to the polar septum. We show that DivIVA copurifies with SpoIIE and that DivIVA may anchor SpoIIE briefly to the assembling polar septum before SpoIIE is subsequently released into the forespore membrane and recaptured at the polar septum. Finally, using super-resolution microscopy, we demonstrate that DivIVA and SpoIIE ultimately display a biased localization on the side of the polar septum that faces the smaller compartment in which σF is activated.
| A central feature of developmental programs is the establishment of asymmetry and the production of genetically identical daughter cells that display different cell fates. Sporulation in the bacterium Bacillus subtilis is a simple developmental program in which the cell divides asymmetrically to produce two daughter cells, after which the transcription factor σF is activated specifically in the smaller cell. Here we investigated DivIVA, which localizes to highly negatively curved membranes, and discovered that it localizes at the asymmetric division site. In the absence of DivIVA, cells failed to asymmetrically divide and prematurely activated σF in the predivisional cell, largely unreported phenotypes for any deletion mutant in a sporulation gene. We found that DivIVA copurifies with SpoIIE, a protein that is required for asymmetric division and σF activation, and that both proteins preferentially localize on the side of the septum facing the smaller daughter cell. DivIVA is therefore a previously overlooked structural factor that is required at the onset of sporulation to mediate both asymmetric division and compartment-specific transcription.
| Asymmetric cell division and differential gene expression are hallmarks that underlie the differentiation of a progenitor cell into two genetically identical, but morphologically dissimilar daughter cells [1]–[5]. The rod shaped Gram-positive bacterium Bacillus subtilis, which normally divides by binary fission to produce two identical daughter cells, undergoes such a differentiation program, termed sporulation, when it senses the imminent onset of starvation conditions (reviewed in [6]–[8]). During sporulation, B. subtilis first divides asymmetrically by elaborating a so-called “polar septum” that produces two unequal-sized daughter cells: a larger “mother cell” and a smaller “forespore” (Fig. 1A) that each receive one copy of the genetic material. After asymmetric division, the daughter cells remain attached and a compartment-specific transcription factor called σF is exclusively activated in the forespore. This activation step is critical because it sets off a cascade of transcription factor activation events, each in an alternating compartment, resulting in the expression of a unique set of genes in each daughter cell, which ultimately drives the rest of the sporulation program [9], [10]. Subsequently, the forespore is engulfed by the mother cell and eventually the forespore achieves a partially dehydrated state of dormancy in which its metabolic activity is largely arrested and is released into the environment when the mother cell ultimately lyses- the released cell is termed a “spore” (or, formally, an “endospore”) [11]. Several factors that are required for the switch from medial to asymmetric division have been identified, but the mechanisms underlying this switch remain largely unknown. Similarly, the biochemical basis for the activation of σF has been well elucidated, but the cell biological basis for how this activation is achieved exclusively in the forespore is less well known.
At the onset of sporulation, FtsZ, the bacterial tubulin homolog that provides the force for membrane invagination during cytokinesis, initially assembles at mid-cell into a ring-like structure called the “Z-ring” [12]–[14]. At this time, an integral membrane protein called SpoIIE is also produced in the pre-divisional cell and co-localizes with FtsZ via a direct interaction [15]–[17]. Instead of constricting at mid-cell, though, the Z-ring next unravels and extends outward towards each pole via a helix-like intermediate and finally reassembles as two separate Z-rings near the two poles of the bacterium; SpoIIE similarly redeploys to the two polar positions with FtsZ [18]. This redeployment of the Z-ring requires SpoIIE and increased expression of ftsZ from a second sporulation-specific promoter [18]–[23]. Next, one of the two polar Z-rings constricts [24], [25], thereby elaborating the polar septum on one end of the bacterium. Although FtsZ constricts at this site and eventually dissipates into the cytosol, SpoIIE somehow remains associated with the polar septum [15], [16], [26]–[28]. A recent report demonstrated that SpoIIE is released into the forespore membrane soon after septum formation is complete and that it is recaptured at the polar septum [29]. Interestingly, the total level of SpoIIE before release and after recapture was similar suggesting that SpoIIE is exclusively released into the forespore membrane after septum formation, but the mechanism by which FtsZ could preferentially release SpoIIE into the forespore membrane is not known.
After formation of the polar septum SpoIIE performs a second function in which it activates σF [23], [30]. Prior to asymmetric division, σF is synthesized in the pre-divisional cell, but is held inactive by an anti-sigma factor called SpoIIAB [31], [32]. After asymmetric septation, SpoIIE, whose C-terminus harbors a phosphatase domain [33], [34], dephosphorlyates an anti-anti-sigma factor (SpoIIAA), which then binds and sequesters SpoIIAB, thereby relieving σF inhibition [23], [35]–[37]- somehow, this activity is manifested only in the forespore compartment. Some evidence has suggested that this compartment exclusivity is ultimately due to a preferential localization of SpoIIE on the forespore side of the polar septum [38], [39], but how and when this asymmetric localization initially arises has not been clear.
In this study, we examined the subcellular localization of DivIVA, a peripheral membrane protein made of coiled-coil domains that spontaneously assembles into a higher order structure, at the onset of asymmetric division. During vegetative growth of B. subtilis, DivIVA localizes to nascent cell division sites at mid-cell at the very onset of membrane constriction [40]. It has been proposed that negative membrane curvature, such as that which arises on either side of a division septum where it meets the lateral edge of the cell, provides a geometric cue that drives the localization of DivIVA to assemble into ring-like structures on both sides of a division septum [40]–[43]. During normal growth, DivIVA rings serve as platforms that recruit the MinCD complex [40], [44]–[46], which inhibits FtsZ assembly, to either side of the nascent cell division septum [47]. As a result, aberrant FtsZ assembly immediately adjacent to a newly forming septum (and thus, the formation of “minicells” devoid of DNA) is inhibited and membrane constriction occurs once, and only once, at mid-cell [40], [47]. At the onset of sporulation, DivIVA performs a second function: DivIVA rings collapse into patches at the two hemispherical cell poles [40] where it anchors the origins of replication of the two replicated chromosomes (via a DNA-binding protein called RacA) [41], [48]–[50], thereby assuring that both the forespore and mother cell receive one copy of the chromosome.
Here, we report that DivIVA localizes to the polar septum and plays an additional role at the onset of sporulation. Deletion of the divIVA gene or depletion of DivIVA protein after its chromosome anchoring function resulted in a severe asymmetric septation defect due to an inability of cells to redeploy FtsZ and SpoIIE from medial to polar positions. As a result, cells arrested at this stage of sporulation, unlike other division mutations reported to cause an asymmetric division defect, prematurely activated σF in a compartment-unspecific manner. We discovered that DivIVA and SpoIIE exist in a complex with one another in sporulating cells and that, when co-produced in vegetative cells, SpoIIE did not persist at division septa in the absence of DivIVA, consistent with a model in which DivIVA is required to briefly anchor SpoIIE at the polar septum during sporulation once FtsZ begins to constrict and subsequently leaves the septum, before SpoIIE is released into the forespore membrane. Finally, employing super-resolution microscopy, we observed that DivIVA initially localized to both sides of the polar septum at the very onset of membrane invagination, but that it preferentially persisted at the forespore side once septation was completed. In contrast, SpoIIE preferentially localized to the forespore side of the polar septum at the onset of membrane constriction, and maintained this biased localization pattern until the completion of polar septum formation. Eventually, SpoIIE was released into the forespore membrane. We propose that DivIVA performs a previously unappreciated role in asymmetric division and compartment-specific activation of σF during sporulation.
DivIVA was previously reported to localize to medially-placed division septa that form during vegetative growth in Bacillus subtilis, but not to the asymmetrically-placed “polar” septum that forms at the onset of sporulation [50]. This implied that DivIVA is somehow able to detect a feature that is exclusive to vegetative division septa, and is absent at asymmetric septa. DivIVA, however, reportedly localizes to division septa by detecting a geometric localization cue, not a chemical cue such as a pre-localized protein: namely, it is thought to preferentially embed onto negatively curved (concave) patches of membrane, such as those that are present where a division septum meets the lateral edge of a rod-shaped cell [41], [42]. Since negative membrane curvature is a feature that is shared by medially-placed division septa that are formed during vegetative growth and asymmetrically-placed polar septa during sporulation, we hypothesized that, if present during sporulation, DivIVA should indeed also localize to the polar septum in sporulating cells. To verify if DivIVA is present at the time of asymmetric septation, we first observed the total levels of DivIVA protein at various time points in synchronized cultures of sporulating B. subtilis by immunoblotting with antibodies specific to DivIVA. Entry into the sporulation pathway was followed by epifluorescence microscopy and enumerating the number of cells that had elaborated a polar septum. Although not all cells in the culture typically initiate sporulation, by 2.5 hours after initiation of the program approximately half of the cells had elaborated a polar septum, yet DivIVA protein levels were largely unchanged, and remained constant for at least the first four hours after induction of sporulation, suggesting that DivIVA persists well into sporulation (Fig. 1B). To examine the subcellular localization of DivIVA in sporulating cells, we first constructed a DivIVA-GFP fusion with a flexible linker that had been previously used to construct a nearly fully functional DivIVA-CFP fusion [44], produced under the control of its native promoter at the ectopic, non-essential amyE locus. As reported for the CFP fusion, cells harboring DivIVA-GFP as the only copy of DivIVA were of similar size as wild type cells and produced very few minicells, suggesting that the fusion protein is largely functional (Fig. S1). The localization of DivIVA-GFP to the polar septum was observed in 93% of sporulating cells that we examined (n = 105; Fig. 1C; the remaining 7% of cells counted displayed either no or weak DivIVA-GFP signal). Reconstruction of deconvolved Z-stacks of a cell that was beginning to elaborate a polar septum (cell #3, Fig. 1C) revealed that DivIVA formed a ring-like structure (Fig. 1D, top row), similar to DivIVA ultrastructure found at vegetative septa as reported previously [40], [43]. Representation of the reconstructed fluorescence signal of DivIVA-GFP and DAPI (indicating the chromosome) as a surface revealed two separate populations of DivIVA: one at the extreme poles that fulfills the chromosome anchoring role of DivIVA patches, and a second, ring-like localization of DivIVA at the nascent polar septum through which the chromosome was threaded (Fig. 1D, bottom row). Time lapse epifluorescence microscopy revealed that soon after formation of the polar septum initiated, DivIVA-GFP localized at that site and remained associated with the completed septum (Fig. S2B).
Next, we examined if the DivIVA ring remains at the site of septation even after membrane constriction (similar to what has been observed at vegetative septa) or if it collapses with the rest of the division machinery as cytokinesis proceeds. We observed the localization of DivIVA-CFP and FtsZ (the major component of the divisome that drives membrane constriction [51]), fused to YFP, in sporulating cells that produced both proteins and had just initiated asymmetric septation. In these cells, FtsZ-YFP localized as a band whose width was less than the width of the entire cell (Fig. 1E, top row), consistent with a pattern of constriction of the division machinery. In the same cells, though, DivIVA-CFP remained as two foci at the site of division that did not overlap with the fluorescence signal from FtsZ-YFP, consistent with the formation of a static ring that did not constrict. Reconstruction of deconvolved Z-stacks revealed that FtsZ-YFP formed a collapsing disk-like structure during constriction of the nascent polar septum, whereas DivIVA-CFP remained as a ring-like structure with an outer diameter that was larger than that of FtsZ-YFP (1E, bottom row). We conclude that, similar to the situation at vegetative septa, DivIVA localizes to the polar septum at the onset of sporulation and remains, at least initially, at that site even after elaboration of the septum.
To study the role that DivIVA may play at the polar septum, we sought to monitor sporulation in cells harboring a deletion of the divIVA gene using fluorescence microscopy. However, ΔdivIVA cells are severely elongated relative to wild type cells during vegetative growth and display a severe sporulation defect, presumably because the morphology of the cells is so different. Moreover, DivIVA is required at the onset of sporulation to anchor the origins of the replicated chromosomes to the two poles via an anchoring protein called RacA, thereby rendering the straightforward cytological analysis of a simple ΔdivIVA mutant during sporulation difficult. We therefore engineered a strain in which DivIVA could be proteolytically degraded after its role during vegetative growth and chromosome anchoring had finished, but before the polar septum was elaborated. To this end, divIVA at its native locus was replaced with divIVA-FLAG fused to an altered ssrA peptide tag named ssrAEc. Additionally, the sspB gene from E. coli, which encodes an adaptor that specifically delivers SsrAEc-tagged proteins to the ClpXP proteolytic machinery, was produced under the control of an inducible promoter from an ectopic site on the chromosome, so that DivIVA-FLAG-SsrAEc could be specifically degraded upon addition of inducer [52]. The dimensions of cells harboring divIVA-FLAG-ssrAEc were similar to that of wild type (Fig. S1), indicating that this allele of divIVA was functional. Finally, to identify individual cells that had properly entered the sporulation pathway, GFP was produced in this strain under the control of a sporulation-specific promoter (PspoIIG).
Two hours after the induction of sporulation, 80% of otherwise wild type cells (n = 140) had both produced GFP (indicating that they had initiated sporulation) and had elaborated a polar septum (Fig. 2A); similar fractions of cells harboring divIVA-ssrAEc as the only allele of divIVA (73%, n = 142; Fig. 2B) or sspB only, in the absence or presence of the inducer IPTG, (81%, n = 124; and 82%, n = 189, respectively; Fig. 2C–D) both entered the sporulation pathway and elaborated polar septa, indicating that neither the SsrAEc tag on DivIVA, nor the presence of the SspB adaptor affected entry into sporulation or asymmetric division. When expression of sspB was induced 45 minutes after the synchronized induction of sporulation, DivIVA-SsrAEc was largely undetectable in 15 minutes (Fig. S3; compare t60 between +IPTG and −IPTG). Surprisingly, when examined by fluorescence microscopy, only 7% of the cells that had initiated sporulation (evidenced by production of GFP; n = 147) elaborated a polar septum. Interestingly, these cells also displayed a condensed and elongated chromosome architecture as evidenced by DAPI staining that appeared untethered at the poles, similar to that observed in a ΔracA strain ([48]; Fig. 2F: compare the cell marked with white arrow and gray arrow), suggesting that the classically defined “Stage I” of sporulation (in which chromosomes replicate once and condense) had been achieved, but that “Stage II”, in which the polar septum is formed, was blocked. In the absence of inducer, only 42% of these cells (n = 172; Fig. 2E) elaborated a polar septum (we suspect due to leaky expression of sspB that led to overall reduced levels of DivIVA-SsrAEc as seen in Fig. S3), suggesting that SspB-mediated degradation of DivIVA-FLAG-SsrAEc blocked asymmetric division.
To eliminate the possibility that this observed defect in polar septation was an unrelated consequence of the strategy involving the timed degradation of DivIVA-FLAG-SsrAEc, we repeated the experiment in strains harboring a deletion of divIVA, but whose elongation phenotype was suppressed by the additional deletion of minCD [48], [50], [53]. Again, to monitor cells that had entered the sporulation pathway, we introduced a GFP reporter produced under the control of a sporulation promoter. Whereas 90% (n = 120) of otherwise wild type cells and 56% (n = 114) of ΔminCD cells producing GFP formed asymmetric septa (Fig. 2G–H), only 5% (n = 126) of ΔdivIVA cells and 14% (n = 140) of ΔdivIVA ΔminCD cells producing GFP elaborated asymmetric septa (Fig. 2I–J), indicating that the absence of DivIVA resulted in an asymmetric division defect. To test if this defect was due to the inability of RacA to interact with DivIVA, we examined asymmetric septation in cells harboring a deletion in racA. The absence of RacA alone had a modest asymmetric septation defect (Fig. 2K; 75% septation, n = 100). Deletion of both racA and minCD reduced the frequency of asymmetric septation to 32% (n = 125; Fig. 2L), possibly due to an additive effect of removing both MinCD and RacA function, but still the defect was not as severe as the ΔdivIVA defect. Taken together, we conclude that DivIVA plays a previously unappreciated role in the asymmetric placement of the polar septum at the onset of sporulation and that in the absence of DivIVA, cells are arrested at the classically defined “Stage I” of sporulation in which chromosome condensation occurs, but polar septation is prevented.
At the onset of sporulation, FtsZ, the bacterial tubulin homolog that drives membrane constriction during cytokinesis, initially assembles as a ring at mid-cell, but then redeploys towards the two poles and forms two polar-localized rings, at which time only one ring constricts to form the polar septum. This redeployment requires an increase in expression of the ftsZ gene [18], which is mediated by the activation of a second sporulation-specific promoter (“P2”) that is dependent on the Spo0A transcription factor, the master regulator of entry into sporulation [21]. To test if DivIVA affects expression levels of ftsZ, we monitored ftsZ transcription by placing the lacZ gene, which encodes β-galactosidase, under the control of the P2 promoter of ftsZ and measured at different time points the β-galactosidase activity in synchronized sporulating cells that harbored this construct. β-galactosidase activity of the P2 promoter reached its peak between 1 h–1.5 h after the induction of sporulation in otherwise wild type cells (Fig. 3A). In cells harboring a deletion of either minCD alone or divIVA and minCD, the profile of β-galactosidase was nearly identical to that of wild type cells, and β-galactosidase activity in ΔdivIVA cells even continued to rise after t1.5, suggesting that the asymmetric septation defect in ΔdivIVA ΔminCD cells is not due to the absence of the transcription burst of ftsZ during sporulation. Next, we checked the steady state levels of FtsZ protein by immunoblotting cell extracts prepared from various strains of B. subtilis. At the time that sporulation was induced steady state FtsZ protein levels relative to σA in the absence of MinCD, DivIVA, or MinCD and DivIVA were similar to that of wild type, and remained similar even 90 min after the induction of sporulation (Fig. 3B, Fig. S4A). Although the burst in transcriptional activity did not evidently result in a sustained increase in steady state levels of FtsZ protein (measured at a population level at the time points that we tested), the data nonetheless indicate that the steady state levels of FtsZ protein were not reduced in the absence of DivIVA, and that the failure of FtsZ rings to redeploy in the absence of DivIVA is not due to a reduction in FtsZ level.
To test if the asymmetric septation defect in the absence of DivIVA was due to the failure of FtsZ to redeploy to polar sites or due to the inability of FtsZ rings to constrict after redeployment, we monitored the subcellular localization of the division machinery in sporulating cells. To avoid complications arising from the co-production of natively produced FtsZ and ectopically produced FtsZ-GFP, we examined the localization of an FtsZ-associated protein that promotes FtsZ assembly (ZapA) fused to GFP that is frequently used as a proxy for FtsZ localization [54], [55]. ZapA-GFP robustly re-deploys to the polar septum approximately 60 min after the induction of sporulation (Fig. S4B). To identify cells that had initiated sporulation, we also introduced a cassette that expressed mCherry under the control of an early sporulation-specific promoter (PspoIIA). Additionally, to ensure that ΔdivIVA mutants may not simply be slightly delayed in elaborating polar septa, we observed all cells two hours after the induction of sporulation (at a time when polar septation was usually completed in wild type cells and engulfment was initiating) and enumerated the total number of cells harboring either a completed septum or a polar-localized ZapA-GFP. In otherwise wild type cells, two hours after the induction of sporulation, polar septa were elaborated, or were about to be elaborated, as evidenced by ZapA-GFP localization at a polar position (Fig. 3C, arrow; Fig. S4B), in 93% of cells (n = 100) that had produced mCherry. In ΔdivIVA cells, only 5% of sporulating cells produced a polar septum (Fig. 3D), but in ΔminCD cells, approximately one third of cells either elaborated a polar septum or displayed ZapA-GFP at a polar site (Fig. 3E). In ΔdivIVA ΔminCD cells, though, only 5% of cells that initiated sporulation elaborated polar septa (Fig. 3F) at this time point. In almost half of these sporulating cells, ZapA-GFP remained at mid-cell (18%) or immediately adjacent to mid-cell (31%) and had not redeployed to a polar site (Fig. 3F). In the remaining 46%, no ZapA-GFP structure was detected, suggesting that the protein was diffusely localized in the cytosol. We conclude that the asymmetric septation defect in the absence of DivIVA is due to the failure of FtsZ to redeploy and assemble into Z-rings at polar sites.
Since DivIVA typically forms a platform to recruit other proteins to particular subcellular sites (MinJ to division septa, or RacA to the cell poles [44], [45], [48], [56]), we speculated that DivIVA could anchor a sporulation protein at the polar septum. Given the asymmetric septation defect caused by the absence of DivIVA, we wondered if DivIVA could influence the activity of SpoIIE, a transmembrane protein that is involved first in shifting the division septum from the medial to the polar site at the onset of sporulation; and then in activating the first compartment-specific sigma factor, σF, specifically in the forespore [23], [30]. A functional SpoIIE-GFP fusion initially localizes as a ring at mid-cell, likely via a direct interaction with FtsZ, and then redeploys to polar sites with FtsZ ([18]; Fig. 4A; Fig. S1D). However, unlike FtsZ, which constricts during septation and eventually dissipates in the cytosol after membrane constriction is finished, SpoIIE remains at the polar septum until septum formation is complete and is then redistributed throughout the forespore membrane to perform its second function in σF activation [29]. However, the mechanism by which it initially persists at this site is not known [15], [26], [27], [29]. In 66% of observed sporulating cells (n = 161), SpoIIE-GFP persisted roughly uniformly along the entire length of the polar septum. In the remaining 34%, we observed two separate populations of SpoIIE at the polar septum: one near the center of the polar septum and another that remained near the lateral edge of the cell (Fig. 4A, cell 1). To see if SpoIIE co-constricts with FtsZ at the polar septum, we examined the localization of both proteins in sporulating cells producing FtsZ-mCherry and SpoIIE-GFP. At the onset of asymmetric division, both FtsZ-mCherry and SpoIIE-GFP co-localized as bands whose widths were approximately equal to the width of the cell (Fig. 4C, top row, arrowhead). However, as FtsZ constricted to form the polar septum, FtsZ-mCherry localized as a band whose width was less than the width of the cell, whereas in 91% of these observed cells (n = 100) SpoIIE-GFP remained as a band whose width was approximately equal to the width of the cell (Fig. 4C, arrow). Reconstruction of deconvolved Z-stacks revealed that, before FtsZ constriction initiated, both FtsZ-mCherry and SpoIIE-GFP assembled into ring-like structures (Fig. 4C, bottom row, arrowhead), but upon constriction of FtsZ at the polar septum, FtsZ-mCherry formed a disk-like structure, whereas SpoIIE-GFP remained as a ring-like structure with an outer diameter that was larger than that of FtsZ-mCherry (Fig. 4C, arrow). This localization pattern of SpoIIE was similar to that observed for DivIVA-CFP at the polar septum (Fig. 1E) and consistent with a model in which FtsZ constricts, while DivIVA and SpoIIE remain associated (SpoIIE albeit briefly) at the lateral edge of the polar septum.
To test if SpoIIE and DivIVA interact in vivo, we constructed a cell that produced, under the control of their native promoters, a functional DivIVA with a C-terminally appended FLAG tag (Fig. S1) in addition to the untagged version of DivIVA; as well as SpoIIE-GFP (also functional in vivo; (Fig. S1D; [27]). We then purified DivIVA-FLAG, using anti-FLAG antibodies, from detergent-solubilized cell extracts and examined various fractions collected during purification by immunoblotting. Although DivIVA localizes to the membrane, its association with the membrane is likely tenuous [57]; accordingly it appeared largely in the soluble fraction of cell extracts even without the inclusion of detergent (Fig. S5). However, the polytopic membrane protein SpoIIE was initially insoluble, but was effectively solubilized by addition of detergent (Fig. S5). The results in Fig. 4D indicate that SpoIIE-GFP co-purified with DivIVA-FLAG, as did the native, untagged version of DivIVA. As a negative control, the unrelated protein σA was not retained on the column. To ensure that an unrelated membrane-associated protein did not co-purify with DivIVA-FLAG, we repeated the experiment using a strain that overproduced the membrane-associated protein SpoVM-GFP, and observed that SpoVM-GFP also did not co-purify with DivIVA-FLAG. Furthermore, when the purification was performed with cell extract which did not produce DivIVA-FLAG, neither SpoIIE-GFP nor DivIVA was retained on the column, suggesting that retention of SpoIIE-GFP was specifically dependent on the presence of the FLAG-tagged DivIVA. Finally, we performed the reciprocal pulldown experiment in which we purified functional SpoIIE-FLAG (Fig. S1D) and observed that DivIVA, but not σA, co-purified with SpoIIE-FLAG. As a negative control, when the SpoIIE-GFP was purified with anti-FLAG antibodies, SpoIIE-GFP, DivIVA, and σA were not retained on the column, suggesting that the specific co-purification of DivIVA was mediated by SpoIIE-FLAG. We therefore conclude that SpoIIE and DivIVA interact with each other in vivo and that this interaction likely takes place at the polar septum.
To test if DivIVA localization at the polar septum is dependent on SpoIIE, we examined the localization of DivIVA-GFP in sporulating cells in the absence of SpoIIE. Even though SpoIIE is involved in shifting the septum formation site to the polar position during sporulation, deletion of spoIIE does not completely abolish asymmetric division and about 30% of the cells elaborate a polar septum [18]. In cells harboring a deletion of spoIIE, DivIVA-GFP localization was unaffected and it continued to localize at the polar septum (Fig. 4B). Taken together, we conclude that DivIVA is in complex with SpoIIE, and that the localization of DivIVA to the polar septum does not depend on SpoIIE.
Next, we examined the subcellular localization of SpoIIE-GFP, produced under control of its native promoter, in the presence and absence of DivIVA. In otherwise wild type cells, SpoIIE was found at the polar septum or at a potential site of asymmetric septation in 91% of cells 1.5 h after the induction of sporulation (Fig. 4E). ΔdivIVA cells, though, due to the defect in polar septation, displayed SpoIIE-GFP in only 12% of cells (Fig. 4F). Once polar septation was restored by deletion of minCD, nearly half of the cells displayed SpoIIE-GFP at the polar septum (Fig. 4G). However, in ΔdivIVA ΔminCD cells, again due to the defect in polar septation, SpoIIE-GFP was found at the polar septum in only 6% of cells (Fig. 4H). Rather, like the localization of ZapA (and by extension, FtsZ), in a majority of cells SpoIIE-GFP was observed either at mid-cell (25%) or at a site immediately adjacent to mid-cell (69%). To ensure that the SpoIIE localization defect in the absence of DivIVA was not due to a defect in transcription levels of the spoIIE gene, we placed the lacZ gene under the control of the spoIIE promoter and measured β-galactosidase activity in sporulating cells at different time points. The results in Fig. 4M indicated spoIIE transcription was largely unaffected in cells harboring a deletion in divIVA, minCD, or both, as compared to wild type. Thus, in the absence of DivIVA, SpoIIE failed to redeploy from mid-cell to its customary polar positions.
Does the interaction between SpoIIE and DivIVA play a role in transiently sequestering SpoIIE at the polar septum? Whereas the dependence of DivIVA-GFP localization on SpoIIE was readily measured by deleting the spoIIE gene (Fig. 4B), the converse experiment was not straightforward to perform, since deletion of divIVA resulted in the failure to elaborate polar septa (Fig. 4H). We therefore produced SpoIIE-GFP under the control of an inducible promoter in vegetative cells, in the absence of other sporulation factors, and examined its localization either in the presence or absence of DivIVA. When produced in vegetatively growing wild type cells, SpoIIE localized to division septa and persisted at 82% (n = 146) of mature septa after cytokinesis had completed ([26]; Fig. 4I). It should be noted, however, that we observed the eventual release of SpoIIE from mature septa (unlike DivIVA) in wild type cells after approximately 2–3 cell generations, suggesting that an interaction between SpoIIE and DivIVA, even in vegetative cells, may be transient. In cells harboring a divIVA deletion, SpoIIE readily localized to future division sites (presumably dependent on FtsZ), but persisted at only 12% (n = 122) of mature septa (Fig. 4J). To ensure that this reduction in localization was not due to the infrequent septum formation, we suppressed the cell elongation phenotype of the ΔdivIVA strain by introducing a deletion in minCD [48], [50], [53]. In ΔminCD ΔdivIVA cells, cells were approximately of wild type length and septa were elaborated much more frequently. In these cells, SpoIIE-GFP still failed to persist at mature septa (19%, n = 100; Fig. 4L). As a control, the deletion of minCD alone did not abolish the persistence of SpoIIE at division septa (43%, n = 100; Fig. 4K). Taken together, we conclude that although DivIVA is not required for the initial recruitment of SpoIIE to the future site of cell division (consistent with a model in which FtsZ initially recruits SpoIIE [16], [17]), the transient persistence of SpoIIE at mature septa after cytokinesis has finished depends on DivIVA.
To verify if the absence of DivIVA affects the second function of SpoIIE, which is to activate σF specifically in the forespore, we used a strain in which gfp was under the control of a σF-controlled promoter (PspoIIQ) and monitored the forespore-specific production of GFP in the presence or absence of DivIVA. In an otherwise wild type strain, 97% (n = 131) of cells that produced GFP produced it exclusively in the forespore (Fig. 5A). In cells harboring a deletion of divIVA, of the cells that produced GFP, 95% (n = 108) of them displayed uncompartmentalized activation of σF (Fig. 5B). Since ΔdivIVA cells are morphologically so dissimilar to sporulating wild type cells, we also examined σF activation in ΔminCD ΔdivIVA cells. In this strain, only 6% of GFP-producing cells displayed forespore-specific activation of σF (n = 106), whereas deletion of minCD alone resulted in proper forespore-specific σF activation in 71% (n = 143) of the cells (Fig. 5C–D), suggesting that DivIVA is required for compartment-specific activation of σF. This unspecific activation of σF in the absence of DivIVA was unlike the phenotype seen in the absence of other division factors, such as FtsZ, FtsA, DivIC, and FtsL, in which asymmetric division was impaired, but σF activation was prevented as well [58]–[61]. Analysis of a σF responsive promoter (PspoIIQ) fused to lacZ indicated that the total amount of σF activity at a population level was not significantly affected in the absence of DivIVA (Fig. S6A). Thus, the absence of DivIVA instead primarily affected the compartment specificity of σF activation.
A previous study had reported that a spoIIEV697A mutant allele caused premature activation of σF, which in turn resulted in an asymmetric septation defect in that strain [62]. Does the uncompartmentalized activation of σF in the absence of DivIVA, then, result in the asymmetric septation defect that we observed in these strains? If so, then deletion of sigF should correct the asymmetric septation defect in the absence of DivIVA. We therefore introduced a sigF deletion in strains that also harbored a divIVA deletion and monitored asymmetric septation by fluorescence microscopy. Since sigF deletion results in a disporic phenotype in which two polar septa are elaborated, we grouped cells containing either one or two polar septa together in one category as being able to form at least one polar septum. At 1.5 hours after the induction of sporulation, wild type cells elaborated a polar septum in 67% of cells (n = 100; Fig. S6B); in the absence of σF, 51% of the cells were still able to elaborate at least one polar septum at this time point (n = 100; Fig. S6C). In a ΔminCD ΔsigF strain 46% of the cells (n = 100; Fig. S6D) were able to form polar septa, a similar fraction as the ΔsigF strain (minicells were distinguished from forespores by observing the presence of DNA in forespores using DNA stain). In contrast, in the absence of DivIVA and σF there were no observable polar septa, and in a ΔminCD ΔdivIVA ΔsigF strain only 7% of the cells displayed asymmetric septation (Fig. S6E–F; n = 100), suggesting that deletion of ΔsigF was not able to suppress the polar septation defect caused by the absence of DivIVA. We therefore conclude that DivIVA is required for the forespore-specific activation of σF, and that the defect in polar septation observed in the absence of DivIVA is not due to the promiscuous activation of σF in a compartment-unspecific manner.
The biochemical basis of how SpoIIE activates σF has been extensively studied [23], [35]–[37], but the cell biological mechanism underlying how it exerts this function in a forespore-specific manner has remained largely unclear. Recently, Guberman et al. developed an algorithm which refines diffraction-limited images by interpolating the space between adjoining pixels and concluded that SpoIIE preferentially localizes to the forespore side of mature polar septa. To more directly visualize the localization of SpoIIE and DivIVA both at mature polar septa, and those septa that are just beginning to form, we employed a super-resolution fluorescence microscopy technique called Structured Illumination (SIM), which can potentially increase the resolution of fluorescence microscopy two-fold [63]. Indeed, this increase in resolution has previously allowed us to distinguish the localization of DivIVA-GFP on either side of a ∼80 nm wide division septum in vegetative cells [40].
We first observed the localization pattern of DivIVA-GFP and SpoIIE-GFP using a commercially available SIM setup (DeltaVision OMX). As shown in Fig. 6A, membrane invagination at the onset of asymmetric division could be visualized by the increase in membrane staining near a pole along the lateral edges of the cell. DivIVA-GFP, produced under the control of its native promoter, localized initially on either side of this site of membrane invagination, similar to its reported behavior at nascent vegetative septa [40]. However, shortly after the completion of polar septation, DivIVA-GFP was found preferentially on the forespore side of the septum- a preference that persisted even after the onset of engulfment when the polar septum began to curve (Fig. 6B–C). This localization pattern was unlike the localization displayed by DivIVA-GFP during vegetative division, where DivIVA-GFP localized on both sides of a division septum and persisted on both sides long after septation was completed. The mean fluorescence intensity of polar DivIVA patches in cells harboring an adjacent polar septum (8286±3321, n = 50) was similar to that of cells harboring no adjacent septa (9046±3072, n = 50), consistent with a scenario in which no significant net exchange of DivIVA molecules appears to take place from the pole to the polar septum (compare patches in cell 1 and cell 2 in Fig. 1C).
We next observed the localization of SpoIIE-GFP under the control of its endogenous promoter in sporulating cells using the same SIM setup. At nascent polar septa, we observed that in many cells SpoIIE-GFP preferentially localized on the face of the polar septum that abutted the forespore (Fig. 6D). After septum formation was completed (Fig. 6E), SpoIIE localized to the forespore side of the septum, and finally, as reported previously, SpoIIE was released from the polar septum and localized uniformly in the membrane surrounding the forespore (Fig. 6F; [29]).
We were initially unable to properly quantify a large enough dataset of cells displaying the forespore-preferential localization of SpoIIE due to a laser-induced phototoxicity effect, using several different commercially available SIM setups, which deformed many of the polar septa that we observed (examples are shown in Fig. S7A). Curiously, we had not observed such severe phototoxic effects when examining vegetative septa. We therefore used a different implementation of this super-resolution technique called MSIM [64] that greatly reduced this phototoxic effect. To observe initial SpoIIE-GFP localization in a larger number of cells, we used a previously described engulfment-deficient strain (ΔspoIIDM) in which the polar septum remains flat [24]. Using MSIM, we again observed that DivIVA-GFP at nascent asymmetric septa localized initially on both forespore and mother cell sides (Fig. 6G). After polar septation was complete, in 81% of sporulating cells (n = 22), DivIVA-GFP persisted only on the forespore side of the mature septum (Fig. 6H), whereas in only 14% of cells, DivIVA-GFP localized on both sides of the polar septum.
In the case of SpoIIE-GFP, MSIM revealed that SpoIIE preferentially localized on the forespore side in both nascent and mature asymmetric septa (Fig. 6I–J; Fig. S9)- the shift in the GFP channel towards the forespore side was revealed more readily by the linescan graph of normalized fluorescence intensity of both membrane stain and SpoIIE-GFP at the septum. In total, 58% displayed clear forespore side specific SpoIIE-GFP localization (n = 38), while the remaining 42% displayed ambiguous localization in which the GFP channel largely overlapped that of the membrane stain. Of note, we did not detect any cells in which localization of SpoIIE-GFP was preferentially on the mother cell side of the polar septum.
One complication of this analysis, though, was that we were unable to determine if the forespore-proximal localization of SpoIIE-GFP in the cells occurred immediately upon septation or if we were imaging the cells after SpoIIE-GFP was released into the forespore membrane and subsequently recaptured. In order to eliminate from our analysis those cells that had recaptured SpoIIE-GFP, we examined SpoIIE-GFP localization in the absence of SpoIIQ, the protein responsible for the recapture of SpoIIE to the polar septum [29]. Whereas SpoIIE-GFP localized exclusively to the flat polar septum in 81% of engulfment-defective (ΔspoIIDM) sporulating cells (n = 100), in the absence of SpoIIQ, SpoIIE-GFP localized to the polar septum in only 52% of the sporulating cells (n = 118). In the remaining cells, SpoIIE-GFP was uniformly distributed around the forespore membrane, consistent with the inability of these cells to recapture SpoIIE at the polar septum ([29] (Fig. S8)). To verify on which face of the polar septum SpoIIE localized to initially before release into the forespore membrane, we employed yet another implementation of SIM, termed instant structured illumination microscopy (ISIM). This technique eliminates the need for the digital post-processing required in other SIM implementations, directly providing a ∼1.4-fold increase in resolution in the raw images, followed by subsequent deconvolution which then provides the full 2-fold resolution improvement relative to conventional widefield fluorescence imaging, allowing us to collect super-resolution images more rapidly [65]. Similar to the MSIM results, when using ISIM, DivIVA-GFP initially localized to both sides of the polar septum in 67% (n = 36) of engulfment-defective (ΔspoIIDM) sporulating cells (Fig. S10), while the remaining 33% displayed forespore-proximal localization. However, 76% of mature polar septa displayed DivIVA-GFP on the forespore-proximal side, while in the the remaining 24%, DivIVA-GFP was found on both sides (Fig. S10). Next, we examined SpoIIE-GFP localization using ISIM. SpoIIE-GFP appeared to localize preferentially on the forespore-side of the asymmetric septum (Fig. 6K–L) in engulfment-deficient (ΔspoIIDM) cells. In cells that additionally harbored a deletion of spoIIQ, in those cells which did not release SpoIIE-GFP into the forespore membrane, we observed that SpoIIE-GFP localized preferentially to the forespore-side in 64% (n = 45) of nascent polar septa and 67% (n = 69) of mature polar septa (Fig. 6M–O; Fig. S9), consistent with a model in which SpoIIE preferentially localizes to the forespore-proximal face of the polar septum before being released into the forespore membrane.
To further ensure that the forespore-biased localization of DivIVA-GFP and SpoIIE-GFP were not due to erroneous image registration by the image processing software, or due to variations in the microscope stage when the positions of emission filters changed, we conducted the experiments with fluorescent beads that fluoresce at both red and green wavelengths and aligned the beads' fluorescence in both channels as an internal control. In the example in Fig. S7B, the arrowhead indicates the location of a bead adjacent to a sporulating cell containing a nascent asymmetric septum. As shown in the overlay, there was no shift between red and green channel for the bead, whereas SpoIIE-GFP in the sporulating cell lying adjacent to the bead was preferentially localized to the forespore-proximal side of the polar septum (Fig. S7B). We conclude that DivIVA initially localizes to either side of nascent polar septa and localizes preferentially to the forespore side once septation is complete. SpoIIE, however, preferentially localizes to the forespore-proximal side of the polar septum from the very onset of membrane invagination and remains associated with the forespore-proximal face of the polar septum until it is released into the membranes surrounding the forespore.
DivIVA of B. subtilis is an extensively studied protein made of coiled-coil domains that resembles eukaryotic tropomyosins [53], [66], and has two well known functions. First, during vegetative growth, DivIVA arrives at mid-cell shortly after cytokinesis initiates and recruits the components of the Min system to mid-cell to prevent aberrant septation on either side of the site of membrane constriction [40], [44], [45], [47]. It has been proposed that an increase in negative membrane curvature, which arises on both sides of the division septum as the membrane constricts, drives this localization of DivIVA to mid-cell [41], [42]. Second, during sporulation, DivIVA localizes to the extreme cell poles and anchors the two origins of replication of the replicated chromosomes to each pole so that the forespore and mother cell each ultimately receive one copy of the chromosome [48]–[50]. In this report, we demonstrate a third role for DivIVA in which it localizes to the polar septum during sporulation and resides in complex with a multifunctional transmembrane protein called SpoIIE. SpoIIE is initially recruited to mid-cell at the onset of sporulation by FtsZ, after which it is required for the efficient redeployment of the Z-ring from mid-cell to polar positions to initiate asymmetric division. After asymmetric division commences, SpoIIE is initially recruited to the polar septum by FtsZ, but a significant population of SpoIIE remains associated with the polar septum even after FtsZ constricts and leaves that location. Shortly thereafter, SpoIIE is released exclusively into the forespore membrane to perform its second function in activating the first forespore-specific transcription factor σF [23], [30]. Subsequently, SpoIIE is recaptured at the forespore face of the polar septum [29] where it may participate in remodeling the peptidoglycan at the polar septum [22]. In the absence of DivIVA, we found that the two main functions of SpoIIE were disrupted, in that asymmetric septation did not occur, and that σF was prematurely activated in a compartment-unspecific manner. Interestingly, the deletion of several other cell division genes have only been reported to cause an asymmetric cell division defect, but not the premature activation of σF [58]–[61]. We therefore propose that, in addition to SpoIIE, DivIVA is required for the redeployment of FtsZ to the polar division sites and for the compartment-specific activation of σF in the forespore.
Our studies began with the observation that DivIVA localizes to the polar septum during sporulation. In an effort to determine the function of DivIVA at this location, we deleted divIVA and suppressed the elongation defect caused by this deletion by introducing a deletion in minCD as well. As previously observed, deletion of minCD alone had a mild defect in asymmetric division [67]–[70], but the additional deletion of divIVA resulted in the nearly complete abrogation of asymmetric division. That this defect was not primarily dependent on the absence of MinCD was established in an experiment in which we achieved the controlled degradation of DivIVA alone at a time point after it had completed its chromosome anchoring function, but before cells had initiated asymmetric division. Removal of DivIVA alone in this manner similarly abrogated asymmetric division at the onset of sporulation. Ultimately, we observed that the defect in asymmetric septation in the absence of DivIVA was due to the inability of these cells to redeploy FtsZ and SpoIIE from medial to polar positions. Interestingly, although polar septation was not directly measured at the time, Cha and Stewart had hypothesized that DivIVA may play an active role in the asymmetric positioning of the polar septation in the very first report of the divIVA locus in B. subtilis [68]. Additionally, a recent report suggested that a severe sporulation defect in a strain in which Spo0A rapidly accumulated at the onset of sporulation was perhaps due to the premature suppression of divIVA expression, consistent with our hypothesis that DivIVA plays an additional, indispensible role during sporulation [71]. Previously, Ben-Yehuda and Losick had demonstrated that, in vegetatively growing B. subtilis, the slight overproduction of FtsZ and the production of SpoIIE were sufficient to generate polar septa [18]. We propose that, in this context, DivIVA, which is abundant during vegetative growth, is also required for the redeployment of FtsZ to generate polar septa during vegetative growth.
It is currently unclear to us how DivIVA mechanistically mediates asymmetric division. At the onset of sporulation, the replicated chromosomes are extensively remodeled so that it may form the “axial filament” structure that extends from one pole to the other in what has traditionally been named “Stage I” of sporulation. Interestingly, in the absence of DivIVA, cells appeared to be arrested at this stage which, to our knowledge, is only the second description of a gene whose deletion results in a so-called “Stage I” arrest, not simply a delay [72]. Proper chromosome segregation has also been implicated in formation of the polar septum, in a manner that depends on the DNA-binding protein Spo0J [73]. Interestingly, DivIVA was shown to interact with Spo0J in a sporulation-specific manner, and independent of the Min system [74], [75]. Recently, the absence of RefZ, a DNA-binding protein that binds to origin- and terminus-proximal regions of the chromosome was reported to delay asymmetric septation, and it was proposed that RefZ may either facilitate repositioning of FtsZ at a polar position or promote disassembly of FtsZ at mid-cell sites [76]. It is therefore conceivable that the asymmetric septation impairment in the absence of DivIVA is a result of defective chromosome remodeling or segregation. It is also possible that the asymmetric septation defect in the absence of DivIVA may be through a metabolic pathway that affects septum formation. Deletion of citC, which encodes isocitrate dehydrogenase, also resulted in inhibition of asymmetric septation [72]. Loss-of-function mutations in the spoVG gene suppressed this defect, and overproduction of SpoVG in otherwise wild type cells resulted in delayed σF activation [77]. Curiously, SpoVG homologs were recently shown to be site-specific DNA-binding proteins [78], thereby possibly providing a link between asymmetric septation and chromosome remodeling or segregation.
In addition to its less understood role in redeploying the Z-ring from mid-cell to polar sites, SpoIIE is required for the activation of σF, the biochemical basis of which has been well studied [23], [30]–[32], [34]. However, the cell biological basis for the forespore-specific activation of σF has been less well understood for two reasons. First, although FtsZ is thought to recruit SpoIIE to the polar septum, it had been unclear how SpoIIE remains associated with the polar septum after FtsZ constricts at that site and ultimately dissipates from that location, or how FtsZ may selectively release SpoIIE into the forespore membrane upon completion of constriction. We have observed a previously unappreciated step in which SpoIIE forms a ring-like structure at the polar septum during FtsZ constriction, before the release of SpoIIE into the forespore membrane. Based on our observation that DivIVA and SpoIIE co-localize at the polar septum and the discovery that DivIVA and SpoIIE copurify with each other, we propose that DivIVA is required for this initial sequestering of SpoIIE at the polar septum (Fig. 7). Consistent with this model, SpoIIE, when produced in vegetative cells, stably associated with vegetative division septa only in the presence of DivIVA. In sporulating cells, in the absence of DivIVA, polar septa were not efficiently formed and σF was prematurely activated in a compartment-unspecific manner. Second, the mechanism by which SpoIIE exerts its activity specifically in the forespore was also unclear. SpoIIE is released into the forespore membrane and it is promptly recaptured, nearly quantitatively, at the polar septum, suggesting that SpoIIE is exclusively released into the forespore, and not into the mother cell membrane [29]. Another report suggested, based on the interpolation of fluorescence signal in diffraction-limited images, that at mature polar septa SpoIIE was preferentially detected on the forespore side of the polar septum [39] -this localization of SpoIIE was likely after its recapture at the polar septum. In this report, we examined cells that displayed nascent polar septa, as well as cells that had elaborated mature polar septa, using three super-resolution techniques called SIM, MSIM, and ISIM, and observed that the biased localization of SpoIIE on the forespore side could be observed at the very onset of membrane invagination. In contrast, we observed DivIVA initially on either side of nascent polar septa, similar to its localization pattern in vegetative division septa. However, unlike its behavior in vegetative cells where it was ultimately equally distributed on both sides, we observed that at mature polar septa, DivIVA, like SpoIIE, also preferentially localized to the forespore side. The mechanism for this shift in distribution is currently unclear. It may, for example, involve the selective degradation of DivIVA on the mother cell face of the polar septum or may involve a transfer of pole-localized DivIVA molecules to the forespore-proximal side of the polar septum which we were unable to measure using our current experimental setup. Such a transfer has indeed been recently reported for vegetative septa, which may occur in 5–20 mins [79]. Curiously, the presence of Min proteins, which are typically recruited by DivIVA, on the forespore-proximal face of the polar septum has been implicated in maintaining the polarity of DNA translocation [80]. The basis for the early bias in localization of SpoIIE is currently not known, but this pattern highlights the establishment of asymmetry, at a compartment-specific level, in the developing sporangium long before polar septation is completed. We propose that the interaction of SpoIIE with DivIVA allows for its retention at the forespore side of the newly forming polar septum until septation is complete. After completion of polar septum formation, similar to the release of MinCD after completion of vegetative septa [47], SpoIIE may be liberated to redistribute along the membrane surrounding the forespore. It is tempting to speculate that the initial interaction of SpoIIE with DivIVA may inhibit the phosphatase activity of SpoIIE until it is released into the forespore membrane, which may also explain the premature activation of σF in the predivisional cell in the absence of DivIVA. Curiously, the N-terminus of MinC shares structural similarity with that of SpoIIAA, the protein that is dephosphorylated by SpoIIE [81]. Thus, once released into the forespore membrane, the phosphatase activity of SpoIIE may be uninhibited and σF may be activated. Therefore, asymmetrically placed DivIVA may act as a molecular beacon that signals the completion of septum formation for the subsequent activation of σF. An outstanding question, though, will be to determine how the asymmetry of SpoIIE is established at such an early time point before the mother cell and forespore compartments are even separated.
All strains used in this study are congenic derivatives of B. subtilis PY79 [82]. Genotypes of strains used are provided in Table S1. B. subtilis competent cells were prepared as described previously [83]. β-galactosidase activity of cell samples collected at time points indicated was measured with modifications as described previously [84], [85]. To express divIVA-linker-cfp from the amy locus, PdivIVA-divIVA without the stop codon was PCR amplified from PY79 chromosomal DNA using primers that abutted HindIII and NheI sites (“oP10” 5′-AAAAAGCTTTCGTGTTTTCTGAGACA and “oP11” 5′-AAAGCTAGCTTCCTTTTCCTCAAATAC). linker-cfp was PCR amplified from DS4152 chromosomal DNA [44] using primers abutting NheI and BamHI sites (“oP54” 5′-AAAGCTAGCGGT TCCGCTGGCTCCGCTGCTGGT TCTGGCCTC and “oP55” 5′-AAAGGATCCTTACTTATAAAGTTCGTCCATGCCAAGTGTAATGCC). Both fragments ligated into integration vector pDG1662 to create pPE17. To express divIVA-linker-gfp from the amy locus, pPE17 was digested with HindIII and XhoI to liberate PdivIVA-divIVA and part of the linker sequence. Next, gfp mut2 was PCR amplified from pKC2 [86] using primers that abutted the remaining linker sequences, and 5′ XhoI and 3′ BamHI sites (“oP62” 5′-AAACTCGAGGGTTCCGGAATGAGTAAAGGAGAAGAACTTTTC and “oP47” 5′-AAAGGATCCTTATTTGTATAGTTCATCCATGCC). Both fragments were ligated into pDG1662 to create pKR227. To replace divIVA at the native locus with divIVA-FLAG-ssrAEc, the final 400 nucleotides of divIVA (omitting the stop codon) were PCR amplified using primers that abutted BamHI and XbaI sites (“DivIVA-C4005′Bam” 5′-AAAGGATCCAGTCAGAAAAGATTACGAAATTG and “DivIVA-C4003′ Xba” 5′ AAATCTAGATTCCTTTTCCTCAAATACAGCG), and ligated into Campbell integration vector pKG1268 [52] to create pKR226. To express divIVA-FLAG from amyE, PdivIVA-divIVA was PCR amplified from PY79 chromosomal DNA using primers that abutted 5′ HindIII and 3′ BamHI sites, as well as a 3′ FLAG tag sequence (“DivIVAprom5′Hind” 5′-CCCAAGCTTTCGTGTTTTCTGAGACAGCAG and “DivIVA3′FLAGBam” 5′-CGCGGATCCTTACTTGTCGTCATCGTCTTTGTAGTCTTCCTTTTCCTCAAATACAG), and ligated into pDG1662 to create pKR200. To express spoIIE-gfp from amyE under the control of an IPTG-inducible promoter, spoIIE-gfp was PCR amplified from SB201 [87] using primers that abutted SalI and SphI sites (“oP44” 5′- AAAGTCGACACATAAGGAGGAACTACTATGGAAAAAGCAGAAAGAAGAGTGAACGGG and “oP24” 5′- GCCGCATGCTTATTTGTATAGTTCATCCATGCC), and ligated into integration vector pDR111 to create pPE19. To express IPTG-inducible spoIIE-FLAG from amyE, plasmid pPE48 was created by appending 3×-flag tag sequence (GATTATAAGGATCATGATGGTGATTATAAGGATCATGATATCGACTACAAAGACGATGACGACAAG) followed by a stop codon to the 3′ end of the spoIIE coding sequence in pPE19 via QuikChange mutagenesis (Agilent). Strain KR610 (ΔspoIIE::tet) was created by the long flanking homology method [88] using primers (“spoIIEKO-1” 5′-GCAAGTAGCCTTGTTGACAC, “spoIIEKO-2” 5′-CAATTCGCCCTATAGTGAGTCGTTCCTCTCATCTCCCACCTG, “spoIIEKO-3” 5′-CCAGCTTTTGTTCCCTTTAGTGAGCGCTTCCGTATAAATCAAATTTC, and “spoIIEKO-4” 5′-TTTCAAGACATTCACTTCAGAAG). For immunoblot analysis, cells were grown as described below, harvested, and cell extracts for immunoblot analysis were prepared by lysozyme treatment as described previously [89]. Extracts were separated by SDS-PAGE and immunoblotted using antisera raised against purified DivIVA-GFP, σA, GFP (Covance, Inc.), or E. coli FtsZ (courtesy of Sue Wickner; [90]) as indicated. Where indicated, band intensities were quantified using ImageQuant software (GE).
B. subtilis overnight cultures grown at 22°C in casein hydrolysate (CH) medium were diluted 1∶20 into fresh CH medium and grown until OD600 reached ∼0.5 at 37°C. For induction of sporulation, cells were spun down and resuspended in Sterlini and Mandelstam (SM) medium as described previously [91], for the time indicated. Sporulation efficiency was measured by growing cells in Difco sporulation medium (DSM; KD Medical) for at least 24 h at 37°C. The number of heat-resistant colony forming units (cfu) was obtained after incubation at 80°C for 10 min. For strains harboring genes that disrupted the thr locus, L-threonine (40 µg/ml final concentration) was added to the culture as a nutritional supplement. Where specified, IPTG (1 mM final concentration) was added at the indicated time to induce genes; xylose (0.5% final concentration), where indicated, was added at the time of resuspension.
Cells were visualized as described previously [92]. Briefly, culture pellets (from 1 mL culture) that were grown as described above was washed with PBS and resuspened in 50–100 µL PBS containing 1 µg/ml (final concentration) of the fluorescent dye FM4-64 or 46 µg/ml (final concentration) TMA-DPH to visualize membranes and, where indicated, 2 µg/ml (final concentration) of DAPI to visualize DNA. Cells (5 µl) were then placed on a poly-L-lysine coated glass bottom culture dish (Mattek Corp.; poly-L-lysine did not appreciably affect localization of DivIVA-GFP at the polar septum (Fig. S2A)). A pad made of 1% agarose in distilled water (or SM media containing IPTG and FM4-64 for time lapse microscopy) was cut to size and placed above the cells. Cells were viewed at room temperature (or 32°C for timelapse) with a DeltaVision Core microscope system (Applied Precision) equipped with a Photometrics CoolSnap HQ2 camera and an environmental chamber. Seventeen planes were acquired every 200 nm at room temperature and the data were deconvolved using SoftWorx software. Imaris software was used for three dimensional surface rendering of fluorescence data. For 3D-SIM, cells were prepared as described above and imaged using Delta Vision OMX Blaze (Applied Precision). For MSIM, cells (with fluorescent beads when indicated) were labeled with FM4-64 as described above and placed on top of a glass slide and a freshly prepared poly-L-lysine coated coverslip (#1.5 thickness, VWR) was placed on top of the cell suspension; coverslips and slides were cleaned as described previously [93]. Fluorescence images were acquired using a custom MSIM system equipped with a 60× 1.45NA oil objective (Olympus) and appropriate filters (Chroma, zt405/488/561 (dichroic) LP02-488RE-25, NF03-561E-25 (emission filters). Total exposure times for each 2D slice were either 1 s or 2 s depending on the signal intensity. 2D slices were acquired every 200 nm along the Z-axis for construction of 3D volumes, comprising a total axial range of 10–12 µm for each sample. The longer wavelength channel (red) was collected prior to the shorter wavelength channel (green) for all samples. Multicolor 100 nm diameter Tetraspeck microspheres (Life Technologies, T-7279) were immobilized on the coverslip surface to enable precise alignment of each image channel. Alignment of the two channels was completed by translating the red channel relative to the green channel to maximize overlap of the reference microspheres using ImageJ. Images were deconvolved with a 3D Richardson-Lucy algorithm implemented in the python programming language (40 iterations, using a Gaussian PSF with x, y, z FWHM values 3.8×3.8×2.4 pixels, code available at code.google.com/p/msim) [64]. ISIM imaging was performed using a previously described system [65]. Cells were prepared as described above. Fluorescence from labeled membranes was excited at 561 nm and GFP signal from fusion proteins was excited at 488 nm. A 525 nm bandpass filter (Semrock, FF01-525/50-25) was used for GFP collection and a 561 nm notch filter (Semrock, NF03-561E-25) was used for collecting membrane fluorescence. Exposure times for imaging the membrane probe were 40 ms or 80 ms (depending on sample brightness) and 400 ms for imaging GFP fusion constructs. 3D stacks were collected with a z step size of 200 nm. Resulting images were deconvolved using Richardson-Lucy deconvolution, and image channels were aligned using the multicolor beads for reference as described above. Line scans were completed using the line scan function in ImageJ, setting the line width to 5-pixels. Line scan measurements were exported to Microsoft Excel for rendering and display.
FLAG-tagged proteins were immunoprecipitated from strains coproducing FLAG- and GFP-tagged proteins using the FLAG Immunoprecipitation kit (Sigma). Samples were processed largely as described previously [84]. Briefly, a 20 ml culture of cells was induced to sporulate for 1.5 h, harvested, resuspended in 500 µl protoplast buffer (0.5 M sucrose, 20 mM MgCl2, 10 mM potassium phosphate [pH 6.8], 0.1 mg/ml lysozyme), and incubated at 37°C for 15 min to remove cell wall [94]. Protoplasts were harvested and cell extracts were prepared by resuspension in 1 ml of lysis buffer (50 mM Tris [pH 7.4], 150 mM NaCl, 1 mM EDTA, 1% Triton X-100). 100 µl of the extract was retained for analysis as the “load” fraction. The remaining cell extract was added to 20 µl of anti-FLAG affinity resin and incubated overnight at 4°C with light shaking. The antibody resin was centrifuged and supernatant was collected as the “unbound” fraction. The resin was then washed extensively with lysis buffer. Proteins associated with the resin were eluted through competitive elution with FLAG peptide-containing lysis buffer. Fractions were analyzed by immunoblotting using specific antisera as described above.
|
10.1371/journal.pgen.1007617 | Tudor-domain containing protein 5-like promotes male sexual identity in the Drosophila germline and is repressed in females by Sex lethal | For sexually reproducing organisms, production of male or female gametes depends on specifying the correct sexual identity in the germline. In D. melanogaster, Sex lethal (Sxl) is the key gene that controls sex determination in both the soma and the germline, but how it does so in the germline is unknown, other than that it is different than in the soma. We conducted an RNA expression profiling experiment to identify direct and indirect germline targets of Sxl specifically in the undifferentiated germline. We find that, in these cells, Sxl loss does not lead to a global masculinization observed at the whole-genome level. In contrast, Sxl appears to affect a discrete set of genes required in the male germline, such as Phf7. We also identify Tudor domain containing protein 5-like (Tdrd5l) as a target for Sxl regulation that is important for male germline identity. Tdrd5l is repressed by Sxl in female germ cells, but is highly expressed in male germ cells where it promotes proper male fertility and germline differentiation. Additionally, Tdrd5l localizes to cytoplasmic granules with some characteristics of RNA Processing (P-) Bodies, suggesting that it promotes male identity in the germline by regulating post-transcriptional gene expression.
| Like humans, all sexually reproducing organisms require gametes to reproduce. Gametes are made by specialized cells called germ cells, which must have the correct sexual identity information to properly make sperm or eggs. In fruit flies, germ cell sexual identity is controlled by the RNA-binding protein Sxl, which is expressed only in females. To better understand how Sxl promotes female identity, we conducted an RNA expression profiling experiment to identify genes whose expression changes in response to the loss of Sxl from germ cells. Here, we identify Tudor domain containing protein 5-like (Tdrd5l), which is expressed 17-fold higher in ovaries lacking Sxl compared to control ovaries. Additionally, Tdrd5l plays an important role in males as male flies that are mutant for this gene cannot make sperm properly and thus are less fertile. Moreover, we find that Tdrd5l promotes male identity in the germline, as several experiments show that it can shift the germ cell developmental program from female to male. This study tells us that Sxl promotes female identity in germ cells by repressing genes, like Tdrd5l, that promote male identity. Future studies into the function of Tdrd5l will provide mechanistic insight into how this gene promotes male identity.
| Sex determination is an essential process in sexually reproducing species, as the production of eggs and sperm depends on the sexual identity of the germ cells and somatic cells of the gonad. In some animals, such as the medaka fish and the house fly, the sexual identity of the soma determines the sexual identity of the germline. But in other animals, such as fruit flies and mammals, the intrinsic sex chromosome constitution (XX vs. XY) of the germ cells is also important for proper gametogenesis (reviewed in [1]). In such cases, the “sex” of the germ cells must match the “sex” of the soma in order for proper gametogenesis to occur. While studies have revealed a great deal about how sex is determined in the soma, how germline sexual identity is determined by a combination of somatic signals and germline autonomous properties is much less well understood.
In Drosophila, somatic sexual identity is determined by the X chromosome number [2], (reviewed in [3]), with two X’s activating expression of the key sex determination gene Sex lethal (Sxl), promoting female identity. The Sxl RNA-binding protein initiates an alternative RNA splicing cascade to allow female-specific splicing of transformer (tra) and, subsequently, doublesex (dsx) and fruitless (fru). dsx and fru encode transcription factors that control somatic sexual identity (reviewed in [4]). Sxl is also the key gene controlling autonomous sex determination in the germline, as Sxl is expressed in the germline in females, and loss of Sxl causes female (XX) germ cells to develop as germline ovarian tumors [5], similar to male (XY) germ cells transplanted into a female soma [6–8]. Further, expression of Sxl is sufficient to allow XY germ cells to make eggs when transplanted into a female soma [9]. However, how Sxl is activated in the female germline and how it regulates female germline identity remain unknown, except for the fact that both are different than in the soma [6,10–16]. To understand how Sxl promotes female germ cell identity, it is essential to discover its targets in the germline.
In this work, we report an RNA expression profiling (RNA-seq) experiment conducted to identify genes regulated downstream of Sxl in the germline. We found that a previously uncharacterized tudor domain containing protein, Tudor domain protein 5-like (Tdrd5l), is a target of Sxl in the germline. Tdrd5l is strongly expressed in the Drosophila early male germline and is repressed by Sxl activity in the early female germline. It promotes male identity in the germline, and its loss results in germline maintenance and differentiation defects in males, thus reducing their fertility. Tdrd5l protein localizes to cytoplasmic granules related to RNA Processing (P-) Bodies, suggestive of a function in post-transcriptional regulation of gene expression.
To investigate how Sxl acts to promote female identity in germ cells, we conducted an RNA-seq experiment comparing ovaries with and without Sxl function in the germline. To exclude the major gene expression changes that occur during the later stages of gametogenesis, the RNA-seq experiment was done in the bag of marbles (bam) mutant background. bam is essential for germline differentiation in both males and females; therefore, by using bam mutants we focus our experiment on gene expression changes in the early germline, where Sxl is expressed most robustly, instead of later stages of gametogenesis (Fig 1A and 1B). The use of bam mutants also gives us the ability to compare similar tissue samples, since ovarian tumors from bam mutants and bam, Sxl double mutants are very similar [17]. Thus, the two genotypes used for the RNA-seq experiment are both in the bam-mutant background, with the experimental genotype knocking down Sxl in the germline using RNAi (nanos-Gal4, UAS-SxlRNAi, bam), and a control genotype expressing a control RNAi (nanos-Gal4, UAS-mCherryRNAi, bam). Sxl protein in the germline was dramatically reduced in the Sxl RNAi samples relative to controls (Fig 1B and 1C). Additionally, we conducted RNA-seq on bam mutant males compared to females, similar to what has been done previously [18] in order to identify male-enriched vs. female-enriched RNAs in the undifferentiated germline. Libraries were prepared from three biological replicates of each genotype and sequenced with 100bp paired-end reads. The raw reads for all libraries were of very high quality—over 95% of the raw reads received a high quality score. In addition, for each library more than 85% of reads were uniquely mapped to the Drosophila genome, and all replicates had high replicate correlation.
As Sxl can regulate alternative splicing in the soma, we analyzed our RNA-seq data for changes in exon usage using DEXSeq [19]. We filtered these results for statistical significance (padj < 0.05) and for changes in exon expression that were 2-fold or greater (all DEXSeq data is presented in S1 Table). This identified 44 exons from 34 independent genes. A similar number of exons were upregulated in Sxl- (20, 45%) vs. downregulated (24, 55%). However, manual curation of these data did not reveal strong candidates for relevant biological regulation of alternative splicing by Sxl. The main exception was the Sxl RNA itself, which in Sxl RNAi samples exhibited the characteristic inclusion of the male-specific exon that requires Sxl protein in order to be excluded in the female soma. Thus, the residual Sxl RNA present after germline Sxl RNAi must be insufficient to produce enough Sxl protein for Sxl autoregulation, further reducing Sxl function in the germline and demonstrating the effectiveness of the germline Sxl RNAi approach. This analysis also identified the alternative, testis-specific promoter for the male germline sex determination factor Plant homeodomain containing protein 7 (Phf7, [20] as being regulated downstream of Sxl, as has previously been described [21]. However, based on this analysis, it does not appear that regulation of alternative splicing is a primary mechanism for Sxl regulation of gene expression in the germline.
We also analyzed changes in overall gene expression levels in the presence of absence of Sxl function (DE-Seq [22]). 94 genes were differentially expressed between the two samples (with padj < 0.05), with 40 being upregulated (43%) and 54 being downregulated (57%) in the Sxl RNAi samples (S2 Table). 24 of these genes differed 4-fold or more between the two samples (6 upregulated in Sxl RNAi, 18 downregulated). We also observed a bias toward X chromosome localization for the differentially expressed genes, with 31.9% being present on the X chromosome compared to only 15.1% of all genes surveyed having X chromosome localization.
We next asked whether Sxl acts as a global regulator of X chromosome gene expression/dosage compensation in the germline, as it does in the soma. Despite the X chromosome bias for differentially expressed genes, relatively few X chromosome genes overall (30 genes or 1.2% of X chromosome loci surveyed) were called as differentially expressed in our analysis. Further, if Sxl was a global regulator of X chromosome gene expression, we would expect that the overall level of X chromosome gene expression, relative to autosomal gene expression, would be altered in Sxl loss of function. However, the overall ratio of average gene expression between the X chromosome and autosomes was little changed between control and Sxl RNAi samples (X:A of 1.19 for controls and 1.24 for Sxl RNAi). We conclude that Sxl is not a global regulator of X chromosome gene expression in the germline.
Lastly, we evaluated the extent to which loss of Sxl in the undifferentiated (bam-mutant) female germline leads to masculinization of these cells. We conducted principle component analysis on the individual replicates of all four of our experimental conditions (Fig 1D). We find that bam Sxl-RNAi female samples are much more similar to bam female samples and control bam mCherry-RNAi samples than they are to bam male samples. Thus, this analysis does not support a strong masculinization of the bam-mutant female germline in the absence of Sxl. However, it is also known that bam does not affect males and females in exactly the same manner [23] and, in addition, the bam male samples also contain male somatic cells while the other samples contain female somatic cells. Both of these factors may contribute to the segregation of the bam male sample away from the others. To circumvent this problem, we analyzed the 94 genes differentially expressed between bam Sxl-RNAi and control samples to determine whether there was a “male” signature. We determined whether genes differentially expressed in Sxl-RNAi compared to controls were also differentially expressed in bam males compared to bam females. Of the 94 genes differentially expressed in Sxl RNAi, a high fraction (44 genes) were also differentially expressed between bam male vs. female samples (47%, which is considerably higher than the fraction of total genes in the genome called as different between bam male and female, 13%). However, these genes did not always change in the expected direction; only 61% of genes altered in both Sxl RNAi and bam males changed in the same direction in both, while 39% changed in the opposite direction (Fig 1E). Thus, the Sxl-RNAi sample does not appear globally “masculinized” relative to controls and it may be that Sxl’s role in repressing male identity in the early germline is restricted to a few specific targets that are important for the male germline.
One such candidate is a previously uncharacterized gene, CG15930, that was strongly upregulated in Sxl-RNAi ovaries and is normally enriched in testes. CG15930 exhibits strong homology to mouse Tdrd5 [24], which is essential for male germ cell development and spermatogenesis in mice [25]. However, since CG15930 is not as similar to Tdrd5 as Drosophila tejas [26], and therefore not a paralog of Tdrd5, we named this gene Tudor domain containing protein 5- like (Tdrd5l). We chose this gene for further study.
The RNA-seq expression profiles show that Tdrd5l has a dynamic expression pattern characteristic of genes with sex-specific functions. Tdrd5l is 18-fold enriched in bam testes compared to bam ovaries, and is upregulated 17-fold in bam, Sxl-RNAi ovaries relative to bam, control-RNAi ovaries (Fig 2A), and is clearly enriched in bam, Sxl-RNAi ovaries by RT-PCR (Fig 2B). This is in contrast to nanos, which is expressed at similar levels in all of the genotypes in our RNA-seq experiment (Fig 2A), consistent with its role in the germline of both sexes. Changes in Tdrd5l expression were restricted to total RNA levels, and no change in exon usage was detected. In situ hybridization to wild-type gonads revealed that Tdrd5l expression is highly enriched in the testis, particularly at the apical tip of the testis where the germline stem cells and proliferating gonial cells reside (Fig 2C and 2D). The finding that Tdrd5l is expressed at high levels in testes relative to ovaries, and is repressed by Sxl in the ovary, suggests that it plays a role in male germline development or function.
To determine the expression pattern of Tdrd5l protein, we generated a genomic transgene that includes a hemagglutinin (HA) epitope tag at the N-terminus of the Tdrd5l protein. The tag was inserted immediately following the start codon of the gene within a 20kb BAC that extends well upstream and downstream of the genes neighboring Tdrd5l (S1 Fig), and is therefore likely to recapitulate endogenous expression. Anti-HA immunostaining shows that Tdrd5l protein is expressed in the germline of the testis, and is also present in the ovary at lower levels (Fig 2E and 2F). HA::Tdrd5l is observed in male germline stem cells, spermatogonia and spermatocytes. The protein is seen in distinct foci that are smaller and more numerous in germline stem cells (Fig 2G, arrows), but appear larger in spermatocytes (Fig 2E arrows). The foci are predominantly cytoplasmic, with many abutting a perinuclear germline structure called the nuage. The accumulation of Tdrd5l into cytoplasmic punctae is characteristic of ribonucleoprotein complexes (“RNA bodies”) (reviewed in [27]), and suggests that it may be involved in mRNA decay or translational repression. Interestingly, HA::Tdrd5l co-localizes with decapping protein 1, (YFP-DCP1, Fig 2H, arrows), which plays a major role in mRNA degradation, and is also required for osk mRNA localization to the posterior of the oocyte [28], (reviewed in [29]). As DCP-1 is localized to “Processing bodies” (P-bodies), Tdrd5l appears to be present in a subset of these structures.
HA::Tdrd5l protein expression is upregulated in ovaries that are mutant for Sxl function in the germline (Fig 3A and 3B). Note that the punctae seen in males are also present in Sxl RNAi ovaries, though they are fewer in number (Fig 3D, arrows). Thus, the de-repression of Tdrd5l in Sxl RNAi ovaries is also detected at the protein level. Interestingly, the Tdrd5l mRNA has 2 putative Sxl binding sites [30,31], one within the 3rd intron and the other in the 3’ UTR (S1 Fig). This suggests that Sxl may directly regulate Tdrd5l expression by binding to one or both of these sites and influencing Tdrd5l RNA processing in the nucleus or translation in the cytoplasm. To assess Sxl’s direct regulation of Tdrd5l, we mutated both Sxl binding sites in the a HA::Tdrd5l transgene (HA::Tdrd5lddel). We found HA::Tdrd5lddel flies show upregulation of HA::Tdrd5l in the female germline (Fig 3C), similar to the upregulation caused by loss of Sxl from the germline. To quantify this regulation, we again took advantage of the bam-mutant background that enriches for Sxl-expressing cells and exhibits a similar tissue phenotype with and without Sxl function. We again found that the WT HA::Tdrd5l transgene exhibited higher protein expression when Sxl function was reduced by RNAi (Fig 3E). Further, the HA::Tdrd5lddel transgene exhibited an increased protein expression in the presence of Sxl function, and removing Sxl function caused no further increase in expression (Fig 3E & S2 Fig). These data indicate that Sxl directly regulates the expression of Tdrd5l protein in the female germline.
The male-biased expression pattern of Tdrd5l suggests it may have an important function in the male germline. Knocking down Tdrd5l function in the germline by RNAi, however, produced no observable phenotype. To conduct a more comprehensive study of Tdrd5l function we generated Tdrd5l mutant alleles using CRISPR-Cas9 genome editing. We generated several independent predicted null alleles of Tdrd5l (S1 Fig), and analyzed male fertility and testis morphology of both young males and aged males. We determined that young (5 days old) Tdrd5l mutant males have a 50% reduction in fecundity compared to controls (Fig 4F), suggesting that Tdrd5l is required for proper male fertility.
To characterize the germ cell defects that may lead to decreased fecundity, we evaluated several aspects of germ cell differentiation in young and aged mutant males. While testes of newly eclosed males appeared similar to wild-type, testes of older males (15–20 days old) exhibited a dystrophic “skinny testis” phenotype with a dramatic reduction in the germline (Fig 4A–4C). This was observed in 7% of animals raised at 25°C and 18% of animals raised at 29°C. Additionally, 15% of males exhibited a displaced hub phenotype (Fig 4E arrow), where the hub is no longer located at the apical tip of the testis as seen in wild-type (Fig 4D arrow). These combined phenotypes suggest a defect in proper germline differentiation and maintenance. Analyzing the expression of critical germ cell differentiation genes such as bam and spermatocyte arrest (required for meiotic cell cycle progression) [32] did not shed further light on the germ cell defect. Therefore, while the morphological defects present at a low penetrance, their overall effects culminate into a substantial reduction in fecundity; a phenotype which supports Tdrd5l importance in male germline development.
Tudor domain-containing proteins have well known functions in small RNA pathways, transcriptional regulation, and the assembly of snRNPs (reviewed in [33]). The closest Drosophila homolog to Tdrd5l is tejas, and the closest mammalian homolog to Tdrd5l is mouse TDRD5. Both tejas and TDRD5 have been shown to function in the piRNA pathway and to repress the expression of transposons in the germline [25,26,34]. Interestingly, we found no changes in transposon expression in Tdrd5l mutants in our RNA-seq analysis. We also analyzed the expression of a wide variety of transposons by quantitative rtPCR and found little to no difference between wt and Tdrd5l-mutant testes. Defects in the piRNA pathway can also lead to increased accumulation of the Stellate protein in the male germline [35,36]. We examined the expression of Stellate in Tdrd5l mutant males, as well as in Tdrd5l mutant males also heterozygous for mutant alleles of tejas, ago3 or aubergine (two PIWI proteins with key functions in the piRNA pathway). None of these testes showed the increased Stellate expression observed in tejas homozygous mutant testes (S3 Fig, [26]). Therefore, while Tdrd5l’s activity is important for the proper development of the male germline, it has a distinct function from regulation of transposon expression.
The sex-specific nature of Tdrd5l’s expression and mutant phenotype suggests it may play a role in promoting male germline sexual identity. However, unlike the male-specific Phf7 gene [20], expression of UAS-Tdrd5l in the female germline did not, by itself, result in defects in the female germline. To further investigate Tdrd5l’s role in sexual identity, we decided to conduct our experiments using the sensitized genetic backgrounds frequently used for the investigation of genes involved in sexual identity. Females mutant for transformer (tra)—a key player in the somatic sex determination pathway—undergo a transformation so that the somatic gonad of XX tra mutants develops as male instead of female. However, because the germline is XX, and therefore incompatible with spermatogenesis, the germline of these testes is highly undeveloped, causing these animals to be sterile (Fig 5B). A strong test of a gene’s ability to promote male identity in the germline is to determine whether it is sufficient to induce XX germ cells to enter spermatogenesis in these animals. Indeed, expression of Tdrd5l in the germline of XX tra mutants resulted in a robust rescue of spermatogenesis. 18% of these animals had highly developed testes that were wild type in size, containing all of the stages of germ cell differentiation up to spermatocytes (Fig 5C, note: since these animals lack a Y chromosome and the spermatogenesis genes located there, they were not expected to be fertile). This is strong evidence signifying that Tdrd5l promotes male identity in the germline.
Similarly, if Tdrd5l promotes male identity in the germline, we would expect that it would be able to enhance the ability of other mutations to masculinize the female germline. Homozygous mutants of ovarian tumor (otu) and sans fille (snf) have been shown to masculinize the female germline and cause germline tumors in ovaries, similar to Sxl-RNAi [37–40], while females heterozygous for mutant alleles of otu and snf are fully fertile and have normal ovary morphology. However, ectopic expression of Tdrd5l in the germline of females heterozygous for otu or snf resulted in the formation of ovarian tumors. 25% of snf/+; nos > Tdrd5l ovaries have large, pervasive germline tumors similar to the homozygous snf mutants (Fig 5D–5F). Another 25% of these ovaries show a complete loss of the germline (Fig 5G), which phenocopies strong sex determination mutants [37,41]. Additionally, 40% of otu/+; nos > Tdrd5l ovaries exhibit either ovarian tumors or complete loss of germline. This evidence supports Tdrd5l as a male-promoting factor in the germline. We therefore conclude that Tdrd5l functions in germline sexual identity; it promotes male identity in the germline and is repressed by Sxl in the female germline.
It has been known for many years that Sxl is necessary for female germline identity [7,8], and Sxl has also been shown to be sufficient to allow XY germ cells to undergo oogenesis [9]. It is likely that Sxl plays multiple roles in the germline, both to promote female identity in the early germline, perhaps as early as in the embryonic germline [9], and in regulating the differentiation of the germline during oogenesis [42]. Here we examined the role of Sxl more specifically in the undifferentiated germline through the use of bam mutants. Principle component analysis indicated that, under these conditions, samples with Sxl function reduced in the germline clustered close to control female samples, and far from male samples. While some of the male/female differences may be contributed by the somatic cells present in these samples, we conclude that reducing Sxl function in the undifferentiated germline does not lead to a dramatic masculinization at the whole-genome level. In contrast, we propose that the role for Sxl in the early germline may be restricted to a relatively small number of changes in sex-specific germline gene expression that are important for female vs. male germline function.
Recently, a genomic analysis of ovaries mutant for the RNA splicing factor sans fille (snf) was conducted [21]. This is considered to also be a Sxl germline loss-of-function condition as one important change in snf-mutant ovaries is a loss of Sxl expression and an ovarian tumor phenotype that can be rescued by Sxl expression [17,42,43]. In contrast to our results, an increased expression of spermatogenesis genes was observed in snf tumorous ovaries compared to wild type ovaries. It is likely that changes in these “differentiation” genes were not observed in our bam-mutant samples since germline differentiation is arrested at an earlier stage in bam mutants, allowing us to focus on the undifferentiated germline. Thus, these two analyses can help separate the role of Sxl in regulating early germline sexual identity vs. later aspects of sex-specific germline differentiation. Interestingly, one “differentiation” gene was identified in both RNA-seq analyses: the testis-specific basal transcription factor TATA Protein Associated Factor 12L (Taf12L or rye). This may indicate that Taf12L could play a role in the undifferentiated germline as well as the later stages of spermatocyte differentiation. In addition, both analyses found evidence for differential regulation of the important male germline identity factor Phf7 [20], where an upstream promoter is utilized preferentially in the male germline and is repressed downstream of Sxl in females [21]. This indicates a role for Phf7 in both the early and differentiating germline. Finally, we did not observe strong candidates for targets of alternative RNA splicing regulated by Sxl. The only strong candidate for alternative RNA splicing was the Sxl RNA itself, where the male-specific exon was retained in the residual Sxl RNA from the Sxl RNAi samples. This provides further evidence that Sxl autoregulation occurs in the germline as it does in the soma, as has previously been proposed [44]. It is likely that Sxl may also act at the level of translational control in the germline, as our evidence indicates here for regulation of Tdrd5l. Future experiments to identify Sxl-associated germline RNAs will be important for investigating this mechanism of action, as has recently been conducted [9].
In addition to its role in sex determination in the soma, Sxl also acts to initiate global X chromosome gene regulation and dosage compensation through translational control of male-specific lethal-2 [45], and it is possible that Sxl plays a similar role in the germline. Whether or not the germline even undergoes dosage compensation is controversial, and thoughtful work has led to opposite conclusions [46,47]. Further, if dosage compensation does exist in the germline, it must utilize a separate mechanism from the soma, as the somatic dosage compensation complex members msl1 and msl2 are not required in the germline [48,49]. However, Sxl could retain an msl-independent role to regulate global X chromosome gene expression in the germline. We did not observe evidence for this. First, few X chromosome genes were differentially expressed between Sxl- and control samples (30 genes or 1.2% of X chromosome genes tested). Second, the ratio of average gene expression between X chromsome genes and autosomes was very similar in Sxl- females compared to control females (X/A for controls: 1.24, Sxl-: 1.19) and this was similar when considering only genes in particular expression categories (e.g. all genes with some expression in both samples). Thus, it appears likely that Sxl’s role in the germline is distinct from that in the soma; it acts to control sex-specific gene regulation and sexual identity in both the germline and the soma, but acts as a general regulator of X chromosome gene expression and dosage compensation only in the soma.
We show here that Tdrd5l is both a target for Sxl regulation and is important for male germline identity and spermatogenesis. Tdrd5l expression is highly male-biased, both at the RNA and protein levels (Fig 2). When female germ cells are sensitized by partial loss of female sex determination genes, expression of Tdrd5l exacerbates the masculinized phenotype in these germ cells (Fig 5). Significantly, expression of Tdrd5l is sufficient to promote spermatogenesis in XX germ cells present in a male soma (XX tra-mutant testes, Fig 5). Thus, Tdrd5l clearly has a role in promoting male identity autonomously in the germline, which must coordinate with non-autonomous influences from the soma for proper germline sex determination.
Loss of Tdrd5l also has a strong effect on male fecundity, even though it is not absolutely required for spermatogenesis. The 50% reduction in fecundity is a strong effect and indicates that Tdrd5l plays an important role in the male germline. However, the fact that some spermatogenesis still proceeds suggests that other factors act in combination with Tdrd5l to control this process. One good candidate is Phf7 which we have previously demonstrated to have a similarly important, but not absolute, requirement for spermatogenesis [20]. However, our analyses of Tdrd5l, Phf7 double mutants did not reveal any synergistic effect on male germline development. This suggests that other important players in promoting male germline identity remain to be identified.
Our data indicate that Tdrd5l regulates male germline identity by influencing post-transcriptional gene regulation. Other tudor-domain containing proteins have been shown to act in RNA-protein bodies to influence RNA stability and translational regulation [reviewed in 33]. Further, Tdrd5l localizes to cytoplasmic punctae, specifically a subset of the punctae that also contain Decapping Protein1 (DCP-1), suggesting that these bodies are related to Processing bodes (P-bodies) that are known to control post-transcriptional gene regulation [28,50], (reviewed in [29]). Interestingly, Tdrd5l’s closest homologs, Tejas in flies and TDRD5 in mice, have been shown to regulate piRNA production and transposon regulation [25,26]. Further, their localization to VASA-containing nuage is thought to influence transposon control [26,51]. However, we observe no changes in transposon expression regulated by Tdrd5l, and Tdrd5l does not co-localize with VASA in nuage. We have also not observed any genetic interaction between tejas and Tdrd5l. Thus, we propose that Tdrd5l plays a distinct role in regulating male germline identity and spermatogenesis, and this may be in the regulation of mRNAs rather than transposons. One possibility is that the role of mouse TDRD5 in transposon regulation and spermatogenesis has been separated into the roles of Tejas in transposon regulation and Tdrd5l in regulating male identity and spermatogenesis in flies. It is widely known that regulation of germline identity is dependent on post-transcriptional mechanisms involving tudor-domain proteins such as the original Tudor protein [52], which helps define the germ plasm, an RNA body that regulates germline identity (reviewed in [53]). Further, the regulation of sex-specific gametogenesis is also dependent on RNA bodies and their requisite tudor-domain proteins, such as TDRD5 in mouse [54], (reviewed in [27]). Our studies indicate that initial germline sexual identity is similarly regulated by post-transcriptional mechanisms, including RNA bodies containing other, distinct, TUDOR-domain proteins such as Tdrd5l.
Gonads were dissected from 1–3 day old flies raised at 25°C. Ovaries were dissected from virgin females. 3 biological replicates were dissected for each genotype. Total RNA was isolated from all genotypes using RNA-bee (Tel-Test). Contaminating DNA was removed from the RNA using Turbo-DNA-free (Ambion). 200ng of RNA was used to prepare each library using the illumina TruSeq RNA Library Prep kit v2. 100bp paired-end read sequencing was done by the Johns Hopkins Genetic Resources Core Facility. The bam mutant male and female libraries were sequenced in one lane and the Sxl-RNAi, control-RNAi libraries were sequenced in separate lane, therefore having 6 libraries per lane.
Quality of raw reads was assessed using the fastQC kit (Babraham Institute). RNA-Seq reads were mapped to the Drosophila genome using Ensembl BDGP6 release 85, and Bowtie 2.2.9, TopHat 2.1.1, and HTSeq 0.9.1 [55–57]. Differential gene expression analysis was done using DESeq using ensemble annotation BDGP6 [22]. Adjusted P value of 0.05 used for significance cutoff. Differential exon analysis was done using DEXSeq [19].
The fly stocks used were obtained from Bloomington Stock Center unless otherwise indicated. bam1 [58], bamΔ86 (BDSC# 5427), nos-Gal4 [59], the control RNAi used was p{VALIUM20-mCherry}attP2 (BDSC# 35785), uas-Sxl-RNAi = TRiP.HMS00609 (BDSC# 34393), uas-CG15930-RNAi = TRiP. GL01046 (BDSC# 36882), Snf148, otu17, tej48-5, attP40{nos-Cas9} (NIG-FLY# CAS-0001), YFP:dDCP1 (a kind gift from Ming-Der Lin, [28]). Fecundity tests were carried out by setting up crosses with one Tdrd5l mutant male and 15 virgin females of the control stock. The control stock used is nos-Cas9 isogenized to FM7KrGFP fly stock to reproduce the treatment of the Tdrd5l mutant fly lines while screening for transformants. Each male was mated with virgin females for 4 days. Females were then discarded and each male was placed with another 15 virgin females in a new bottle. This was repeated twice more for a total of four mating bottles per male. All offspring were counted by day 18. Total offspring per male was calculated by averaging the number of offspring from each of the four mating bottles for each male.
Adult ovaries and testes were fixed, blocked and stained as previously described [60]. All images were taken with a Zeiss LSM 510 confocal microscope.
Primary antibodies and the concentrations used are as follows: chicken anti-Vasa 1:10,000 (K. Howard); rabbit anti-Vasa 1:10,000 (R. Lehmann); mouse anti-Sxl 1:8 (M18, DSHB); mouse anti-Armadillo 1:100 (N2 7A1, DSHB); rat anti-HA 1:100 (3F10, Roche); guinea pig α-TJ (1:1,000; generated by J. Jemc using the same epitope as previously described [61]); mouse anti-HTS 1:4 (1B1, DSHB). DSHB: Developmental Studies Hybridoma Bank. Secondary antibodies were used at 1:500 (Alexa-fluor). Samples were mounted in vectashield mounting solution with DAPI (vector Industries).
For RT-PCR and qRT-PCR, total RNA was isolated from ovaries and testes using RNA-bee (Tel-Test). Contaminating DNA was removed from the RNA using Turbo-DNA-free (Ambion). RNA was converted to cDNA using Superscript II (Invitrogen). qRT-PCR was done using 2 biological replicates and in technical triplicate, using SYBR green detection. In-situ hybridization was carried out as previously described [62]. DIG-labelled sense and antisense probes were synthesized by in vitro transcription of PCR product generated from RP98-1M22 BAC (BACPAC Resources Center).
Mutant alleles of Tdrd5l were created using CRISPR-Cas9 mediated genome editing. Small guide RNA (sgRNA) was designed and cloned following the Perrimon lab protocol [63], using the U6b-sgRNA-short vector described therein. The sgRNA was injected by Best Gene inc. into nos-Cas9(II-attP40) flies. The HA::Tdrd5l transgenic flies were generated by BAC recombineering [64,65], using the CH322-188C18 BAC obtained from the BACPAC Resources Center. A 3xHA epitope tag was added to the N-terminus of the gene (S1 Fig). This construct was also modified to delete the Sxl binding sites in the intron and the 3’UTR. The Sxl binding site in Tdrd5l 3’UTR was deleted using QuickChange II site-directed mutagenesis kit (agilent). The intronic binding site was removed by deleting the entire intron.
|
10.1371/journal.pgen.1007023 | High rate of adaptation of mammalian proteins that interact with Plasmodium and related parasites | Plasmodium parasites, along with their Piroplasm relatives, have caused malaria-like illnesses in terrestrial mammals for millions of years. Several Plasmodium-protective alleles have recently evolved in human populations, but little is known about host adaptation to blood parasites over deeper evolutionary timescales. In this work, we analyze mammalian adaptation in ~500 Plasmodium- or Piroplasm- interacting proteins (PPIPs) manually curated from the scientific literature. We show that (i) PPIPs are enriched for both immune functions and pleiotropy with other pathogens, and (ii) the rate of adaptation across mammals is significantly elevated in PPIPs, compared to carefully matched control proteins. PPIPs with high pathogen pleiotropy show the strongest signatures of adaptation, but this pattern is fully explained by their immune enrichment. Several pieces of evidence suggest that blood parasites specifically have imposed selection on PPIPs. First, even non-immune PPIPs that lack interactions with other pathogens have adapted at twice the rate of matched controls. Second, PPIP adaptation is linked to high expression in the liver, a critical organ in the parasite life cycle. Finally, our detailed investigation of alpha-spectrin, a major red blood cell membrane protein, shows that domains with particularly high rates of adaptation are those known to interact specifically with P. falciparum. Overall, we show that host proteins that interact with Plasmodium and Piroplasm parasites have experienced elevated rates of adaptation across mammals, and provide evidence that some of this adaptation has likely been driven by blood parasites.
| Malaria caused by the parasite Plasmodium falciparum remains the third-most deadly infectious disease of humans. Over the last 75,000 years, partial genetic resistance to malaria has evolved several times, earning malaria the title of "one of the strongest selective forces on the human genome." Yet, these human adaptations are only the most recent maneuvers in an ancient evolutionary war between host and parasite. Relatives of Plasmodium infect a variety of mammalian species today, and these large groups of hosts and parasites have likely coevolved for over 100 million years. Here, we identify 490 host genes that have been experimentally linked to the outcome of parasite infection in mammals, representing approximately 5% of all conserved mammalian proteins. In many cases, these proteins have also been linked to viral or bacterial infections. We show that parasite-interacting proteins have experienced ~3 times more adaptive substitutions than expected over mammalian evolution, and that blood parasites have left their own significant mark on our ancestors' genomes. We also identify one target of long-term adaptation to Plasmodium—the red blood cell protein alpha-spectrin—that may be involved in human susceptibility to malaria, demonstrating the value of considering modern disease in its long-term evolutionary context.
| Malaria is one of the world's most notorious infectious diseases, responsible for billions of illnesses and millions of deaths in the last fifty years alone [1]. Human malaria is caused by five species in the genus Plasmodium, which are evolutionarily related to Babesia, Theileria, and other parasites in the order Piroplasmida. Approximately fifty Plasmodium species cause malaria in primates, rodents, and bats [2], while Piroplasms infect a wider range of mammals (Fig 1). Although some wild animals appear to host malaria parasites without ill effects (e.g. [3]), others are known to suffer serious symptoms and death, especially when exposed to novel parasites [4, 5]. The severity of infection may then depend on the population history of exposure and individual acquired immunity, as it does in humans [6].
Parasites and other pathogens are important drivers of adaptive evolution in their hosts [7]. In the specific case of humans and Plasmodium, genetic variation in about 35 red blood cell or immune proteins has been associated with protection from severe complications of malaria, if not outright resistance (reviewed in [8–10]). Some of these protective genes, including HBB, DARC, and GYPA, have been supported by population genetic evidence of selection in African or Southeast Asian populations within the last 75,000 years (e.g. [11–13]). Malaria has consequently been labeled "one of the strongest selective forces on the human genome" [9, 10], though this statement has never been quantified.
Human adaptation to malaria is likely occurring within the broader context of mammalian adaptation to widespread blood parasites. The common ancestor of modern humans existed perhaps 200,000 years ago, while the common ancestor of placental mammals dates back ~105 million years [14, 15]. For comparison, the parasite genus Plasmodium experienced a major radiation ~129 million years ago [16]. Plasmodium and Piroplasms have likely infected mammals for as long as mammals have existed, but the evolutionary consequences of this long-standing relationship have never been investigated.
Despite their age and diversity, Plasmodium and Piroplasms cause disease through similar mechanisms, including transmission to and from mammalian blood by the bite of a mosquito or tick. Plasmodium cells migrate first to the liver, multiply within hepatocytes, and emerge several days later to invade red blood cells (RBCs) [17]. Babesia parasites invade RBCs directly, while Theileria parasites infect both red and white blood cells [18, 19]. Although Piroplasms like Babesia and Theileria are thought to lack a liver stage [20], their infections cause substantial liver damage through increased coagulation and other mechanisms [21–23]. Parasitized cells also adhere to capillaries lining the liver, lung, brain, and other tissues, which can impair circulation and lead to life-threatening organ dysfunction (e.g. [18, 24–26]). Finally, each parasitic infection triggers a complex immune response from the host, including the removal of infected RBCs from circulation by the spleen (e.g. [27]).
The complexity of these host-parasite interactions makes it difficult to precisely measure their evolutionary impact. One important reason is that our knowledge of host responses is biased toward convenient samples, like blood cells, from specific groups, like humans and Plasmodium. We particularly lack information across the extant diversity of parasite and host species [2]. A second key reason is that host genes relevant to malaria are likely to be pleiotropically involved with other selected phenotypes, including responses to other pathogens [28]. In particular, some Plasmodium-associated genes in humans are also associated with viruses or bacteria, making it difficult to attribute their evolution specifically to pressure from Plasmodium [29, 30]. Parsing the contribution of various pathogens to host evolution thus requires a broader understanding of many host genes, many tissues, and many pathogens.
In this work, we examine patterns of adaptation and functional pleiotropy in a set of ~500 Plasmodium- or Piroplasm-interacting proteins (PPIPs) manually curated from the literature. These PPIPs represent about 5% of the mammalian proteome, as defined by the set of 9,338 proteins conserved across 24 well-sequenced mammal species. Previously, evolutionary analysis of an externally defined gene set has proven useful for detecting polygenic adaptation [7, 31–33]. Here, because PPIPs represent a relatively small fraction of all conserved mammalian genes, we use permutation tests to compare PPIPs to a background of non-PPIP controls. That is, we compare PPIPs to many sets of other mammalian proteins, which we match to PPIPs by a number of important metrics. This approach has recently been used by [31] to identify viruses as a dominant driver of adaptation in mammals.
Overall, we demonstrate that PPIPs have experienced ~3 times more adaptive substitutions than expected throughout mammalian evolution. The strongest adaptive signals are present in PPIPs with immune functions, which are highly pleiotropic with respect to other pathogens. However, we detect a significant excess of adaptation even in non-immune PPIPs that are not known to interact with pathogens beside Plasmodium. Additional evidence suggests that the red blood cell protein alpha-spectrin, as well as PPIPs highly expressed in the liver, may have played key roles in adaptation to blood parasites. Overall, our work supports the hypothesis that Plasmodium and Piroplasm parasites—not unlike other classes of pathogens—have been important and long-standing drivers of evolutionary change in mammals.
Malaria-like illnesses generate substantial health and economic burdens in humans, livestock, and pets [1, 34]. These costs have motivated a large body of research into host-parasite interactions and host responses to infection. We queried the PubMed database for scientific papers whose abstracts mentioned the name of a host gene along with the terms malaria, Plasmodium, Babesia, Theileria, Rangelia, or Cytauxzoon, the latter four being the best-studied Piroplasmid genera (Methods, PPIP Identification). To focus on mammalian evolution, we limited our search to 9,338 protein-coding genes that are conserved in 24 mammalian species with high-quality reference genomes (Fig 1; Methods, Mammalian Orthologs; [31]). Most of these mammalian species belong to one of four orders—primates, rodents, artiodactyls, or carnivores—and represent a range of susceptibilities to our focal parasites (Fig 1).
This search returned ~35,000 papers associated with ~5,000 mammalian genes. However, the vast majority of these results were false positives. Many short acronyms that identify genes have multiple meanings, and many papers containing these acronyms do not concern genes or proteins. We manually curated paper titles and abstracts to identify just 786 papers linking 490 proteins to Plasmodium or Piroplasms via four types of phenotypic evidence: (1) biochemical interaction between a mammalian and parasite protein; (2) statistical association between genetic variation and disease susceptibility; (3) knockout or overexpression studies impacting susceptibility; and (4) low-throughput studies showing a change in gene expression during infection (Fig 2A; S2 Table).
Nearly half of the 490 PPIPs (45%) were supported by multiple studies, and 35% by multiple sources of evidence. Expression changes were the most common form of evidence (85% of PPIPs, or 60% of total evidence), with about 43% of PPIPs identified only via expression changes. Expression-based PPIPs likely represent both direct and indirect interactions with parasites, given the size and interconnectedness of gene expression networks (e.g. [35]). We have attempted to limit indirect interactions by excluding high-throughput expression experiments, and we later show that all four evidence types, including expression changes, identify sets of genes with elevated rates of adaptation (S4 Fig). Consequently, we chose to analyze all PPIPs together without making distinctions based on evidence type.
The majority of PPIPs were linked to Plasmodium in studies of humans and mice (Fig 2). A fifth of PPIPs (21%) were linked to Piroplasms in studies of cows, dogs, and other mammal species (Fig 2). Plasmodium- and Piroplasm-interacting proteins overlap substantially, ~14 times more than expected by chance (p<1x10-5, Fig 2B). This overlap is consistent with the similar life cycles of Plasmodium and Piroplasms, as well as the conservation of host responses to these parasites across mammals.
To ask whether PPIPs generally perform functions relevant to malaria, we also tested 17,696 GO functional categories for PPIP enrichment (Methods, Protein Metrics). After correcting for multiple testing, over 1,200 categories contained significantly more PPIPs than expected (S3 Table). The most enriched categories were dominated by immune functions, with 51% of PPIPs falling under immune system process (p = 6.16 x 10−94) and 83% under response to stimulus (p = 7.89 x 10−69) (Table 1). Other highly enriched categories indicate functions more specific to malaria pathology, including cell adhesion (p = 9.89 x 10−32), hemostasis (p = 1.68 x 10−35), and hemopoiesis (p = 2.20 x 10−37) (Table 1). Together, these results reflect the expected functions of host genes ascertained through studies of malaria-relevant processes.
Many immune genes, even outside the adaptive immune system, are activated by signals from multiple pathogens (e.g. [36, 37]). This 'pathogen pleiotropy' poses an important complication when testing the link between blood parasites and host adaptation, even in genes phenotypically linked to these parasites. To quantify the extent of pathogen pleiotropy in mammals, we compared PPIPs to host proteins known to interact with viruses and bacteria (Methods, VIPs and BIPs). In both cases, we focused only on genes conserved across our focal 24 mammal species. For viruses, we obtained a high-quality list of 1,256 manually curated virus-interacting proteins from [31] (S4 Table). For bacteria, we queried the EBI IntAct database [38] for all deposited interactions between humans and bacteria, which returned 1,250 host proteins (S4 Table).
Overall, we find that 36% of all PPIPs also interact with viruses, 23% with bacteria, and 48% with viruses and/or bacteria—many more than expected by chance (Fig 2C; all p<10-4). Unsurprisingly, this overlap is strongest for immune PPIPs (here defined as falling under the GO category immune system process), of which 56% interact with multiple pathogens (p<10-4). However, nearly 40% of non-immune PIPs also interact with multiple pathogens (p<10-4). Some of these "non-immune" proteins may have uncharacterized immune functions, but most are known for their involvement in general cellular processes, including metabolism and signal transduction. This suggests that a diverse array of prokaryotic, eukaryotic, and viral pathogens may interact with a surprisingly small number of host proteins, or alternatively, that these proteins represent a non-specific host response to infection.
Plasmodium and Piroplasms influence several mammalian tissues as they progress through their complex life cycle. To begin investigating the specificity of PPIPs to malaria-like infections, we examined gene expression in a condensed set of 34 human tissues collected by the GTEx Consortium from uninfected individuals [39] (Methods, Protein Metrics). We first found that PPIPs have an average of 8.2% higher total expression than randomly selected sets of non-PPIPs (p<0.001; S1A Fig). To fairly evaluate PPIP expression enrichment in each tissue, we designed a matched permutation test that compares PPIPs to many, similarly-sized sets of control genes with similar total expression (Methods, Permutation Tests). Throughout this work, we use matched permutation tests to compare PPIPs to many sets of other genes that are controlled for confounding factors. This approach allows us to isolate the evolutionary effects of interactions with Plasmodium or Piroplasms from potentially correlated factors, such as high total expression.
Compared to matched control proteins, we found that PPIPs were significantly differentially expressed in 20 of 34 tissues, despite the noisiness of this measurement (Fig 3). PPIP expression was underrepresented in 16 tissues, particularly those involved in reproduction, and overrepresented in four tissues: blood, liver, lung, and spleen. All four of these overrepresented tissues are key sites of parasite replication, containment, and/or tissue damage in Plasmodium or Piroplasm infections in mammals. This result may be due in part to an ascertainment bias in sampling certain tissues, especially blood, but is also consistent with the biology of host interactions with Plasmodium and Piroplasm parasites.
We have already shown that PPIPs have three unusual properties—immune enrichment (Table 1), excess interactions with other pathogens (Fig 2C), and high mRNA expression (S1A Fig)—that may influence their rate of evolution. In order to evaluate PPIP adaptation against an appropriate background, we assessed several additional metrics for PPIPs and other proteins in order to control for them in our permutation tests (S5 Table).
First, we examined three additional broad measures of gene function in humans: the density of DNAseI hypersensitive elements; protein expression, as measured by mass spectrometry; and the number of protein-protein interactions (Methods, Protein Metrics). For each of these metrics, PPIPs have significantly higher mean values than sets of random controls, indicating that PPIPs are more broadly functional in humans (S1B–S1D Fig; all p<0.001). We next tested four measures of genomic context that have been linked to the rate of sequence evolution: GC content; aligned protein length; the regional density of protein-coding bases; and the density of highly conserved, vertebrate elements [40–43](Methods). Most of these metrics do not differ between PPIPs and other genes (S1F–S1H Fig), with the exception of conserved element density, which is slightly but significantly lower in PPIPs (mean = 8.0% vs. 8.8%; p = 0.002; S1E Fig).
Based on these results, we expanded our permutation test to match all five of these significantly varying measures of gene function and genomic context while generating sets of non-PPIP control genes. Each non-PPIP was considered an acceptable match for a given PPIP if its values for all five metrics fell within specific ranges of the PPIP values (Methods, Permutation Tests). About 10% of PPIPs were too dissimilar from other proteins to be matched and were excluded from subsequent analysis, but these proteins show similar rates of adaptation to other PPIPs (S6 Table). On average, each retained PPIP could be matched to 34 control genes, allowing the generation of many different sets of ~440 matched controls. This permutation procedure effectively equalized distributions between PPIPs and control genes for all tested functional and evolutionary metrics (compare S1 Fig to S2 Fig).
Finally, one of the largest differences between PPIPs and other proteins is the frequency with which they are discussed in the scientific literature (S1I Fig). The average PPIP has 6.9 times more PubMed citations than the average mammalian protein (Methods, Protein Metrics). This difference was too large to match in the permutation test without excluding the majority of PPIPs. However, we show that the citation frequency of non-PPIPs has no relationship with protein adaptation (p ≥ 0.17; S3 Fig). This indicates that a high rate of citation for PPIPs is not statistically associated with their rate of adaptation.
After controlling for each metric of function and genomic context (S2 Fig), we asked whether PPIPs exhibit unusual patterns of amino acid substitution and polymorphism in mammals. Importantly, PPIPs have a typical ratio of non-synonymous to synonymous polymorphism in a combined sampling of great ape species (Fig 4A; mean pN/(pS+1) = 0.21 in PPIPs vs. an average of 0.20 in matched controls; p = 0.10; Methods, Protein Metrics). That is, PPIPs do not appear more or less evolutionarily constrained than other similar proteins, bolstering the null expectation that they should evolve at average rates.
In contrast, we find that PPIPs have a significantly elevated ratio of non-synonymous to synonymous substitutions across 24 mammal species (dN/dS = 0.186 in PPIPs vs. an average of 0.128 in matched controls, p<10−4). If we make a very conservative (and in many ways unreasonable) assumption that matched controls have experienced no adaptation in the history of mammals, then this 31% excess in dN/dS in PPIPs, despite unremarkable pN/(pS+1), implies that at least 31% of all amino acid substitutions in PPIPs were adaptive. However, given that matched controls have likely experienced at least some adaptation, the proportion of adaptive substitutions in PPIPs is likely to be even larger.
We investigated adaptation in PPIPs in more detail using the BS-REL and BUSTED tests available in the HYPHY software package [44–46] (Methods, Estimating Adaptation). Both tests use maximum likelihood models to estimate the proportion of codons in a protein with dN/dS > 1, consistent with adaptation in some proportion of the protein. BS-REL estimates the exact proportion in each branch of the tree, whereas BUSTED estimates whether it is greater than zero in at least one branch (i.e., whether a gene has experienced positive selection at some point in the history of mammalian evolution).
Both models find additional evidence of excess adaptation in PPIPs. Nearly 36% of PPIPs have BUSTED evidence (at p≤0.05) of adaptation in some part of the mammalian phylogeny, versus 24% of matched controls (p<10−4; Fig 4C). PPIPs also have BS-REL evidence for adaptation on more branches of the mammalian tree (p = 1.87× 10−4; Fig 4D), as well as for more codons per protein (p<10−4; Fig 4E). This excess is robust to the BUSTED p-value threshold used to define adaptation, and increases as the threshold becomes more stringent (Fig 4F, p = 7x10-5). PPIPs identified via different kinds of evidence all have more adaptation than expected by chance, although PPIPs that physically interact with parasite proteins have a greater excess of adaptation than PPIPs that change expression during infection (p = 0.032, S4 Fig).
These matched permutation tests show that PPIPs have experienced an elevated rate of adaptive substitutions in mammals. Although the hundreds of published experiments that define PPIPs (Fig 2A) support the idea that Plasmodium and Piroplasms may have driven this adaptation, it remains critical to address PPIP pleiotropy with other pathogens (Fig 2C).
Based on the available information for many host-pathogen interactions (Methods, VIPs and BIPs), we divided PPIPs into three categories: "Plasmodium-only," "Plasmodium + Piroplasms," and "multi-pathogen," which includes PPIPs that also interact with viruses and/or bacteria (Fig 2C). For each category, we again matched PPIPs to controls and found significantly more adaptive substitutions than expected (Fig 5A). Among categories, the excess adaptation for PPIPs versus controls is greater when more diverse pathogen interactions are included (p<0.05). That is, Plasmodium-only PPIPs have 1.9X more adaptation than expected; PPIPs that interact with Plasmodium and/or Piroplasms but are not known to interact with viruses or bacteria have 2.5X more adaptation than expected; and PPIPs that also interact with viruses or bacteria have 3.7X more adaptation than expected (Fig 5A). This result suggests that an increased number and diversity of pathogen interactions drives a cumulative increase in host adaptation.
Importantly, PPIPs that interact with more pathogens are also more likely to have immune functions (Fig 5A, Fig 2C). Only 39% of Plasmodium-only PPIPs, versus 59% of multi-pathogen PPIPs, have a GO annotation for immune system process (Fig 5A). Immune genes are well known to evolve at rapid rates [47–52]. Here, we also find that non-immune PPIPs have adapted at slower rates than PPIPs as a whole (Fig 5B; p = 0.018; see Methods, Permutation Tests for why immune PPIPs are not analyzed directly). These correlations among immune function, adaptation, and multi-pathogen interactions complicate the link between malaria-like parasites and host adaptation.
Fortunately, these correlations can be disentangled by considering the 239 PPIPs that do not have an annotated immune function. In these non-immune PPIPs, there is a breakdown of the link between adaptation and multi-pathogen interactions (Fig 5C). That is, non-immune PPIPs known to interact only with Plasmodium have ~2X more adaptation than expected, and this excess does not significantly increase when PPIPs that interact with additional pathogens are included (all p > 0.17; Fig 5C). We note that this lack of a significant difference is not simply due to reduced power from a reduced sample size, given that subsampling of PPIPs in Fig 5A to the sample sizes in Fig 5C retains complete power to detect differences. Overall, we show that even non-immune PPIPs not known to interact with any pathogens except for Plasmodium still show sharply elevated rates of adaptation. Although we lack complete knowledge of host-parasite interactions, to explain this result independently of Plasmodium as a selective pressure would require the existence of some other pressure or pathogen, whose interactions with mammalian genes overlap remarkably well with those of Plasmodium.
Host adaptation to malaria could potentially be concentrated in any malaria-relevant tissue enriched for PPIP expression, specifically blood, liver, lung, and spleen (Fig 3). We used a threshold analysis to test whether expression in these tissues was linked to elevated adaptation. That is, we compared expression patterns for the bulk of genes to patterns in the 5% of genes with the most adaptive codons for both PPIPs and controls matched for total expression (Fig 6).
In sets of matched control non-PPIPs, the most highly adaptive genes are expressed at significantly lower levels than the bulk of control non-PPIPs in all four malaria-relevant tissues (Fig 6). In contrast, for PPIPs, highly adaptive genes are not expressed at significantly lower levels in any of the tissues, despite the same overall sample size and level of total expression. In fact, in the case of the liver, the highly adaptive PPIPs are expressed at significantly higher levels than other PPIPs, the opposite direction of the pattern observed in controls (p = 9.8 x 10−4; Fig 6). The fact that high expression in the liver is associated with elevated adaptation in PPIPs, but not controls, suggests that the liver may have been a site of particularly strong selective pressures acting specifically on PPIPs.
Plasmodium and Piroplasm infections have been reported from a wide variety of mammalian species (Fig 1). We tested whether PPIP adaptation is similarly widespread across mammals by applying BUSTED and BS-REL models to subsets of the sequence data within individual mammalian orders (Methods, Order-specific Analyses).
When all PPIPs are considered, we find a highly significant excess of PPIP adaptation in rodents and primates (both p<0.001; Fig 7). Before correcting for multiple testing, the signal is marginally significant in carnivores (p = 0.052) and positive, but not significant, in artiodactyls (p = 0.29). However, it is difficult to compare significance among clades for two reasons. First, we cannot account for differences in evolutionary rate in different groups due to, e.g., generation time. Second, our mammalian tree has not sampled equal numbers of species from each the four clades (Fig 1). We further note that we completely lack statistical power to perform clade-specific analysis on subsets of PPIPs, such as Plasmodium- or Piroplasm-only PPIPs (Methods, Permutation Tests). Despite these caveats, clade-specific analyses indicate at least a trend toward high adaptation in PPIPs in all major clades of the mammalian tree.
Direct, biochemical interactions between mammalian and parasite proteins may be particularly important drivers of host adaptation (S4 Fig), although such interactions remain uncharacterized for the majority of PPIPs (Fig 2A). We chose one well-studied PPIP to test for a direct relationship, at the amino-acid level, between host adaptation and biochemical host-parasite interactions.
Of the top ten PPIPs with the strongest BUSTED evidence of adaptation, alpha-spectrin (SPTA1) is the only candidate that has been extensively characterized for molecular interactions with Plasmodium proteins. Alpha-spectrin is a textbook example of a major structural component of the red blood cell (RBC) membrane. In humans, dozens of polymorphisms in this gene are known to cause deformations of the RBC, which may either be asymptomatic or cause deleterious anemia (reviewed in [53]). These deformations are more common in individuals of African descent, leading to the hypothesis that SPTA1 is involved in malaria resistance in humans.
The SPTA1 protein has a well-defined domain structure, and specific interactions with Plasmodium proteins are known for three domains (Fig 8). Repeat 4 is the binding site for KAHRP, the major P. falciparum component of the adhesive 'knobs' that form on the surface of infected RBCs [54]. Another 65-residue fragment containing EF-hand 2 has been shown to bind to PfEMP3, an interaction that destabilizes the RBC skeleton and may allow mature merozoites to egress from the cell [55]. A central SH3 domain can also be cleaved by a promiscuous Plasmodium protease called plasmepsin II [56], which mainly functions in hemoglobin digestion [57]. Furthermore, naturally occurring mutations in the first three SPTA1 domains have been shown to impair the growth of P. falciparum in human RBCs [58–60].
We wished to test whether sites of mammalian adaptation in SPTA1 mapped to any of these Plasmodium-relevant domains. To identify adaptive codons with higher precision and power, we aligned SPTA1 coding sequences from 61 additional mammal species (S7 Table, S10 Table) for analysis in MEME [61] (Methods, Alpha-spectrin). Of the 2,419 codons in this large mammalian alignment, we found that 63 show strong evidence of adaptation (p<0.01), and that these are distributed non-randomly throughout the protein.
Remarkably, three domains—Repeat 1, Repeat 4, and EF-hand 2—are significantly enriched for adaptive codons, after controlling for domain length and conservation (Fig 8; Methods). That is, all three SPTA1 domains with strong evidence of adaptation in mammals are known to either interact specifically with P. falciparum proteins, or harbor human mutations that provide resistance to P. falciparum. This overlap is unlikely to occur by chance (p = 0.015) and is robust to the p-value thresholds chosen (S8 Table). Thus, evidence from SPTA1 suggests a specific connection, at least in this well-studied example, between the mechanics of Plasmodium infection and adaptation in the host red blood cell.
Notably, we do not claim that all adaptation in SPTA1 is due to pressure from malaria. Adaptation has occurred in at least one codon of SPTA1 on every branch of the 85-species mammalian tree, with the top branches including the base of the Camelidae, Loxodonta africana (elephant), Trichechus manatus (manatee), and the base of all Eutheria (S11 Table). The especially high density of adaptive changes in camels may be related to the unusual shape of their red blood cells, which has been shown to extend RBC lifespan during chronic dehydration [62, 63]. Nonetheless, when we focus only on the three domains of SPTA1 that are enriched for adaptive substitutions (Fig 8), we find much stronger evidence of adaptation on primate branches than when the entire SPTA1 protein is considered (S11 Table). This observation is consistent with known molecular interactions between P. falciparum and these specific SPTA1 domains in humans. Together, branch- and domain-specific patterns of adaptation in SPTA1 support malaria as an important, but by no means unique, influence on the evolution of mammalian red blood cells.
In this work, we have identified 490 conserved mammalian proteins that interact with Plasmodium or Piroplasm parasites. This large set of PPIPs is a substantial expansion of the list of host proteins traditionally considered to be associated with Plasmodium and Piroplasms (e.g. [10]), enabling us to investigate both their broad functional properties and long-term evolutionary patterns.
We find that PPIPs are strongly enriched for immune annotations, although about half are involved in diverse non-immune processes (Table 1). PPIPs are also widely expressed, but particularly overrepresented in the blood, liver, lung, and spleen—tissues highly relevant to the Plasmodium and Piroplasm life cycles. We find that PPIPs tend to interact not only with multiple blood parasites (Fig 1B), but also with unrelated bacterial and viral pathogens (Fig 1C). As expected, this multi-pathogen overlap is strongest for immune PPIPs. Somewhat surprisingly, this overlap also extends to non-immune PPIPs, suggesting either that unrelated parasites tend to interact with the same host proteins or that these proteins correspond to some non-specific host response.
Our key result is that PPIPs have been evolving unusually quickly compared to carefully matched non-PPIPs (Fig 4). If we conservatively assume that none of the amino acid substitutions in non-PPIPs have been adaptive, then we may estimate that 31% of amino acid substitutions in mammalian PPIPs have been driven by positive selection. However, because non-PPIPs have also experienced appreciable positive selection (Fig 4C), the true proportion of adaptive substitutions in PPIPs is certainly higher. Regardless of the precise number, it is clear that host proteins that interact with Plasmodium or Piroplasm parasites have evolved at an elevated rate, with a substantial proportion of amino acid changes driven by positive selection (Fig 4).
Across mammals, the rate of PPIP evolution is comparable to that of other proteins previously identified as targets of strong positive selection. For example, the antiviral protein PARP14 and the sperm-expressed protein TEX15—which are not PPIPs—represent two classes of proteins that have diversified rapidly in some mammals [64, 65]. When we consider the proportion of codons under positive selection in all our mammalian orthologs, we find that PARP14 ranks in first place (7.2%), TEX15 in third place (5.8%), and the Piroplasm-associated ENTPD1 in second place (6.3%) (S5 Table). As another example, in colobine monkeys, a shift to foregut fermentation is thought to have driven adaptive substitutions in approximately half of the residues of the antibacterial enzyme lysozyme [66]. Across our broader sampling of mammals, lysozyme ranks in 228th place (1.8% adaptive codons), behind 41 PPIPs (S5 Table). It can be difficult to make precise comparisons across studies that include different mammal species, and with the exception of a recent study of viruses [31], few have systematically assessed the importance of particular selective pressures across many mammals. Nonetheless, PPIPs appear to have experienced similar rates of adaptation as some of the better-known mammalian examples.
Given the extreme pleiotropy of PPIPs in regards to other pathogens, a natural question is whether Plasmodium and Piroplasms are truly drivers of PPIP adaptation. Our set of 490 PPIPs is large enough to begin parsing the specific effects of blood parasites on mammalian evolution, independent of the effects of other pathogens. When we consider PPIPs that interact only with Plasmodium or Piroplasms, but not viruses or bacteria, we find a 2.5X enrichment of adaptation compared to matched controls (Fig 5A). For non-immune PPIPs in particular, additional viral or bacterial interactions do not elevate the excess of adaptation, which remains significantly higher than in matched controls (Fig 5C). This provides some evidence that blood parasites have played a specific role in influencing mammalian protein evolution, both in immune and non-immune genes.
Two additional pieces of evidence are consistent with this idea. First, PPIPs with the highest levels of adaptation are also particularly highly expressed in the liver, opposite to the pattern seen in matched controls (Fig 6). This suggests the possibility that adaptation is related not simply to liver expression, but to parasite interactions that take place in the liver. In Plasmodium infections, parasites initially migrate to the liver, invade hepatocytes, and replicate many times before emerging to infect red blood cells [17]. In Piroplasm infections, liver damage is also common and associated with fatality (e.g. [21, 22, 67]). Although some Piroplasms are thought to lack a liver stage [20, 68], a number of studies have reported the presence of Babesia, Rangelia, or Cytauxzoon parasites within the endothelial cells of the liver, among other tissues [23, 69–71]. The liver may thus represent a critical opportunity for PPIPs to ameliorate the effects of Plasmodium and Piroplasm infection on the host.
Second, in the well-studied case of alpha-spectrin (SPTA1), we were able to directly investigate the correspondence between sites of host protein adaptation and sites of molecular interactions with Plasmodium. We indeed found strong evidence of adaptation in three domains of SPTA1 that are known to participate in molecular interactions with Plasmodium parasites (Fig 8), consistent with host-parasite interactions specifically driving mammalian adaptation. Notably, our evidence of adaptation in SPTA1 was derived from long-term evolutionary patterns in dozens of mammal species, whereas molecular interactions with P. falciparum were identified only in humans. This is analogous to the interspecies evolutionary patterns we observe in the PPIPs identified from intraspecies association studies (S4 Fig). These results suggest that the host cellular machinery underlying extant parasite interactions has been largely conserved over deep evolutionary time, potentially allowing the same proteins to participate in adaptation across multiple time scales.
In the end, despite these three lines of evidence pointing towards Plasmodium and Piroplasms specifically driving PPIP adaptation, this conclusion must remain tentative because of our incomplete knowledge of host-pathogen interactions. At one level, many host genes that interact with pathogens likely remain unidentified [31]. At another level, the taxonomic distribution of pathogens on hosts remains quite poorly understood [2]. This is especially problematic for testing whether adaptation in certain mammal lineages corresponds to the densities of specific parasites. For example, based on veterinary records, we may have expected artiodactyls to adapt specifically to Piroplasms but not to Plasmodium (Fig 1). However, white-tailed deer in North America were recently discovered to carry a Plasmodium species at high frequency [72]. This finding demonstrates that absence of proof is not proof of absence when it comes to the phylogenetic distribution of pathogens, nor to interactions between parasites and host genes. Emerging high-throughput studies of host-pathogen interactions, combined with broader sampling of natural infections, will allow more precise tests of how hosts evolve in response to specific pathogens.
In our case, the fact that non-immune, "Plasmodium-only" PPIPs show a clear excess of adaptation (Fig 5) may reflect either specific interactions with Plasmodium or incomplete knowledge of interactions with other pathogens. Likewise, high rates of adaptation in PPIPs highly expressed in the liver may reflect adaptation to liver-antagonizing blood parasites, or to other viral and bacterial pathogens that also damage the liver. Our analysis of SPTA1 provides the most compelling evidence of a specific association between adaptation in a PPIP and interactions with Plasmodium, but because this is only a single example, we cannot claim that such associations would be found more broadly if other PPIPs were to be studied in similar detail. However, these results are hopeful for our future ability to identify specific selective pressures associated with specific pathogens.
Other future work could also examine PPIPs for evidence of balancing selection, especially as more non-human polymorphism data become available. Several examples of PPIP evolution in humans indicate an important role for the maintenance of polymorphism [11, 73], and it is possible that sampling of PPIP polymorphism in other species has contributed to the elevated divergence shown here. Balancing selection within species and directional selection across species may even be two sides of the same coin, as evidenced by immune and other genes that appear to have experienced both [11, 74] (S5 Table; S4 Fig). Indeed, balanced states have been shown to be a natural consequence of directional selection in fast-changing environments [75]. In human populations with abundant polymorphism data, PPIPs could be used as an important resource for understanding the relationship between these two selective modes.
In conclusion, we show that proteins that interact with Plasmodium and Piroplasms comprise a substantial portion of the mammalian proteome; that they exhibit high rates of adaptation across mammals; and that this adaptation may be partially driven by these blood parasites. We hope that the collection of 490 mammalian PPIPs will continue to prove a powerful and continually growing resource for exploring host-parasite interactions and adaptation.
We queried PubMed for scientific papers containing both a gene name and the term malaria, Plasmodium, Babesia, Theileria, Rangelia, or Cytauxzoon in the title or abstract, as of Feb. 20, 2017. Human gene names were drawn from the HUGO Gene Nomenclature Committee [76] (http://www.genenames.org/) for 9,338 mammalian orthologs (see Methods, Mammalian Orthologs). For each of the genes that returned at least one hit, we manually evaluated the titles of up to 20 associated papers to assess the link between the gene and a malaria phenotype. Many acronyms used to represent genes are also used as abbreviations for techniques, locations, drugs, or other phrases. Consequently, most genes could be eliminated based on their nominal connection with papers addressing non-genetic aspects of malaria.
For papers discussing genes, we examined the abstracts for the presence and type of evidence connecting genes to malaria phenotypes. In cases where the abstract was ambiguous, we examined the full text of the paper.
To limit the number of false positives, we did not classify PPIPs using evidence from RNAseq or other high-throughput experiments. Gene expression is typically regulated via large, interconnected networks (e.g. [35]), such that high-throughput experiments can identify hundreds or thousands of genes whose expression is perturbed by infection. Many of these differentially expressed genes may have very small or indirect impacts on the progression of malaria, making them unlikely to be important targets of malaria-related selection. In contrast, low-throughput expression experiments are typically based on a priori knowledge or hypotheses of the more direct roles of a few host genes in malaria.
Focusing on candidate genes may inflate the rate of false positives in genetic association studies [77]. Here, we make substantial efforts to ensure that any bias potentially related to PPIP identification, such as the popularity of a gene, does not impact our results (S1–S3 Figs). While we cannot guarantee that every PPIP is a true positive, in part because replication has often not been attempted, PPIPs as a whole do appear to represent a meaningful class of genes. In general, misclassification of either PPIPs or non-PPIPs for any reason (false negatives or false positives) would reduce any true difference between the two categories, weakening our results.
We used BLAT to identify homologs of 22,074 human coding sequences in 24 high-depth mammal genomes (Fig 1). We retained orthologs which (1) had best reciprocal hits in all 24 mammal species, (2) lacked any in-frame stop codons, (3) were at least 30% of the length of the human sequence, and (4) had clearly conserved synteny in at least 18 non-human species. Coding sequences for the resulting 9,338 proteins were aligned with PRANK (S1 File), and any codon present in fewer than eight species was excluded from analysis. Additional details are available in [31].
GO annotations were downloaded in October, 2015 from the Gene Ontology website [78] (http://geneontology.org/). Tests of enrichment were performed using Fisher's Exact Test.
Expression data for 53 human tissues were downloaded from the GTEx portal (http://www.gtexportal.org/home/) on March 6, 2017. These 53 tissues were condensed into a set of 34 tissues by averaging the RPKM across multiple tissue samples from adipose, aorta, artery, brain, cervix, colon, esophagus, heart, and skin. Each data point was then transformed as log2(RPKM+1). Total expression for each gene was calculated as the sum of these transformed values across all 34 tissues.
Regions of DNaseI hypersensitivity, combined from 95 cell types, were obtained from the databases of the ENCODE Project Consortium [79] (https://www.encodeproject.org/). The density of DNaseI hypersensitivity regions was calculated in 50 Kb windows centered on each ortholog.
Protein expression levels were obtained from the Human Proteome Map [80] (http://www.humanproteomemap.org/), which used high resolution and high accuracy Fourier transform mass spectrometry experiments. We summed spectral values over 30 tissues and cell types and took the log of these total values.
The log number of interacting partners for each human protein was obtained from the Biogrid Database [81] (http://thebiogrid.org/), curated by [82].
Genomic elements conserved in 46 vertebrate species, derived from PhastCons [43], were downloaded from the UCSC genome browser (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/phastCons46way/). Conserved element density was calculated within 50 kb windows centered on each gene in the human reference. Coding density was calculated from coding nucleotides in the same 50 Kb windows. The length and GC content of each protein was derived from the mammalian alignment.
The citation frequency of each gene was determined by the number of citations linked to its PubMed Gene page (http://www.ncbi.nlm.nih.gov/gene) as of May 11, 2017.
Polymorphism data for great apes—chimpanzee, gorilla, and orangutan—was obtained from the Great Apes Genome Project [83]. For each individual species, the counts of polymorphic sites are low, making the pN/pS ratio a noisy measure. This problem was alleviated by combining all the great ape data, which provided an overall control for the level of purifying selection across multiple species.
Virus-interacting proteins (VIPs) were manually curated in [31] in the same manner as PPIPs. To our knowledge, no similar collection of high-quality interactions is available for other pathogens. Therefore, we queried the EBI IntAct database (http://www.ebi.ac.uk/intact/) for protein interactions between Kingdom Bacteria (taxid:2) or Phylum Apicomplexa (taxid:5794) and humans (taxid:9606). This approach, while much faster than manual curation, is less ideal for two reasons: (1) many interactions are not included in the database (e.g., only 17 human-Plasmodium interactions are included in IntAct), and (2) many of the included interactions are based on high-throughput assays, including yeast two-hybrid experiments, which suffer from both false negatives and false positives [84]. Consequently, we do not perform rigorous analysis specifically for bacterial-interacting proteins, as has been done for PPIPs and VIPs [31]. Rather, we use IntAct data on bacterial interactions only to classify PPIPs as 'multi-pathogen' or not (Fig 5).
PPIPs were compared to other sets of genes using a permutation test of the mean. That is, the mean value for PPIPs was compared to the mean value of many sets (1,000–10,000) of control genes. P-values were defined as the fraction of permutations where the control mean was more extreme (usually, higher than) than the PPIP mean.
PPIPs differ from other genes in a number of ways (S1 Fig). In order to evaluate PPIPs against a fair background, sets of control genes were selected that were matched to PPIPs by important functional and evolutionary metrics (S2 Fig). This matched permutation approach allows the evolutionary effects of interacting with Plasmodium or Piroplasms to be isolated from correlated factors that may also influence evolutionary rate.
For the analyses shown in Figs 4, 5 and 7, each PPIP was matched to a set of control proteins based on similarity in five metrics: mRNA expression, protein expression, protein-protein interactions, DNaseI density, and conserved element density. A control protein was considered a PPIP 'match' if each of its five values fell within a given range, based on the PPIP values (S9 Table). For example, margins of min = 0.1 and max = 0.2 for mRNA expression would mean that, for a control protein to be matched to a PPIP, the mRNA expression of the control must fall between 90–120% of the mRNA expression of the PPIP. The goal was to maximize the number of matched controls per PPIP while creating control sets that were statistically indistinguishable from PPIPs for all five metrics (S2 Fig). To achieve this balance, maximum margins were iteratively chosen that yielded average p-values for all metrics of at least 0.1 over 100 permutations. Once appropriate margins were found, matched control sets of equal size to the PPIP set were obtained by randomly sampling one matched control protein for each PPIP.
Margins for the main permutation test (Fig 4) are given in S9 Table. For subsets of PPIPs (e.g. Fig 5A), the margins were altered to generate well-matched controls in every case. About 90% of PPIPs were typically matched and the rest excluded. Sets of matched controls were chosen based on the distribution of PPIP values included in each test, so whether any given PPIP was matched depended on the other PPIPs in the test (i.e., one extreme PPIP may or may not be balanced out by another). Therefore, the sum of matched PPIPs across categories differs slightly from the total.
Notably, the pool of immune controls is relatively small (966 genes) compared to the pool of non-immune controls (7,548 genes)(S3 Table). This made it difficult to match immune PPIPs to immune controls without discarding many immune PPIPs. Consequently, to test hypotheses of faster immune adaptation, we compared all PPIPs to all controls and non-immune PPIPs to non-immune controls (Fig 5).
The codeml model m8 from the PAML package [85] was used to estimate dN/dS for each gene across 24 mammal species (Fig 4B). However, branch-site tests in PAML rely on assumptions that may be violated in the case of recurrent adaptation to a pervasive selective pressure (see [31]). Consequently, we also implemented maximum-likelihood branch-site tests in the better-performing HYPHY package [44]. The BUSTED algorithm [45] was used to detect overall evidence of positive selection at any branch in the mammalian tree, and BS-REL was used to estimate the proportion of positively selected codons in each gene on each branch. Both of these algorithms rely on the same underlying codon model; details of the model are described in [44, 45] and reviewed in [31]. Unless otherwise specified (i.e., Fig 4F), codons identified by BS-REL were 'counted' as adaptive if the BUSTED p-value for that gene was ≤0.05.
We note that we did not employ the classical McDonald-Kreitman (MK) test to test for adaptation across multiple branches of the mammalian tree. The MK test estimates the proportion of adaptive substitutions in a protein for a single lineage, based on polymorphism within that lineage and comparison to an outgroup [86]. Here, our questions concern the evolution of proteins in multiple lineages, many of which lack polymorphism data. The methods in HYPHY are designed to simultaneously test for adaptation in multiple lineages, explicitly within the context of their phylogenetic relationships, based on a single sequence from each species. Therefore, BS-REL, BUSTED, and MEME are more powerful and appropriate for our data and questions than the MK test.
We split the mammal-wide alignments for each gene into four non-overlapping alignments corresponding to the following clades: primates (human, chimpanzee, gorilla, orangutan, gibbon, macaque, baboon, marmoset, bush baby), rodents (mouse, rat, guinea pig, squirrel, rabbit), carnivores (panda, ferret, dog, cat), and artiodactyls (sheep, cow, pig). We excluded microbat, elephant, and horse, as these species are not closely related to any of the four major groups [15] (Fig 1). However, we included rabbit with rodents because they are more closely related. We ran BUSTED on each alignment to yield a p-value of clade-specific adaptation for each gene.
PPIPs were matched to controls as described above (Methods, Permutation Tests). However, rather than counting BS-REL adaptive codons in all branches if the tree-wide BUSTED p≤0.05, we (1) kept each clade codon count separate, (2) counted codons only on branches within a clade, and (3) counted codons only if the clade-specific BUSTED p≤0.05.
An attempt was made to examine clade-specific adaptation in Plasmodium-only and Piroplasm PPIPs separately. However, down-sampling PPIPs to the numbers actually present in each subset resulted in dramatically increased variance in the estimates of adaptation, which eliminated statistical power to distinguish between PPIPs and controls.
Alpha-spectrin homologs were initially identified in 88 mammal species using NCBI Gene (http://www.ncbi.nlm.nih.gov/gene/?Term=ortholog_gene_6708). The sequence of the longest mRNA transcript for each species was downloaded using E-Utilities, and each transcript was trimmed to the longest ORF using TransDecoder [87] (http://transdecoder.github.io/). Coding sequences with <50% of the human CDS length were removed. The remaining 85 coding sequences were aligned with PRANK [88] using default settings (S7 Table). The alignment was manually inspected and corrected using JalView [89]. A phylogenetic tree for the 85 species was also obtained from phyloT (http://phylot.biobyte.de/) using NCBI Taxonomy (S10 Table).
To analyze positive selection in specific domains of alpha-spectrin, we employed the HyPhy test MEME [61] rather than BS-REL. For a given gene, BS-REL estimates a proportion of codons under positive selection on each branch of a phylogeny, but does not specifically identify the adaptive codons. In contrast, MEME tests each individual codon for positive selection across all branches. MEME also estimates a probability of adaptation for that codon on each branch. Unlike BS-REL, then, MEME is capable of identifying specific codons that have evolved adaptively.
We used the domain designations from SMART [90] (http://smart.embl-heidelberg.de/) to assign 92.2% of SPTA1 codons to one of 25 domains (S8 Table). Then, for each domain, we calculated an 'adaptation score' as:
a/v
where a measures adaptation (the proportion of codons within the domain with MEME p≤0.01*) and v measures variability (the proportion of codons within the domain that vary among species, i.e., are not 100% conserved). This score controls for domain length as well as the presence of invariable sites, as both components represent proportions of codons within the domain. To calculate the significance of each domain's adaptation score (i.e., to ask, is it higher than expected?), we randomly permuted codons among domains 10,000 times.
*We also tested MEME p-value cutoffs of 0.1, 0.5, 0.005, and 0.001 for defining a; these results are available in S8 Table. The results for p≤0.01, which are reported in the main text, are representative across these cutoffs.
|
10.1371/journal.pntd.0006238 | Developing a Buruli ulcer community of practice in Bankim, Cameroon: A model for Buruli ulcer outreach in Africa | In the Cameroon, previous efforts to identify Buruli ulcer (BU) through the mobilization of community health workers (CHWs) yielded poor results. In this paper, we describe the successful creation of a BU community of practice (BUCOP) in Bankim, Cameroon composed of hospital staff, former patients, CHWs, and traditional healers.
All seven stages of a well-defined formative research process were conducted during three phases of research carried out by a team of social scientists working closely with Bankim hospital staff. Phase one ethnographic research generated interventions tested in a phase two proof of concept study followed by a three- year pilot project. In phase three the pilot project was evaluated. An outcome evaluation documented a significant rise in BU detection, especially category I cases, and a shift in case referral. Trained CHW and traditional healers initially referred most suspected cases of BU to Bankim hospital. Over time, household members exposed to an innovative and culturally sensitive outreach education program referred the greatest number of suspected cases. Laboratory confirmation of suspected BU cases referred by community stakeholders was above 30%. An impact and process evaluation found that sustained collaboration between health staff, CHWs, and traditional healers had been achieved. CHWs came to play a more active role in organizing BU outreach activities, which increased their social status. Traditional healers found they gained more from collaboration than they lost from referral.
Setting up lines of communication, and promoting collaboration and trust between community stakeholders and health staff is essential to the control of neglected tropical diseases. It is also essential to health system strengthening and emerging disease preparedness. The BUCOP model described in this paper holds great promise for bringing communities together to solve pressing health problems in a culturally sensitive manner.
| Buruli ulcer (BU) is a neglected tropical disease primarily found in West Africa largely effecting the rural poor. BU has a known cause and cure, but an unknown route of transmission and a poorly understood incubation period. If not treated early and in a timely manner, BU often progresses to an advanced state requiring surgery and prolonged wound care. In the Cameroon, previous efforts to mobilize community health workers and educate community members to identify cases of BU yielded poor results. In this paper, we describe steps undertaken to create a successful BU community of practice (BUCOP) composed of community stakeholders working in concert with clinic staff. The success of the BUCOP was measured in terms of numbers of suspected BU cases referred and confirmed, a decline in treatment drop out, and sustained collaboration among stakeholders both during and following the pilot project. Pilot project success is attributed to an innovative and culturally sensitive approach to BU outreach education, increased levels of patient assistance, and mutual respect among BUCOP members for what each stakeholder contributed to BU detection, treatment, psychosocial support, and spiritual protection.
| Buruli ulcer (BU) is one of several neglected tropical skin diseases that afflict the rural population of sub-Saharan Africa, especially the poor living in areas with limited access to health infrastructure [1,2]. BU stands out as one of the most disabling of all neglected tropical diseases. The large majority of cases of BU have been identified in West Africa, particularly the countries of Benin, Cameroon, the Democratic Republic of the Congo, Cote D’Ivoire, Ghana, and Nigeria [3].
BU is caused by Mycobacterium ulcerans (MU), a microorganism belonging to the same genus of bacteria as tuberculosis and leprosy. BU has a known cause and cure, but an unknown route(s) of transmission and poorly understood incubation period [4, 5, 6 7]. BU manifests as necrotizing cutaneous lesions. Thirty-five percent of lesions are located on the upper limbs, 55% on the lower limbs, and 10% on other body parts [5]. In Africa, about 48% of those affected with BU are children under 15 years of age [4], and males and females are affected equally.
If not treated early and in a timely manner, BU often, but not always, progresses to an advanced state requiring prolonged wound care and skin grafting. If left untreated or treated late, BU does not kill, but may render the afflicted permanently disabled. Early diagnosis and treatment are the only ways to minimize morbidity and prevent disability [8]. Notably, at present most cases (68%) of BU are diagnosed at a late stage—categories II and III of the disease [9]. Beyond non-identification in its early stages, and delay in seeking care, there is some evidence suggesting that distinct phenotypes of BU may be more likely to progress to severe forms [10].
Prior to 2005, when effective antibiotic treatment was discovered for treating early stages of BU, all cases required surgery. Antibiotic treatment with streptomycin injections and oral rifampicin for 56 days proved to be highly successful in early (category I) BU cases. Once treated with appropriate antibiotics, there was a very low rate of BU relapse [11]. Clinical trials of oral treatment for the early stages of BU using rifampicin and clarithromycin have been found promising and recently approved for use by WHO [3]. The key to community management of BU is identifying cases in the early stages of the disease and enrolling the afflicted in treatment programs with minimal delay and sustained adherence to treatment.
In this paper, we describe a pilot project conducted in Bankim District, Cameroon that proved to be highly successful in establishing a BU community of practice (BUCOP) (Fig 1). A community of practice (COP) is an assemblage of stakeholders committed to a common objective, a common basic understanding of a focal problem, and mutual respect for what each stakeholder contributes to a process of problem solving [12,13]. In the case of BU, this entails health staff, community health workers (CHWs), and traditional healers sharing a common understanding of the signs of BU, collaboration in encouraging the afflicted to seek and continue BU treatment, open lines of communication between stakeholders, and mutual respect for what each contributes to a process of healing that includes, but extends beyond the management of BU as a disease.
BU has been identified in 64 of Cameroon’s 179 districts. Bankim District, the focus of this paper, is located in the northwest Adamawa region of Cameroon bordering Nigeria. As noted in Fig 2, the district has one of the three highest prevalence rates for BU in the country [14].
The central treatment and referral hospital for BU in Cameroon is the Ayos District Hospital located in the southern part of the country. Until recently, BU patients from all over Cameroon had to travel considerable distances to be treated at Ayos, a journey many were reluctant to make. Bankim is located over 475 KM away from Ayos and over 10 hours by local transport. Over the last decade, Cameroon’s National BU Control Program has trained clinic staff to provide treatment for BU in its early stages in many regions of the country and established five diagnostic and advanced treatment centers. Another two BU treatment centers are soon to be functional [14].
Bankim is located in a remote region of Cameroon with environmental factors favoring the presence of the MU microorganism responsible for BU. The district is situated in the Mape River Valley, where a dam was built to generate hydro-electric power more than 25 years ago. The Mape Dam splits the area into isolated islands and scattered villages. In the last two decades, increased irrigation has enabled rice cultivation. Rice farming has been identified as a possible risk factor for BU transmission [15,16]. Inhabitants of the region also engage in the growing of maize, cassava, and peanuts as well as various forms of hunting and fishing. Much agriculture is done on plots of land some distance from villages during the months of January to May, and many inhabitants seek employment in Nigeria from November to April. Population movement is both fluid and seasonal.
Health services in Bankim district include a district hospital and 5 satellite clinics. The hospital, at the onset of the project, was staffed by one doctor, 2 nurses, 4 nurse assistants, and one lab technician. The 5 satellite clinics were staffed by one nurse and 1–3 assistants. Two of the 5 satellite clinics had a lab technician conducting basic laboratory analysis. Each large village in Bankim has one community health worker/ volunteer (CHW) who assists hospital staff with outreach activities when requested to do so. There are 86 primary schools in the district and 10 secondary schools. School attendance waxes and wanes depending on season and agricultural activities.
Bankim Health District is a challenging place to initiate a community outreach program for BU due to both its rugged terrain and the wide variety of ethnic groups inhabiting and moving in and out of the region. These groups speak a variety of languages and dialects in addition to French and Pidgin English. The region is inhabited by Tikars, Yambas, Mambilas, Kwanjas, Fulanis as well as ethnic groups hailing from the neighboring Western, North West, Central, and Adamawa regions of Cameroon, and Nigeria. This required BU outreach activities to be carried out in multiple languages and for the research team to seek the approval and support of Christian and Muslim clerics, influential traditional healers, and local chiefs. There are seven paramount chiefs responsible for the welfare of Bankim district. The project staff had to gain the permission from each paramount chief before they could initiate outreach activates in their domain. A proof of concept study was first initiated in the jurisdiction of the paramount chief of Bankim town, a forward thinking, but cautious leader. Once the project was deemed feasible, other chiefs agreed to sanction community based activities conducted during the pilot phase of the project.
This community-based intervention employed a seven-stage formative research process [17] summarized in (Table 1) and adapted for BU. The formative research process covers all aspects of an intervention from the collection of baseline data and problem recognition to the generation and weighing of possible interventions from the vantage point of different stakeholders to project implementation, monitoring and evaluation.
The three phases of the project are summarized in Table 2. Phase one focused on baseline data collection, problem identification, and the generation and assessment of intervention options. Phase two had two parts. Part one entailed a proof of concept study to test the feasibility of promising interventions on a small scale. Part two applied lessons learned in the proof of concept study to a large pilot study. In phase three, outcome, process and impact evaluations of the pilot intervention were carried out.
In the first phase of the project, a team of three Cameroonian anthropology graduate students and their research advisers conducted interviews with current (N = 69) and former BU patients (N = 22) as well as health staff (N = 15), traditional healers (13), and CHWs (19). These interviews probed predisposing, enabling, and health service related factors influencing health care seeking for chronic ulcers, and reasons for treatment delay and drop out among BU patients offered free treatment. The team identified barriers to treatment adherence related to: 1) cultural perceptions of why wounds do not heal in a timely manner; 2) fear of hospital treatment and trust in hospital staff, and 3) pragmatic issues such as poor transportation, housing, and the availability of food for patients and caretakers when treatment requires hospitalization. The three anthropologists were then embedded in separate communities to investigate the current role of community health workers and healers in chronic wound management, existing BU detection activities, patterns of treatment referral, and household wound care decision making in different seasons. Health staff interaction with BU patients, CHWs, and healers were also observed in the community, at the district hospital, and in local clinics. Research methods employed included participant observation, key informant interviews, prospective and retrospective case studies of BU patients, semi-structured interviews incorporating “what if” scenario, and observations of social interactions between health staff and community members.
At the end of four months of intensive ethnographic research, the anthropological team presented their findings at a workshop attended by the doctor in charge of Bankim district hospital, the heads of the Cameroon National BU Program, and representatives of the NGO Fairmed providing support for neglected tropical diseases (NTD) programs in the region. In keeping with stages 2–4 of formative research, the team also presented data on what different stakeholders saw as possible means of more proactively involving community members in BU detection and the kinds of support that might reduce treatment delay and drop out.
Two major concerns were raised at the workshop. The first concern entailed the need to pay respect to local culture while at the same time addressing cultural beliefs and practices that posed barriers to BU detection and treatment. This required gaining the trust of local leaders and traditional healers and enlisting their support in a new BU outreach initiative. The key question posed was: Would it be possible to involve traditional healers in community-based BU outreach such that they become part of the solution, rather than a major part of the problem of BU treatment refusal, delay, and drop out? The second concern raised was pragmatic. What could be done to reduce the difficulties faced by impoverished patients faced with having to travel long distances to clinics for daily outpatient treatment or required to remain at Bankim hospital as an inpatient for months? The question posed was: would providing transport to the clinic, and food and lodging when necessary, increase the local population’s willingness to seek BU treatment early and adhere to treatment guidelines?
Phase two of research entailed a small- scale proof of concept (POC) study testing the feasibility of a package of proposed interventions to enhance BU outreach and establish a BU community of practice. The twin objectives of the intervention were to raise consciousness about BU using mass outreach events, and to use these events as an opportunity to establish collaborative relationships between clinic staff, chiefs, CHW, traditional healers, and recent BU patients who had a positive treatment experience. A second part of the POC was testing different types of patient support.
The POC study produced positive outcomes (reviewed in the Results section) warranting a larger scale pilot study in Bankim district. In the second part of phase two, a three year pilot study was launched after four modifications were made. The first modification entailed the use of a new WHO promotional video for BU in outreach programs prior to presentation of the educational PowerPoint. Videos are very popular in rural Cameroon. Although the new video was not designed for educational purposes, the theme of hope portrayed was in line with the BUCOP outreach program. Research revealed that while community members did not comprehend much of the language contained in the video, they were happy to see images of patients who had recovered from BU after treatment. The messages in the video did not duplicate nor clash with the messages presented in the educational power point presentation.
A second modification was using CHWs to engage in simultaneous translation into local languages. Hospital staff would present PowerPoint slides in French while a community health worker would translate the messages into local languages (Fulani, Pidgin English, Kwanja, Mambila, Yamba, and Tikar). Teams of presenters adopted a familiar presentation style commonly used in Pentecostal churches with messages passed back and forth between languages in a free and easy style. Pilot research revealed that although repetitive for speakers of multiple languages, translation was responded to positively by audiences even though it increased the duration of the program.
A third modification involved expansion of the number of traditional healers involved in the BUCOP. The healers participating in the POC were carefully screened as exemplars to model best practices. In the pilot study, three other healer groups located in the district were invited to participate in the BUCOP. Members of these healer groups were trained using the BU PowerPoint presentation as a common reference point for instruction. Healer groups met every month to discuss cases and each group monitored members’ adherence to a BUCOP contract. Traditional healers were also invited to participate in community outreach activities in their locales, often working along with CHWs. As in the POC study, they were paid a small honorarium for their efforts.
A fourth modification entailed an upgrade of the halfway houses and hospital wards. During the POC, concern was expressed about unhygienic conditions in the temporary halfway houses. Two halfway houses with bore wells were constructed in rural areas and visited by health staff traveling by motorcycle. At Bankim hospital, both a BU ward and a well laid out dressing room were constructed. A second surgeon was stationed at the hospital such that most skin grafts could be carried out in Bankim rather than being referred to Ayos hospital.
As noted in Table 2, the role of the team of anthropologists shifted over the course of the pilot project. During the POC and the first year of the pilot study, the anthropologists played an active role assisting in the implementation of the intervention package and monitoring interventions, enabling mid-course correction. Anthropologists acted as change agents and were consulted by clinic staff when problems arose. Gradually, implementation of all intervention activities was turned over to hospital staff, CHWs, and leaders of traditional healer groups. By the second year of the pilot study, anthropologists assumed the role of participant observers, monitoring outreach and referral activities, and documenting cases of successful partnerships as well as the ways in which members of the BUCOP solved problems. Then, in the third year of the pilot, the social scientists left Bankim for 12 months to see if BUCOP activities would be sustained without their presence as cultural brokers.
Phase three of the project took place one year after the social science team left Bankim. The team returned and conducted outcome, process, and impact evaluations to assess the effectiveness of the intervention in terms of BU detection, treatment adherence, and BUCOP stakeholder collaboration without the presence of social scientists as change agents. Attention was focused on whether COP member partnerships and lines of communication were sustained.
During this evaluation phase of the project hospital records were reviewed and 44 interviews and twelve focus groups were carried out with 22 CHWs, eighteen healers, nine health staff, nine former and eight recent patients, and five government administrators responsible for health activities in the district. The three anthropologists who had carried out stage one formative research investigated shifts that had occurred in community stake holder relationships with clinic staff as well as each other, task sharing, and changes in the social status of health staff, CHWs, and traditional healers. Broad impacts of the project were assessed as well. Chief among these was whether collaborative relations established by the BUCOP were being leveraged and extended to other health initiatives, and whether the BUCOP model constituted a viable means of promoting trust between health center staff and community stakeholders.
Ethical clearance and research authorization for the project was secured from the National Ethics Committee of the Cameroon Ministry of Health. The District Medical and Sub-Divisional Officers granted authorization for local entrance into the region and community leaders beginning with the Paramount Chief of the Tikars approved of the project. Informed consent was obtained orally from all adult participants in the project after being assured that their participation was voluntary, that the information they shared was confidential, and that they had the right to decline to be interviewed at any point during the project. Oral consent was necessitated given both low rates of literacy and the need to communicate details about the project in local languages, some of which are only spoken.
Research results are presented by phase of research.
Four sets of observations made during phase one formative research may be briefly highlighted: 1) local perceptions of chronic ulcers encompassing BU, 2) health care seeking patterns for chronic ulcers, 3) health staff, CHW and healer interactions; and 4) problems in BU identification and treatment warranting intervention.
Chronic ulcers that do not heal, a hallmark of BU, are often but not always attributed to the local disease category Mbouati (Atom in other parts of Cameroon) [18,19,20] a spirit affliction that is also the sign of special power accorded to the afflicted. Informants across all ethnic groups voiced the opinion that both Mbouati and BU co-existed in the region, some believing that only healers could determine the difference between the two and transform Mbouti into a chronic physical ulcer (nbong: a widely used Tika term) amenable to successful treatment. A common perception was that dual illness causality could be responsible for ulcers that do not heal. Such ulcers could either be caused or complicated by a combination of natural and supernatural factors [20]. There was also widespread speculation that in recent years the type of ulcer health staff call BU had increased in the region following the dam project and the introduction of rice cultivation.
Traditional healers are very popular in Bankim and often turned to as a first source of treatment for skin lesions, especially if they do not heal and are linked to witchcraft or Mbouati. Healers commonly use herbs, incantations, talismans, and “vaccination” (cutting, burning wounds, etc.) to treat chronic ulcers. While healers work independently, many are members of healer groups, which answer to the paramount chief of Bankim, recognized to be the chief of healers in the district. Healers have close ties to village chiefs and are called upon to offer blessings and protection to the community.
The hospital was not a place that villagers commonly readily turned to for treatment of BU for three reasons. Community members were afraid of hospital based BU treatment as it was associated with operations and amputation. Second, they had little interest in being referred to Ayos hospital as it was far off and in a place foreign to them. Third, despite BU treatment being free, people feared the indirect costs of treatment and hospitalization. There was also some confusion about what kind of wounds were being treated free. While medicines for BU were supplied free of cost by the Cameroonian NTD program, other chronic ulcers that look like BU are not treated free.
Healers did not have close working relations with clinic staff and did not refer cases to them. In the five years prior to the research project, not one case of BU had been referred to Bankim Hospital or any satellite clinic by a traditional healer. When healers visited Bankim hospital at the request of patients, they did so secretly. Healers also had little contact with CHWs.
CHWs attended meetings at the district hospital when called to do so, but their role was passive. They were given standard WHO educational materials depicting the signs of BU, but were not involved in actively educating community members about BU. For the most part, CHWs only identified possible BU cases when the disease control officer (DCO) from Bankim hospital personally visited their village by motorbike and directly asked CHWs to be shown villagers with chronic wounds. In the five years prior to the research project, CHWs identified only 48 potential BU cases and all of these cases were in advanced stages. BU cases were detected by chance by health staff during vaccination campaigns, and by the DCO when assisting a foreign research team in identifying cases for a clinical trial testing thermotherapy as a possible means of treatment [21]. CHWs did not see themselves as having a clear role in organizing BU outreach activities and they were not in close communication with clinic staff. Health staff saw BU detection and treatment as the responsibility of the DCO and surgeon who headed Bankim hospital.
Four types of interventions were called for on the basis of formative research. First, a new outreach education program (described below) was needed to raise community awareness about BU and its treatment, address rumors undermining confidence in clinic based care, and foster hope as a means of diffusing fear about BU treatment. Second, CHWs and traditional healers needed to be mobilized, and given a proactive role in both BU case detection and referral as well as patient follow up and psychosocial support. Third, visiting a clinic for 56 days of treatment was challenging for members of many households due to travel difficulties and indirect costs. It was clear that education alone was not going to solve the problem of treatment delay and drop out. Transport, and when necessary, lodging and the feeding of patients and their caretakers was required. Fourth, more advanced BU patients unwilling to travel to a referral hospital like Ayos needed to be treated at Bankim Hospital. Upgrades in BU care needed to be made at the hospital and outreach programs needed to inform the local population that high quality treatment for BU was now available locally at the Bankim district hospital.
An intervention package addressing these four intervention priorities was developed after options were weighed in keeping with stage four formative research. The first component of the package, seen as the cornerstone for building a BUCOP, was the introduction of a culturally sensitive community-based BU outreach program requiring clinic staff and community stakeholders to work closely together. An innovative education program was already in the process of being developed by teams of West African social scientists participating in the Stop Buruli Consortium, including the team from Cameroon. The education program developed and tailored for each consortium country (Benin, Cameroon, Ghana) is the subject of a forthcoming publication. In brief, it took the form of an image-rich PowerPoint presentation on BU delivered by local teams equipped with portable generators, laptop computers, LCD projectors and sound systems. The outreach program adopted a question–answer format enabling new issues to be added as they arose. Outreach meetings were interactive, not passive, and questions were invited from community members in attendance. As such, the educational presentation was the product of an iterative process. The social scientists investigated how best to respond to questions in a way that was at once scientific and understandable to local audiences. Messages and visuals were tested and changed as needed. Table 3 briefly summarizes the major sections of the PowerPoint presentation (available at https://www.fairmed.cm/defis/maladies-tropicales-negligees). Different messages were designed to inform and educate the community about BU, reassure community members about the quality of care available at clinics, offer hope of a cure, or display stakeholder collaboration.
Over the course of the POC the education program was developed, and pretested. Community outreach education meetings were held in the evenings in eight communities. CHWs were responsible for organizing meetings and inviting chiefs, local healers, and former patients to attend. Programs were treated as social performances where roles in the BUCOP were enacted. Chiefs and healers were seated in places conveying respect, and they were invited to voice their support for the program using a microphone, itself a symbol of power. Social scientists trained the DCO to deliver the PowerPoint education program during the POC and were on hand to assist in responding to questions from the community.
After the program was complete, hospital staff were then on hand to screen community members for wounds they suspected might be BU. During the POC study, 21 suspected cases of BU were identified by health staff either during or a few days following outreach meetings, (see Table 4 below). Patient’s samples were collected either using wound swabs or fine needle aspiration (for oedematous lesions). Of these, 19 (90.5%) were confirmed to be BU by Ziehl-Neelsen staining and/or polymerase chain reaction tests PCR), of which 10 (53%) were category I–early category II BU cases. In villages too remote to reach with audio and visual equipment, CHWs delivered key messages from the PowerPoint presentation orally, using posters and other visual aids depicting the signs of BU, and holding interactive question–answer sessions. Of cases referred to health staff by CHWs, 40% were deemed unlikely to be BU based on visual inspection. Of the remaining cases referred to health staff, 44 (62%) were confirmed by laboratory test to be BU and treated. Nineteen of these cases were either category I or early category II BU.
Following POC outreach educational activities, many cases of chronic ulcers and skin lesions were brought to the attention of CHWs or health staff by community members themselves. Ninety-eight of these cases were suspected to be BU by health staff, of which 45.9% were confirmed a by laboratory test and treated as BU. Of these cases, 25 were category I or early category II BU. As a point of comparison, between 2009 and 2010 no cases of BU like symptoms had been self-referred to clinic staff by community members. During the POC study, former BU patients were also encouraged to refer suspected cases of BU. They referred 17 cases to health staff, of which 11 (64.7%) were confirmed to be BU and treated.
To gain a sense of just how effective the POC was in detecting cases of BU, it is useful to compare POC results with the results of a house to house NTD survey conducted in Bankim district between late March and mid-April 2010. During the NTD survey researchers visited 9,344 households (48,962 people). The survey only identified 25 suspected cases of BU, of which six were confirmed to be BU by PCR [4].
A second component of the POC study was provision of free transport to clinics for category I and early category II BU patients not requiring hospitalization and living 3–7 KM from clinics. Motorcycle taxis were hired to deliver BU patients to the hospital for the duration of treatment. For those who lived too far to make this feasible, housing was secured for them near clinics in facilities termed halfway houses. Halfway houses were tested in both Bankim town and two rural areas. In Bankim town, patients staying at halfway houses received treatment and hospital staff routinely monitored their wounds. In the rural areas, local clinic staff traveled 8–10 km daily by motorcycle to administer treatment to patients at halfway houses. Patients and caretakers remaining at halfway houses were supplied a food ration. Seventeen patients took up residence in halfway houses during the POC. Anthropologists monitored social interaction between patients of different ethnic groups and found that they bonded around the shared experience of BU and supported one another during treatment.
The third component of the POC was assessing whether traditional healers would participate in BU outreach activities and become proactive members of a BUCOP. In brief, after obtaining support from the paramount chief of Bankim town, a meeting was held with a group of popular traditional healers who were members of a healer association. Ten traditional healers were selected to participate based on their willingness to collaborate with clinic staff in addressing both biomedical and traditional medical aspects of BU treatment. Clinic staff explained to healers that in order for their treatment to be effective, patients needed to be referred to them quickly and wounds needed to be left undisturbed and not treated with traditional medicines. Clinic staff acknowledged that they had no expertise in the mystical aspects of treatment such as removing spiritual affliction, protecting the patient when vulnerable to forces of malevolence, nullifying obstructions to the healing process, or dealing with patient fears of malevolent spirits during treatment. Healers were asked to attend to the spiritual and psychosocial aspects of treatment, while clinic staff used medications and bandaging to take care of the physical manifestations of the disease.
A contract was proposed and signed by members of this group of traditional healers. Two key components of the contract were that traditional healers promised to refer all patients with possible signs of BU to Bankim clinic staff within ten days of seeing them and not treat the skin of these patients. Clinic staff gave healers free access to Bankim hospital, satellite clinics, and halfway houses to offer patients psychosocial support and spiritual protection. Healers were also offered a small amount of money to pay for accompanying patients to a clinic for screening.
The social science team monitored traditional healers’ adherence to the contract. All ten traditional healers adhered to the contract, and six of the ten healers referred 49 suspected BU patients to Bankim hospital. Of the 49 suspected cases referred, 13 (26.5%) were confirmed to be BU by laboratory test. In the case of traditional healers, all cases referred were tested, even when staff thought the case was unlikely to be BU. This policy was followed after a healer challenged a case clinic staff dismissed as not being BU based on visual inspection. The healer was proven correct in her detection of BU.
The treatment adherence rate during the POC for BU patients detected was 94% compared to a rate of 54% in Bankim Hospital in 2009–2010. Increased adherence was due to both enhanced patient support by health staff, CHWs, and traditional healers and the provision of resources better enabling patients to remain in treatment. Notably, all BU patients referred by healers and visited by them in the hospital completed treatment.
During the pilot project an additional 89 CHWs and 55 healers were trained in BU detection. Forty-four well attended community outreach programs were conducted reaching approximately 15,500 people. In Tables 4 and 5 we present data on BU case referral and confirmation by stakeholders during both the POC and pilot interventions. The number of cases referred designates cases referred to health staff and deemed to be possible cases of BU by visual inspection. Confirmed cases were by Ziehl-Neelsen staining and/or polymerase chain reaction tests (PCRs). During the POC approximately 40% of total cases brought to the attention of health staff were dismissed as some other kind of skin lesion, abscess, or ulcer. During the pilot study this percentage dropped to 30% of all cases reported. Table 6 presents data on the category of confirmed BU cases referred by different stakeholders in the BUCOP. Table 7 presents data on adherence to treatment by confirmed cases of BU that initiated therapy. This data is present by the type of stakeholder who referred the case.
Several findings stand out as notable. First, prior to the project no cases of BU were referred to health staff by community stakeholders with the exception of cases identified when a disease control officer occasionally visited a community searching for NTD cases. During the project, 90% of all confirmed cases of BU were identified by community stakeholders. Second, as a result of the project not only was there a significant rise in the number of BU cases referred by community stakeholders, but a significant number of category I BU cases were detected and treated. Five hundred and twenty-two suspected cases of BU that health staff thought warranted laboratory confirmation were referred by community members. Out of these 522 cases, 266 cases (51%) were confirmed to be BU. Twenty-one percent of these cases were category I. Third, more suspected BU cases were referred by family members following outreach educational programs than any other BUCOP stakeholder. Their referrals accounted for 40% of all referrals of suspected cases of BU during the POC, and 46% of all referrals during the pilot study. In many instances, family members checked with CHWs and traditional healers participating in the BUCOP to ask their opinion about the lesion or to request that they accompany them to a clinic. In the same period, 23% of suspected cases were detected and referred by CHWs, of which 17% were confirmed of which 19%% were category I. Health staff only detected 39 cases of BU during their routine activities, of which only five cases (13%) were category I.
A fourth finding was that traditional healers actively participated in both the POC and pilot project and kept to the terms of the contract they signed. Traditional healers referred 19% of all suspected cases of BU during the pilot project, 15% of all confirmed cases, and 18% of confirmed category I cases. Fifth, the confirmation rate of suspected cases sent for testing was impressive. While 30% to 40% of lesions brought to health staff for inspection were dismissed as other skin aliments, a high percentage of the remaining 60–70% of cases were found to be BU. Of cases tested, 43% of those referred by CHWs, 32% by traditional healers, 52% by former patients, and 52% by family members were found to be positive. The rate of confirmation for traditional healers was lower because unlike the other stakeholders every case referred was sent for testing. If 60%–70% of their cases had been sent for testing after staff screening, their confirmation rate would have been similar to other BUCOP stakeholders. Sixth, treatment adherence rates for confirmed cases referred by all stakeholders were > 90%. One can compare this to a 54% BU treatment adherence rate two years prior to the project.
The goal of the pilot project was not just to mobilize individual stakeholders to become more actively involved in BU detection and referral, but to create an interactive and supportive BUCOP. Phase three evaluation research revealed that a functioning BUCOP had indeed emerged in Bankim. Lines of communication between CHWs, traditional healers, and health staff were well established and collaboration in BU outreach activities were ongoing. The status of CHWs increased markedly as a result of CHWs becoming actively involved in arranging outreach education and screening programs for their communities. A closer working relationship with clinic staff empowered them to play a more proactive role in both referring and following up suspected BU cases. The status of former BU patients also changed. The community appreciated their testimonials during large outreach meetings. Former patients came to be seen less as victims and more as survivors having BU treatment experience they were willing to share.
Clinic staff noted being surprised by the large number of BU cases referred by community stakeholders as a result of the project. They came to value the BUCOP, stating that it both enabled them to have a much closer working relationship with CHWs and traditional healers, and it enhanced the reputation of the hospital. One testament to growing trust in the hospital was consultation by members of the Fulani ethnic group. Members of this group had previously been reluctant to seek treatment for BU. Outreach education delivered in pidgin by a trained Fulani community health worker (also a traditional healer) and a few treatment success stories have paved the way for greater contact with this community. Additionally, the director of Bankim hospital has been praised by Cameroonian health officials for forging close working relationships with community stakeholders. Better working relations between health staff and CHW, and increased acknowledgment of each stakeholder’s contribution to BU program success served as important non–financial incentives enhancing stakeholder motivation [22].
A major question posed during phase one of the project was whether traditional healers would be willing to actively participate in a BUCOP. The impact evaluation found that healers are presently seen by clinic staff and health officials as having an important role in community-based BU management. Not only have traditional healers referred cases and honored their treatment contract, but they have routinely assisted CHWs during outreach activities, offered psychosocial support to patients who are hospitalized, and encouraged patients who dropped out of therapy to return to treatment. Health staff are now invited to traditional healer group meetings to discuss cases, and healers are invited to hospital meetings when outreach activities are being planned. Health officials were initially reluctant to give traditional healers identification badges designating them as members of the BUCOP. By the second year of the pilot project, officials felt that healers had proven their commitment to the BUCOP. Traditional healers were offered badges along with referral cards and health officials left it up to traditional healer groups to both monitor member adherence to the contract and to sponsor new healers who wished to join the BUCOP after receiving training.
A key issue investigated was what healers gained and lost from participation in the BUCOP. Some members of the NTD community initially expressed the opinion that traditional healer participation might be primarily motivated by funds received when referring patients. This opinion proved to be a false. Research revealed that healers lost far more financially then they gained by referring patients for treatment. At the time of the impact evaluation, funds for healers to accompany patients to the clinic were exhausted. Yet healers continued to refer as well as visit patients at the clinic without charging any fee in cash or in kind. During healer groups, cases were discussed and when one healer did not have the funds to take a patient to the clinic or visit them, another member of the group often offered to do so.
So what did healers get out of the becoming members of the BUCOP? Research revealed that the social and symbolic capital healers gained outweighed any financial loss they incurred for referring patients. Healers who visited Bankim hospital were warmly received by health staff who saw them as an asset in reassuring patients about their treatment. Healers took pride in offering spiritual protection to patients so that BU medications could act effectively, and for offering patients psychosocial and spiritual support while under treatment. When BU patients were cured, healers shared credit with health staff for treatment success. While healers received no payment for their BU related activities, they did receive gifts when patients were cured. Moreover, other patients residing at the hospital requested their assistance. In short, their reputation increased and was not diminished by collaboration with clinic staff.
An additional impact of the BU project was its contribution to the integration and control of other neglected tropical skin diseases. Although not intended to do so, BU outreach programs attracted a large number of cases of yaws, a disease not seen at clinics in the district for over a decade and assumed to be eradicated in this region of the Cameroon. Cases of yaws identified during BU outreach programs alerted health staff of the presence of the disease and they then conducted follow up school based screenings in these communities. Eight hundred and fifteen cases of confirmed cases of yaws were successfully treated [23].
We do not claim that the relationships established in the Bankim BUCOP can be formed in all contexts. Context must be taken into consideration. For example, traditional healers in Bankim are organized into groups which exercise some modicum of authority over their members. Such is not the case in all African contexts. Mobilizing individual healers would prove more challenging than mobilizing groups. A challenge that may be briefly mentioned is that when the reputation of a health facility rises so do patient load and resource need. In Bankim, the increased reputation of the hospital has drawn patients from outside the district and even neighboring Nigeria. Resources to maintain quality of care will need to be sustained or the reputation of the hospital will decline.
In a recent review of the impact of two decades of health social science research on NTDs, Bardosh [24] notes that while this research has generated important insights into health care seeking and community response to disease centered programs, it has not been effectively used in program development and implementation. This pioneering study illustrates how a multistage formative research process can contribute to the creation of a BUCOP. We would argue that the COP model of clinic staff–community stakeholder collaboration presented here has great potential for other community-based disease outreach programs in Africa and beyond. It’s inclusion of community stakeholders like healers, volunteers, and former patients extends the COP model recently advocated for establishing productive relationships between local experts, NGOs, ministries of health and government health staff, and foreign advisers [25, 26]. In Bankim, now that collaborative relationships have been established, they are being leveraged for other public health endeavors such as vaccination programs. Establishing a COP also has great potential for emerging disease preparedness. Should a disease like Ebola strike Bankim, collaborative relationships between clinic staff, CHWs, chiefs, and healers will enable a rapid coordinated response. If the 2015 Ebola outbreak in West Africa taught us anything, it is recognition that close ties to community stakeholders constitute an essential part of health system strengthening [27,28,29,30, 31,32]. Establishing trust and lines of communication with community leaders enables swift action and increased opportunities for local problem solving.
|
10.1371/journal.pgen.1002199 | Genome-Wide Association Analysis of Incident Coronary Heart Disease (CHD) in African Americans: A Short Report | African Americans have the highest rate of mortality due to coronary heart disease (CHD). Although multiple loci have been identified influencing CHD risk in European-Americans using a genome-wide association (GWAS) approach, no GWAS of incident CHD has been reported for African Americans. We performed a GWAS for incident CHD events collected during 19 years of follow-up in 2,905 African Americans from the Atherosclerosis Risk in Communities (ARIC) study. We identified a genome-wide significant SNP (rs1859023, MAF = 31%) located at 7q21 near the PFTK1 gene (HR = 0.57, 95% CI 0.46 to 0.69, p = 1.86×10−08), which replicated in an independent sample of over 8,000 African American women from the Women's Health Initiative (WHI) (HR = 0.81, 95% CI 0.70 to 0.93, p = 0.005). PFTK1 encodes a serine/threonine-protein kinase, PFTAIRE-1, that acts as a cyclin-dependent kinase regulating cell cycle progression and cell proliferation. This is the first finding of incident CHD locus identified by GWAS in African Americans.
| In the United States, African Americans are at high risk for coronary heart disease (CHD). Although environmental and social factors have a role, genetic factors also contribute to CHD risk and mortality. Research to identify genetic factors for CHD susceptibility has been carried out mostly in Europeans and European Americans and little has been done in African Americans. Genome wide association studies (GWAS) provide a means to identify susceptibility loci without any a priori assumptions about the functional importance of a gene. In this study, we used GWAS to identify a novel genomic region associated with incident CHD events in African Americans from the ARIC study and replicated this finding in a large sample of African American women. This region contains several genes, including PFTK1, that regulate cell cycle progression and cell proliferation. This is the first report of a susceptibility locus for incident CHD identified by GWAS in African Americans.
| Coronary heart disease (CHD) is the leading cause of death worldwide [1]. In the United States, African Americans are the most vulnerable population with regard to CHD risk factors and mortality. A recent American Heart Association report showed that African Americans are twice as likely to die from a heart-related disease compared to other ethnicities [2]. The presence of multiple CHD risk factors is 50% more likely in African Americans than in the population of European ancestry. Hypertension and diabetes are more prevalent and the highest rate of obesity is found in African American women [2]. The factors underlying these disparities are not well understood. Socioeconomic status and health care accessibility play an important role [3]. However, genetic factors are known to influence the risk of CHD [4] and population differences in the frequency and effects of these genetic factors also likely have a role.
Progress in the discovery of susceptibility genes for multiple chronic diseases and their risk factors has been made possible in recent years through genome-wide association studies (GWAS); African Americans were noticeably absent from most of these studies. GWAS have the advantage of discovering new genetic variants underlying a disease without a priori knowledge of gene location or function. More than 20 variants for CHD have been discovered in samples of European-descent so far [5]–[11].
In this study, we took advantage of rich longitudinal data on CHD incident events in the Atherosclerosis Risk in Communities (ARIC) study and performed a GWAS of incident CHD events in African Americans. Results that reached pre-specified genome-wide significance were investigated in African Americans from the Women's Health Initiative (WHI) [12].
Genome-wide association analysis for incident CHD was carried out in African Americans from the ARIC study. After 19 years of follow-up, 362 individuals developed CHD and 2,543 individuals were free of CHD. Descriptive statistics for multiple cardiovascular disease risk factors at the baseline examination are provided in Table 1. Three loci reached the pre-specified genome-wide significance threshold (p<5×10−8) (Table 2) and were considered for replication in a sample of African American women from WHI. Results are shown in Table 2. For all variants, the direction of effect was consistent in both studies, but only the variant rs1859023 reached statistical significance in WHI (Table 2). The rs1859023 minor allele A (frequency = 0.31) had a protective effect on CHD risk with a hazard ratio (HR) of 0.57 (95% CI 0.46 to 0.69, p = 1.86×10−08) in ARIC and 0.81 (95% CI 0.70 to 0.93, p = 0.005) in WHI. After adjustment for CHD risk factors (smoking, diabetes, LDL, BMI and hypertension), the result was only slightly less significant (HR = 0.59, 95% CI 0.48 to 0.73, p = 8.53×10−07 in the ARIC study). After exclusions of coronary revascularization procedures from the ARIC case definition, the association became stronger (HR 0.53, 95% CI 0.43 to 0.65, p = 2.95×10−09), possibly due to a more homogenous definition of CHD. The pooled hazard ratio of incident CHD for combined ARIC and WHI data was 0.72 (95% CI 0.64 to 0.81, p = 1.29×10−08).
rs1859023 was not significantly associated with any of the traditional CHD risk factors at the baseline ARIC examination (LDL, HDL, SBP, DBP and BMI, data not shown). However, average thickness of the carotid artery, a common measure of subclinical atherosclerosis, was significantly different among rs1859023 genotypes (mean difference in carotid artery thickness per copy of the minor allele β = −0.0168, SE = 0.0047, p = 3.39×10−04) which suggest a role for this variant in the atherosclerotic process. When including carotid artery wall thickness as a covariate in the analysis of incident CHD attenuated but did not abolish the association with rs1859023 (HR = 0.59, 95% CI 0.48 to 0.73, p = 1.58×10−06). We attepted to replicate this observation by testing the assocation of rs1859023 and right coronary fatty streak area (% of intimal surface area) in the PDAY study consisting of 1,452 African Americans and 1,342 European Americans but the results did not reach statistical significance using a one-sided test. In the ARIC European Americans, rs1859023 was not significantly associated with incident CHD. The smallest p-value in the region surrounding rs1859023 (±50 kb) was at rs17869240 (p = 0.01, MAF = 0.07).
rs1859023 is located in an intergenic region in proximity to the PFTK1, CLDN12 and GTPBP10 genes (Figure 1). PFTK1, also known as CDK14, encodes a serine/threonine-protein kinase PFTAIRE-1 that acts as a cyclin-dependent kinase regulating cell cycle progression and cell proliferation [13]. It is highly expressed in heart tissue [14]. In the ARIC study, rs1859023 is in loose linkage disequilibrium (LD) with other SNPs in the region (r2< = 0.4). In the Yoruba HapMap data, there are only 3 SNPs with LD greater than r2 = 0.8, all located within a very short distance of rs1859023 (Figure 1) which implies that the tagging region of rs1859023 is very narrow in African-derived populations. Given the location of the rs1859023 5′ to the PFTK1 gene, these data imply that rs1859023 may affect gene expression. To test this hypothesis, we undertook expression QTL analyses using the resources provided at the SCAN – SNP and CNV Annotation Database (http://www.scandb.org/newinterface/about.html) [15]. rs1859023 predicts the expression of 6 genes with p-values less than 10−4 (Table 3). Interestingly, the evidence of rs1859023 predicting the gene expression is found only in the Yoruba population.
A variant close to PFTK1, rs10499903, located ∼60 kb from rs1859023 was associated with ankle brachial index (ABI) in European Americans from the Framingham Heart Study (FHS) [16]. Given the different ethnic backgrounds of the two studies, and expected difference of LD patterns within the region between two samples, the Framingham result strengthens our finding and further suggests a role for the PFTK1 gene region in the atherosclerotic process.
Some of European-discovered CHD genes have been reported to also influence CHD in African Americans [17]. However, this is the first reported GWAS finding of a CHD risk locus in African Americans. Although the sample size is less than contemporary GWAS publications in European-Americans (i.e. often exceeding 100,000 individuals (e.g. [18]), we have combined all of the available well-powered incident CHD data in African Americans with genotype data and are able to present results based on 133,415 person-years of follow-up. In addition, supporting evidence is provided by the association with subclinical atherosclerosis and expression QTL analyses. In conclusion, we have identified a region near the PFTK1 gene as being associated with incident CHD and subclinical atherosclerosis in African Americans. Further studies are needed to examine the cellular or metabolic mechanisms underlying this association, and large population-based studies of minority populations are necessary to more fully understand the impact of genetic factors on multiple phenotypes in those that bear a disproportionate burden of disease.
The ARIC (Atherosclerosis Risk in Communities) study is a population-based prospective cohort study of cardiovascular disease and its risk factors [19]. ARIC includes 15,792 persons aged 45–64 years at baseline (1987–89), randomly chosen from four US communities. Of these individuals, 4,266 are self-reported African Americans. Cohort members completed four clinic examinations, conducted approximately three years apart between 1987 and 1998, and followed with annual phone interviews since 1987.
Incident CHD in ARIC was ascertained by contacting participants annually, identifying hospitalizations and deaths during the prior year, and by surveying discharge lists from local hospitals and death certificates from state vital statistics offices for potential cardiovascular events [19]. A CHD event was defined as a validated definite or probable hospitalized MI, a definite CHD death, an unrecognized MI defined by ARIC ECG readings, or coronary revascularization. Participants were excluded from analyses if they had a positive or unknown history of prevalent stroke, transient ischemic attack/stroke symptoms, or CHD at the initial visit and/or being of non–African American ethnicity. Real-time, B-mode ultrasound was used to evaluate the carotid arterial intima-media wall thickness as an indicator of atherosclerosis in the ARIC study and the detailed description of its measurement is described elsewhere [17].
Genotyping was done in 15,020 ARIC participants using the Affymetrix Genome-Wide Human SNP Array 6.0. A total of 3,182 individuals remained after excluding individuals of non African American ethnicity, subjects who did not consent DNA use, unintentional duplicates with higher missing genotype rates, suspected mixed/contaminated samples, scans from one problem plate, samples with a mismatch between called and phenotypic sex, samples with genotype mismatch with 39 previously genotyped SNPs, suspected first-degree relative of an included individual, and genetic outliers based on average IBS statistics and principal components analysis using EIGENSTRAT. SNPs were excluded due to having no chromosome location, being monomorphic, having a call rate <95% and HWE-p<10−5. In this analysis, we considered only variants with a MAF greater than 10%.
Cox proportional hazards models with adjustment for age, gender and the first three principal components derived from EIGENSTRAT were used to estimate CHD hazard rate ratios (HRs) over a 19-year period (362 cases) under an additive genetic model. These analyses were done using PLINK and an R application for survival regression analyses. We define as “genome-wide significant” all associations with p<5×10−8. We define replication to be a significant (p<0.05) and directionally consistent association in an independent sample.
The WHI has two major components: (1) a clinical trial that enrolled and randomized 68,132 women ages 50–79 into at least one of three clinical trials; and (2) an observational study that enrolled 93,676 women ages 50–79 into a parallel prospective cohort study [12]. WHI participants were recruited from 1993–1998 at 40 clinical centers across the U.S. During follow-up, incident CHD events were adjudicated locally and centrally from medical records including hospital discharge summaries, ICD-9 codes, diagnostic, laboratory, surgical, and pathology reports by trained physicians blinded to randomized intervention and exposure status [20]. In the WHI replication sample, CHD was defined as MI, coronary revascularization, hospitalized angina, or CHD death. Definite and probable nonfatal MI required overnight hospitalization and was defined according to an algorithm based on standardized criteria using cardiac pain, cardiac enzymes and troponin levels, and ECG findings. CHD death was defined as death consistent with underlying cause of CHD plus one or more of the following: hospitalization for MI within 28 days prior to death, previous angina or myocardial infarction, death due to a procedure related to CHD, or a death certificate consistent with underlying cause of atherosclerotic CHD.
Of a total of 26,045 (17%) women from minority groups, 8,515 self identified African American women who had consented to genetic research were eligible for the WHI GWAS project. Genotyping was performed on the Affymetrix 6.0 array. After excluding samples due to genotyping failure, cryptic relatedness, and discrepancy between genetic ancestry and self-reported race, there were 8,421 WHI African Americans. Participants were further excluded from analyses if they had a positive or unknown history of prevalent stroke, transient ischemic attack/stroke symptoms, or CHD at the initial visit. A total of 862 incident first CHD events occurred among 8,155 eligible African American women without baseline CHD (Table 4). The mean age at study entry was 61.6+/−7.0 years (range 50–79). The mean baseline age of the cases was 64.2+/−7.2, and the mean baseline age of the non-cases was 61.3+/−6.9. The mean time to CHD event was 5.29+/−3.19 years. The mean age at CHD event was 69.5+/−7.5 years.
This study was approved by the participating institutional IRBs, and all ARIC and WHI participants provided written informed consent, involving the sharing of data with the scientific community.
The Pathobiolobical Determinants of Atherosclerosis in Youth (PDAY) study is composed of subjects who were 15 to 34 years of age when they died of non-CVD related causes (accidents, homicides or suicides). The purpose of PDAY was to evaluate early development of atherosclerosis [21]. For this replication analysis, we genotyped rs1859023 in 2,794 individuals from PDAY - 1,452 African Americans and 1,342 European Americans and tested the association with fatty streak area in the right coronary artery.
|
10.1371/journal.ppat.1006810 | Memory CD8 T cells mediate severe immunopathology following respiratory syncytial virus infection | Memory CD8 T cells can provide protection from re-infection by respiratory viruses such as influenza and SARS. However, the relative contribution of memory CD8 T cells in providing protection against respiratory syncytial virus (RSV) infection is currently unclear. To address this knowledge gap, we utilized a prime-boost immunization approach to induce robust memory CD8 T cell responses in the absence of RSV-specific CD4 T cells and antibodies. Unexpectedly, RSV infection of mice with pre-existing CD8 T cell memory led to exacerbated weight loss, pulmonary disease, and lethal immunopathology. The exacerbated disease in immunized mice was not epitope-dependent and occurred despite a significant reduction in RSV viral titers. In addition, the lethal immunopathology was unique to the context of an RSV infection as mice were protected from a normally lethal challenge with a recombinant influenza virus expressing an RSV epitope. Memory CD8 T cells rapidly produced IFN-γ following RSV infection resulting in elevated protein levels in the lung and periphery. Neutralization of IFN-γ in the respiratory tract reduced morbidity and prevented mortality. These results demonstrate that in contrast to other respiratory viruses, RSV-specific memory CD8 T cells can induce lethal immunopathology despite mediating enhanced viral clearance.
| Memory CD8 T cells have been shown to provide protection against many respiratory viruses. However, the ability of memory CD8 T cells to provide protection against RSV has not been extensively examined. Unexpectedly, mice with pre-existing CD8 T cell memory, in the absence of memory CD4 T cells and antibodies, exhibited exacerbated morbidity and mortality following RSV infection. We demonstrate that the immunopathology is the result of early and excessive production of IFN-γ by memory CD8 T cells in the lung. Our research provides important new insight into the mechanisms of how memory T cells induce immunopathology. In addition, our findings serve as an important cautionary tale against the use of epitope-based T cell vaccines against RSV.
| Respiratory syncytial virus (RSV) is a major cause of severe disease in young children, the elderly, and immunocompromised populations [1–6]. Furthermore, RSV is the leading cause of infant hospitalizations creating an immense healthcare burden for treatment and prevention [1, 2, 7–11]. There is currently no licensed vaccine for RSV. During a primary RSV infection, the CD8 T cell response is crucial for mediating viral clearance [12, 13]. Depletion of CD8 T cells in mice prior to RSV challenge leads to elevated viral loads, but also ameliorates morbidity [12]. Thus, CD8 T cells contribute to both viral clearance and immunopathology following an acute RSV infection. RSV-specific memory CD8 T cells also contribute to protection from a secondary infection [12]. Antibody-mediated depletion of memory CD8 T cells in RSV-immune mice impairs viral clearance following re-infection as compared to non-treated controls [12]. Thus, vaccines that elicit robust memory CD8 T cell responses may help promote long-lived immunity against RSV.
The induction of neutralizing antibodies remains a primary goal of most RSV vaccines due to their clearly established capacity to reduce the severity of RSV-induced disease [14–17]. In contrast, studies have demonstrated that robust memory CD4 T cell responses can mediate vaccine-enhanced disease following RSV infection [18, 19]. Adoptive transfer of activated effector RSV-specific CD8 T cells, in vitro stimulated T cell lines, or in vitro propagated T cell clones leads to enhanced RSV clearance from the lung following RSV challenge. These effector CD8 T cell transfers were also associated with increased weight loss, indicating that infusion of effector CD8 T cells can induce increased systemic disease [20–23]. However, the role of memory CD8 T cells in providing protection against RSV infection remains unclear. Evaluating the capacity of memory CD8 T cells to mediate protection against RSV infection is critically important because high neutralizing antibody titers alone are insufficient to prevent RSV-induced disease in every individual [14, 24].
Herein, we evaluated the protective capacity of memory CD8 T cells against RSV infection in the absence of RSV-specific CD4 T cell memory and antibodies. We employed a dendritic cell-Listeria monocytogenes (DC-LM) prime-boost immunization regimen to induce high magnitude, RSV epitope-specific CD8 T cell responses in naive mice. A similar prime-boost immunization strategy has been shown to elicit protection against other respiratory viruses including influenza A virus (IAV) and severe acute respiratory syndrome coronavirus (SARS-CoV) [25, 26]. DC-LM immunization induced robust memory CD8 T cell responses that reduced viral titers following RSV challenge. However, despite enhanced viral clearance, immunized mice experienced increased pulmonary disease, weight loss, and mortality. Exacerbated disease and mortality was unique to the context of an RSV infection as immunized mice were protected against challenge with a lethal dose of a recombinant IAV expressing an RSV-derived CD8 T cell epitope. The lethal immunopathology observed in immunized mice was caused by rapid and excessive IFN-γ production by memory CD8 T cells in the airways. Our studies reveal that memory CD8 T cells enhance RSV clearance similar to other viral infections, but are unique in that they mediate severe immunopathology caused by the overproduction of IFN-γ.
Peptide-coated, mature DCs can be utilized to prime a CD8 T cell response that allows for robust secondary expansion following a booster immunization in mice [27]. To induce RSV-specific CD8 T cell memory in the absence of virus-specific CD4 T cells and antibodies, naive mice were immunized with matured splenic DCs loaded with M282-90 (M282) peptide and boosted 7 days later via infection with an attenuated, recombinant LM strain expressing the M282 epitope. A control group without RSV-specific CD8 T cell memory was generated by immunizing mice with DCs not exposed to peptide followed by a boost with an LM that did not express an RSV-derived epitope. DC-LM immunization led to a significant (p<0.001) increase in M282-specific CD8 T cell frequencies following the LM booster inoculation within the peripheral blood leukocytes (PBL) compartment as compared to the control group (Fig 1A). Approximately 20% of all CD8 T cells in the PBL were M282-specific at day 42 post-boost (Fig 1A). Immunized mice challenged with RSV exhibited a significant (p<0.001) reduction in lung viral titers by day 4 post-infection (p.i.) compared to the control group undergoing a primary RSV infection (Fig 1B). As expected, immunized mice exhibited an increased number of total and M282-specific CD8 T cells in the lung by day 5 p.i. as compared to non-infected M282-immunized controls (Fig 1C and 1D). In addition, there was a greater frequency (p<0.001) of IFN-γ+ and IFN-γ+TNF+ CD8 T cells following M282 peptide re-stimulation at days 4 and 5 p.i. as compared to control groups (Fig 1E). Thus, DC-LM immunization elicited robust memory CD8 T cell responses that mediated enhanced viral clearance following RSV challenge.
Pre-existing memory CD8 T cells altered the response of specific cell types following RSV infection (S1 Fig). Both conventional CD4 and regulatory CD4 T cell (Treg) numbers were significantly reduced (p<0.05) at days 4 and 5 p.i. in the lung compared to the acute infection control group. Furthermore, the number of monocytes were significantly (p<0.05) increased at day 5 p.i. as compared to control mice. These changes were in contrast to eosinophil, neutrophil, and natural killer (NK) cell responses, which remained similar to non-infected M282-immunized mice. Overall, DC-LM immunization induced memory CD8 T cells that altered the magnitude of the subsequent cellular infiltrate and enhanced viral clearance following RSV infection.
Since DC-LM immunized mice exhibited decreased viral titers, we next determined if they also experienced reduced disease severity. We evaluated weight loss and airway obstruction, both key disease manifestations that can be assessed following RSV infection in mice [18, 28, 29]. Despite enhanced viral control, M282-immunized mice exhibited a significant (p<0.01) decrease in survival (Fig 2A). Approximately 40% of fatalities were due to M282-immunized mice naturally succumbing to RSV infection, while 60% were euthanized upon reaching a humane weight loss endpoint. This outcome was both unexpected and unusual since an acute RSV infection is rarely fatal in adult BALB/c mice. M282-immunized mice also exhibited significantly (p<0.05) increased weight loss (Fig 2B) and reduced pulmonary function (Fig 2C and 2D). Additionally, we evaluated lungs by histology for evidence of diffuse alveolar damage (DAD), an acute form of lung injury [30, 31]. If severe and extensive enough, DAD is the foundational lesion in the clinical syndrome known as acute respiratory distress syndrome. M282-immunized mice revealed increased (p<0.001) histopathological evidence of characteristics associated with early stages of DAD including cellular sloughing and necrosis, alveolar hemorrhage, early cellular infiltrates, and hyaline membrane formation (Fig 2E and 2F and S2 and S3 Figs).
Previous work has demonstrated that the M282-specific CD8 T cell response contributes to the immunopathology associated with an acute RSV infection [32]. Thus, it was unclear if the increased disease severity observed in M282-immunized mice was unique to the M282-90 epitope. To address this possibility, we evaluated mice immunized against the F85-93 (F85) CD8 T cell epitope following RSV infection [33]. Similar to M282-immunized mice, F85 DC-LM immunization induced a high frequency of RSV F85-specific memory CD8 T cells that mediated a decrease in lung viral titers at day 4 following RSV challenge (S4A and S4B Fig). In addition, F85-immunized mice exhibited increased mortality, weight loss, and pulmonary dysfunction as compared to controls (S4C–S4F Fig). Thus, the severe immunopathology induced by memory CD8 T cells was not specific to either a particular epitope or an RSV protein. We next determined if RSV-specific memory CD8 T cells would also cause increased disease in C57BL/6 mice, as an acute RSV infection in this mouse strain typically causes only mild disease [34]. Therefore, we immunized C57BL/6 mice against the immunodominant M187-195 (M187) CD8 T cell epitope [35]. DC-LM immunization targeting the M187 epitope resulted in approximately 33% M187-specific CD8 T cells in the PBL by day 28 post-boost (S5A Fig). Similar to M82-immunized BALB/c mice, M187-immunized C57BL/6 mice exhibited significantly reduced lung virus titers (p<0.001), decreased pulmonary function, and increased weight loss following RSV infection (S5B–S5E Fig). However, in contrast to M282-immunized BALB/c mice, all of the M187-immunized C57BL/6 mice survived following RSV infection. This data indicates that M187-specific CD8 T cells also contribute to immunopathogenic responses in the C57BL/6 genetic background.
The challenge virus utilized in our study is the RSV A2 strain, which primarily induces a Th1-biased immune response [36]. However, other RSV strains can induce more heterogeneous Th responses. To determine if memory CD8 T cells generated by DC-LM vaccination induce immunopathology independently of the RSV challenge strain, we infected M282-immunized mice with the recombinant RSV A2-line19F strain, which promotes a more Th2-biased immune response than RSV A2 [36]. M282-immunized mice challenged with A2-line19F exhibited mortality, weight loss, and pulmonary dysfunction that was identical to mice challenged with A2 (S6 Fig). These results indicate that M282-specific memory CD8 T cells induce immunopathology independently of the RSV strain utilized for challenge.
To determine if M282-specific memory CD8 T cells would be pathogenic outside of RSV infection, we challenged M282-immunized mice with a recombinant IAV expressing the M282 epitope (IAV-M282). Slutter et al. recently used a similar DC-LM prime-boost strategy to elicit IAV-specific memory CD8 T cells and demonstrated protection against a subsequent IAV infection [25]. Similar to RSV challenge, immunized mice infected with IAV-M282 exhibited a significant increase in the number of total and M282-specific CD8 T cells in the lung by day 5 p.i. (S7 Fig). IAV-M282 infection also resulted in a significant (p<0.05) increase in monocytes and decrease in eosinophils compared to non-infected M282-immunized controls (S7 Fig). CD4 T cell, Treg, neutrophil, and NK cell numbers were not significantly altered in IAV-M282 challenged mice (S7 Fig). As anticipated, all control mice, which do not have virus-specific memory CD8 T cells, succumbed to a lethal IAV-M282 challenge. However, in contrast to the significant (p<0.001) mortality observed in M282-immunized mice following RSV challenge, M282-immunized mice were protected against a lethal IAV-M282 infection (Fig 3A). Pre-existing M282-specific memory CD8 T cells also prevented the prolonged weight loss and pulmonary dysfunction observed in control mice following IAV-M282 infection (Fig 3B–3D). Furthermore, weight loss and pulmonary dysfunction following RSV challenge were significantly (p<0.05) increased at early timepoints compared to IAV-M282 infected mice (Fig 3B–3D). In addition, M282-specific memory CD8 T cells induced a significant (p<0.001) reduction in IAV titers at both day 4 and day 6 p.i. (Fig 3E). These data indicate that pre-existing RSV M282-specific memory CD8 T cells mediate enhanced IAV-M282 clearance and promote increased survival of mice by preventing prolonged disease. To confirm that memory CD8 T cells would also provide protection without extensive immunopathology to a lower viral inoculum, we challenged M282-immunized mice with a sublethal dose of IAV-M282. Similar to a lethal dose of IAV-M282, M282-immunized mice experienced significantly (p<0.05) improved pulmonary function and less weight loss as compared to the control group following a sublethal infection (S8 Fig). Thus, memory CD8 T cells do not promote distinct patterns of disease severity between sublethal and lethal IAV infections. These results indicate that lethal immunopathology associated with a high-magnitude memory CD8 T cell response is unique to the context of an RSV infection.
The DC-LM immunization strategy utilized in our study induced high frequency systemic CD8 T cell memory but resulted in fatal immunopathology following RSV challenge. We hypothesized that a local immunization would promote a large population of antigen-specific resident memory CD8 T cells within the lung that would prevent the immunopathology observed after RSV infection. To evaluate the effects of local immunization, DC-M282-primed mice were either not boosted, given a systemic LM-M282 boost intravenously (i.v.), or given a local IAV-M282 boost intransally (i.n.). Administration of an IAV-M282 boost induced a significant (p<0.001) increase in the total number of M282-specific CD8 T cells in the lung prior to RSV challenge as compared to a systemic LM-M282 boost (S9A and S9B Fig). We next utilized intravascular staining to determine the localization of cells within the lung following immunization [37, 38]. The majority of M282-specific CD8 T cells in the lung of DC-M282-primed mice that were either not boosted or boosted systemically with LM-M282 were located within the pulmonary vasculature (S9C Fig). In contrast, greater than 85% of M282-specific CD8 T cells were localized within the lung tissue in IAV-M282-boosted mice, resulting in a significant (p<0.001) increase in total number as compared to LM-M282-boosted mice (S9C and S9D Fig). In addition, local boost in the lung with IAV-M282 induced a large frequency of RSV-specific CD8 T cells within the lung tissue expressing both CD69 and CD103, which represent the canonical markers of resident memory CD8 T cells. In contrast, systemic LM-M282 immunization failed to elicit resident memory CD8 T cells within the lung tissue (S9E and S9F Fig).
To determine whether the RSV-specific resident memory CD8 T cells generated by local immunization induce less severe immunopathology than their systemically induced counterparts, DC-M282-primed mice that were either not boosted or boosted with either LM-M282 i.v. or IAV-M282 i.n. were challenged with RSV and monitored for morbidity and mortality. In contrast to a systemic LM-M282 boost, a local boost in the lung with IAV-M282 resulted in 100% survival following RSV challenge (S10A Fig). The IAV-M282 boost also resulted in significantly (p<0.05) reduced weight loss and pulmonary dysfunction following RSV infection compared to the LM-M282 boost (S10B–S10D Fig). These results suggest that local prime-boost immunization generates RSV-specific resident memory CD8 T cells that prevent fatal immunopathology and ameliorate disease following RSV challenge.
We next evaluated the primary antiviral effector functions of memory CD8 T cells to determine their contribution to the immunopathology. The primary pathway of CD8 T cell-mediated cytolysis is through the release of perforin and granzymes [39–41]. Therefore, we evaluated survival and disease severity following RSV infection of perforin-deficient M282-immunized mice. Mice deficient in perforin exhibited accelerated mortality as compared to wild-type (WT) controls (S11A Fig). The kinetics of weight loss and pulmonary dysfunction were similar between WT and perforin-deficient M282-immunized mice following RSV challenge (S11B–S11D Fig). Therefore, perforin is not required to mediate exacerbated disease, but may be necessary to prevent additional mortality.
We also evaluated the role of TNF in immunized mice given its previously identified contribution to immunopathology associated with an acute RSV infection [42]. Antibody-mediated neutralization of TNF in the airways at the time of RSV challenge led to survival of all M282-immunized mice (S12A Fig). Neutralization of TNF significantly (p<0.05) reduced both weight loss and pulmonary dysfunction (S12B–S12D Fig). These data illustrate that similar to an acute RSV infection [42], TNF also contributes to the immunopathology associated with memory CD8 T cell responses. Furthermore, TNF neutralization had no significant impact on viral titers at day 4 p.i. (S12E Fig). Assessment of TNF levels at day 2 following RSV challenge revealed similar levels between control- and M282-immunized mice (S12F Fig). TNF levels in the lung were significantly (p<0.01) increased in M282-immunized mice at day 4 p.i. as compared to the control group, but no increase was observed in the serum (S12F Fig). However, the overall amount of TNF in both the lung and the serum had decreased by day 4 p.i., the time when immunized mice began to succumb to the RSV infection. Therefore, these data demonstrate that TNF contributes to general inflammation in the lung during both a primary and recall response to RSV infection, but that TNF is necessary for the lethal immunopathology to occur.
Due to the accelerated memory CD8 T cell response in DC-LM immunized mice, we speculated that the early inflammatory cytokine milieu would be distinct from an acute RSV infection. We initially evaluated IFN-γ levels, as it is a common pro-inflammatory cytokine produced by CD8 T cells following viral infection. IFN-γ protein levels were significantly (p<0.001) increased in both the lung and serum at day 2 p.i. of M282-immunized mice as compared to the control group (Fig 4A and 4B). IFN-γ levels in M282-immunized mice remained elevated above controls at day 4 p.i., but the overall amount was reduced as compared to day 2 following infection. Interestingly, challenge with IAV-M282 resulted in significantly (p<0.001) greater IFN-γ protein levels in the lung at day 2 and day 4 p.i. as compared to challenge with RSV (S13A Fig). In contrast, serum IFN-γ and lung TNF levels in IAV-M282-infected mice were reduced (p<0.001) at day 2 but increased (p<0.001) at day 4 p.i. when compared to the levels observed following RSV challenge (S13B and S13C Fig).
To determine the in vivo source of IFN-γ in RSV-infected immunized mice, we treated mice with brefeldin A (BFA) to capture cells producing IFN-γ via intracellular staining and flow cytometry [43]. Leukocytes producing IFN-γ in vivo were readily identified using this previously established method (S14A Fig). Upon evaluation of the primary leukocyte populations present in the lung following RSV infection, only lymphocytes had produced IFN-γ in immunized mice at day 2 p.i. (Fig 4C). Only a small frequency of CD4 T cells and NK cells secreting IFN-γ were observed at day 2 p.i., whereas approximately 45% of CD8 T cells were producing IFN-γ (Fig 4C). These data also correlate with the rapid and transient increase in the amount of IFN-γ protein we observed in the lung and serum at day 2 p.i., as virtually no IFN-γ-producing cells were recovered on day 5 p.i. (S14B Fig). The IFN-γ secreting CD8 T cells were largely CD11ahi, indicating that the majority were antigen-experienced T cells (Fig 4D) [44]. When comparing M282-specific T lymphocytes to all other CD8 T cells in the lung at day 2 p.i., half of M282-specific CD8 T cells produced IFN-γ, whereas almost 30% of the remaining CD8 T cells were also producing IFN-γ (Fig 4E). These results indicate that both M282-specific and bystander antigen-experienced CD8 T cells secrete IFN-γ in immunized mice early following RSV infection. Lastly, we assessed IFN-γ production by CD8 T cells in the respiratory tract (lung and BAL), mediastinal lymph node (mLN), and periphery (spleen and PBL). The majority of CD8 T cells secreting IFN-γ were localized to the lung and BAL (Fig 4F). In contrast, a relatively low frequency of CD8 T cells produced IFN-γ in the spleen and PBL, and virtually no IFN-γ production was observed in the mLN (Fig 4F). Taken together, these data demonstrate that antigen-experienced CD8 T cells in the airways of immunized mice secrete IFN-γ early following RSV infection leading to increased IFN-γ levels in both the lung and the periphery.
Due to the increased systemic IFN-γ levels largely produced by antigen-experienced CD8 T cells in the respiratory tract, we next determined if IFN-γ was necessary to mediate the severe immunopathology in immunized mice. To evaluate the role of IFN-γ, we treated M282-immunized mice with either control IgG or anti-IFN-γ neutralizing antibody administered i.n at the time of RSV challenge. While high mortality was observed in the IgG-treated group, neutralization of IFN-γ led to the survival of all immunized mice (Fig 5A). In addition, weight loss and respiratory dysfunction were significantly (p<0.05) reduced in immunized mice following IFN-γ neutralization as compared to the IgG-treated mice (Fig 5B–5D). Neutralization of IFN-γ did not significantly impact virus titers in the lung at day 4 p.i., suggesting that IFN-γ does not contribute to pathogen clearance in this prime-boost immunization model (Fig 5E). Neutralization of IFN-γ resulted in significantly (p<0.001) decreased TNF levels in the lung at day 2 p.i. compared to IgG-treated controls (Fig 5F). Overall, our results suggest that antigen-experienced CD8 T cells rapidly secrete IFN-γ, which mediates lethal immunopathology following RSV infection in DC-LM-immunized mice by promoting the production of TNF by other cell populations.
Current RSV vaccine development and assessment is focused upon induction of a strong humoral immune response [45]. In contrast, the capacity of cellular immunity to provide protection against an RSV infection has received less attention. Here we evaluated the capacity of memory CD8 T lymphocytes to protect against an RSV infection. Our results demonstrate that memory CD8 T cells, in the absence of RSV-specific CD4 T cell memory or antibodies, promote enhanced viral clearance following RSV challenge. However, pre-existing RSV-specific memory CD8 T cells also mediate exacerbated disease severity and lethal immunopathology. The CD8 T cell response has been previously shown to contribute to weight loss and illness following an acute RSV infection [12]. Therefore, the CD8 T cell response plays a crucial role in both viral clearance and immunopathology during primary and secondary responses.
A number of previous reports have shown that the adoptive transfer of activated effector RSV-specific CD8 T cells, in vitro stimulated T cell lines, or in vitro propagated T cell clones results in enhanced RSV clearance from the lung following RSV challenge. These effector CD8 T cell transfers were also associated with increased weight loss, indicating that infusion of effector CD8 T cells leads to the induction of increased systemic disease [20–23]. Thus, our studies contrast substantially with these previous studies as we have examined the protective capacity of in vivo generated RSV-specific memory CD8 T cells that have not undergone in vitro activation or restimulation.
Vallbracht et al. reported that mutation of the M282 epitope sequence within the RSV genome results in reduced T cell-mediated immunopathology following an acute infection [32]. However, the enhanced disease severity associated with pre-existing RSV-specific memory CD8 T cells we observe here was not limited to either a specific epitope or RSV protein. DC-LM immunization targeting either the M282 or F85 epitope resulted in exacerbated disease and high mortality following RSV challenge. Furthermore, immunization against the immunodominant M187 epitope in C57BL/6 mice also caused increased weight loss and pulmonary dysfunction, but no mortality following RSV infection. The lack of mortality in the immunized C57BL/6 mice may be due to M187-specific CD8 T lymphocytes having superior cytolytic function with limited immunopathology as compared to M282-specific CD8 T cells [46].
Interestingly, the lethal immunopathology associated with pre-existing memory CD8 T cells was unique to the context of an RSV infection. Memory CD8 T cells are protective and do not exacerbate disease severity with other respiratory viral infections such as IAV or SARS-CoV [25, 26]. Consistent with memory CD8 T cells being able to provide protection against IAV infection [25], we show that RSV M282-specific memory CD8 T cells mediate protection following a lethal IAV-M282 infection. A systemic increase of IFN-γ was seen in immunized mice early 2 days following RSV infection. Antigen-experienced CD8 T cells in the airways were the primary source of IFN-γ. The presence of IFN-γ-producing CD8 T cells only in the airways suggests that production was in response to antigen stimulation given the strong tropism of RSV to the epithelium and alveolar cells of the respiratory tract [47–49]. Interestingly, peak IFN-γ production occurred prior to the significant accumulation and/or expansion of virus-specific CD8 T cells within the lung by day 5 p.i. This observation is similar to work by Liu et al. showing IFN-γ production by CD8 T cells prior to expansion of the CD8 T cell response following lymphocytic choriomeningitis virus (LCMV) infection [43]. Neutralization of IFN-γ prevented all mortality and reduced disease severity in immunized mice without impacting RSV titers. These data indicate that IFN-γ promotes the severe immunopathology, but does not contribute to viral clearance.
A robust IFN-γ response may explain why memory CD8 T cells protect against a lethal IAV-M282 and not an RSV infection. Antibody-mediated neutralization of IFN-γ does not impact RSV viral titers during an acute infection indicating a minor role in viral clearance [42]. Induction of a Th1-biased immune response is important to prevent the pathology associated with a Th2-skewed response [42, 50–52]. Nonetheless, IFN-γ still contributes to minimal immunopathology associated with an acute RSV infection [42, 52]. However, treatment of mice with recombinant IFN-γ at early timepoints has been shown to protect against lethal IAV infection with no difference in viral clearance [53]. This result is consistent with our study showing that DC-LM immunized mice receiving a lethal IAV infection have significantly increased IFN-γ protein levels in the lung at early time points as compared to RSV challenged mice. Together, these results suggest that supplemental IFN-γ limits the severe pathology associated with a lethal IAV infection without contributing to viral clearance. Therefore, IFN-γ can play distinct roles by either contributing to immunopathology or ameliorating disease dependent upon the respiratory virus infection.
Other CD8 T cell functions were not required to mediate the exacerbated disease observed in DC-LM-immunized mice. Immunized mice deficient in perforin exhibited similar disease severity as WT controls following RSV challenge. In addition, all perforin-deficient mice eventually succumbed to infection whereas typically 10–20% of WT immunized mice survive. It has been previously demonstrated that IFN-γ and TNF levels in perforin-deficient mice are elevated following an acute RSV infection, which likely contributes to the accelerated mortality we observed in M282-immunized perforin-deficient mice [42]. A role for regulating CD8 T cell expansion and cytokine production by perforin to prevent mortality has also been demonstrated for secondary CD8 T cell responses against LCMV [54, 55]. Lastly, the neutralization of TNF also improved survival and ameliorated disease in immunized mice following RSV infection. The cytokine TNF is known to contribute to both weight loss and inflammation during acute RSV infection [42]. Therefore, TNF plays a critical role in mediating immunopathology during both primary and secondary RSV infections.
Our results are in contrast to work by Lee et al. showing enhanced RSV clearance with reduced disease severity mediated by vaccine-elicited memory CD8 T cells [56]. The DC-LM immunization utilized in our studies resulted in dramatically increased numbers of RSV-specific memory T cells compared to the immunization strategy employed by Lee et al. [56]. This disparity in RSV-specific memory CD8 T cell numbers prior to RSV challenge may account for the observed difference in outcome between our studies. It remains controversial whether a strong CD8 T cell response is desirable during RSV infection in humans. Experimental RSV infection of adult humans revealed that greater frequencies of virus-specific CD8 T cells in the BAL correlated with reduced clinical disease symptoms [57]. In contrast, a greater ratio of CD8 to CD4 T cells in the airways is associated with acute lung injury during common respiratory tract infections such as RSV [58].
RSV-specific CD8 T cell responses have also been examined in humans following RSV infection. Consistent with their critical role in clearing an acute infection, a study by Weilliver et al. found that children with a fatal primary RSV infection had fewer CD8 T cells in their lung tissue than normal controls suggesting that patients with severe lower respiratory tract illness may have an insufficient cell-mediated immune response [59]. In contrast, a report by Heidema et al. examining CD8 T cells in the airways of infants with a severe primary RSV infection was able to readily identify RSV-specific CD8 T cells in the airways with no significant difference in the number recovered between infants with more versus less severe disease [60]. Thus, it is currently unclear if CD8 T cells contribute to immunopathology during an acute RSV infection in infants.
The role of RSV-specific memory CD8 T cells in humans has been difficult to address due to the difficulty in obtaining CD8 T cells from the airways. A recent report by Jozwik et al. using the human RSV challenge model has sought to address this issue. In adult volunteers experimentally infected with RSV, Jozwik et al. found that a higher baseline frequency of RSV-specific CD8 T cells in the airways correlated with a lower cumulative symptom score following RSV challenge [57]. Our results obtained by administration of a local IAV-M282 boost are consistent with this notion. Our data suggest that systemic RSV-specific memory CD8 T cells are more prone to causing immunopathology, possibly in part due to their delay in reaching the lungs. However, the use of the human RSV challenge model cannot evaluate the protective capacity of memory CD8 T cells in the absence of RSV-specific memory CD4 T cells and antibodies as we have done here using our animal model. Thus, we believe our results are applicable to humans in cases where vaccination of an RSV seronegative individual would primarily elicit a CD8 T cell response.
Our study defines a clear role for memory CD8 T cells following RSV infection. Pre-existing CD8 T cell memory contributes to enhanced viral clearance upon RSV challenge, but also mediates severe immunopathology in contrast to many other viral infections. The outcome is compelling given the high rate of mortality, which is unusual in this RSV infection model. These data highlight how complex and unique the RSV-induced immune response is in contrast to other respiratory viral infections. Our results indicate that epitope-based cellular vaccines against RSV may have detrimental consequences. Our data also support that caution must be exercised during evaluation of any RSV vaccine candidate, particularly when robust memory T cell responses are involved in order to prevent the induction of immunopathology.
Female BALB/cAnNCr mice between 6–8 wk old were purchased from the National Cancer Institute (Frederick, MD). Female H-2d perforin-deficient mice were provided by Dr. John Harty (University of Iowa, Iowa City, IA) [61, 62].
All experimental procedures utilizing mice were approved by the University of Iowa Animal Care and Use Committee under Animal Protocols #4101196 and #7041999. The experiments were performed under strict accordance to the Office of Laboratory Animal Welfare guidelines and the PHS Policy on Humane Care and Use of Laboratory Animals.
Memory CD8 T cells were induced using a DC-LM, prime-boost immunization regimen. BALB/c mice were injected intraperitoneally (i.p.) with 5 x 106 B16 melanoma cells that express fms-related tyrosine kinase 3 ligand (B16-FLT3L). After 14 days, mice were injected i.v. with 1–2 μg lipopolysaccharide (LPS) to mature DCs. 24 hrs later, spleens were harvested and digested in HBSS containing 60 U/mL DNase I (Sigma-Aldrich) and 125 U/mL collagenase (Invitrogen) while gently shaking for 20 mins at 37°C. Spleens were made into single-cell suspensions and incubated with a 2 μM concentration of either M282-90 or F85-93 peptide for 2 hrs at 37°C while rocking. DCs were isolated using anti-CD11c microbeads (Miltenyi Biotec) and sorted via positive selection on an autoMACS separator (Miltenyi Biotec). Mice were primed with 5 x 105 peptide-pulsed DCs. DC-immunized mice were boosted with 5 x 106 actA-deficient LM that express either M282 or F85 administered i.v. 7 days later. 28–42 days following the LM boost, mice were infected with either RSV or IAV-M282. Control mice were primed with DCs incubated without peptide and boosted with an actA-deficient LM that does not express any RSV-derived epitopes [63]. The recombinant LM strains were created using pPL2 integration vector [64]. Target DNA was inserted at digested BamH1 and PstI sites and ligated in Escherichia coli XL1-Blue cells. Recombinant chloramphenicol-resistant plasmids were conjugated in E. coli SM10 cells along with the 10403S strain of LM that is resistant to streptomycin [65] on brain heart infusion agar plates. Growth from previous step were streaked out on selective brain heart infusion agar plates to select chloramphenicol- and streptomycin-resistant colonies that contain pPL2 integrated into the 10403S LM strain. Recombinant LM were grown in tryptic soy broth (35.6g/L) containing 50 mg/mL streptomycin.
The A2 strain of RSV was a gift from Dr. Barney Graham (National Institutes of Health, Bethesda, MD). The A2-line19F strain was a gift from Dr. Martin Moore (Emory University, Atlanta, GA). RSV strains were propagated in HEp-2 cells (ATCC). Mice were infected i.n. with 1.0–1.7 x 106 PFU of purified RSV. For RSV purification, 50% polyethylene glycol was added to crude RSV for a final dilution of 1:5. The RSV preparation was mixed at 4°C for 2 hrs and centrifuged at 7300 g for 30 mins in a swing bucket rotor. Pellets were resuspended in 20% sucrose solution and placed on top of 60% and 35% sucrose layers and centrifuged at 170,000 g for 1 hr. Purified RSV at the interface between the 35% and 60% sucrose layers was collected and stored at -80°C. All solutions were created in a buffer containing 0.15 M NaCl, 0.05M Tris-HCl, and 0.001M EDTA. For mock infections, mice were administered an equivalent volume of sterile PBS. Recombinant IAV-M282 was kindly provided by Dr. Ryan Langlois (University of Minnesota, Minneapolis, MN). The virus was created using standard reverse genetics as previously described [66], rescued, and grown in 10 day-old embryonated chicken eggs (Charles River). M282 epitope was inserted into the mRNA nucleotide position 186 encoding the neuraminidase stalk region, which is known to tolerate such insertions [67]. For lethal heterologous IAV infections, mice were challenged i.n. with a 5 LD50 dose representing 1 x 103 PFU of recombinant IAV-M282 virus. For sublethal IAV infections, mice were infected i.n. with a 0.1 LD50 dose representing 20 PFU of IAV-M282. In certain experiments, mice were boosted with IAV-M282 and given a 0.1 LD50 dose i.n. 7 days following the DC-M282 prime i.v.
Whole lungs were harvested from mice, weighed, mechanically homogenized, and supernatant was stored at -80°C until further use. 1:10 serial dilutions of supernatants were performed and incubated on Vero cells (ATCC) in 6-well plates for 90 mins at 37°C. Plates were overlaid with a 1:1 mixture of 2X Eagle minimum essential medium (2X EMEM, Lonza, Walkersville, MD) and 1% SeaKem ME agarose (Cambrex, North Brunswick, NJ). Following 5 days of incubation at 37°C, 5% CO2, plates were stained with a 1:1 mixture of 2X EMEM and 1% agarose containing 0.015% neutral red (Sigma-Aldrich). Plaques were counted after 24–48 hrs. For determination of IAV titers, lungs were processed in the same manner as for RSV plaque assay. MDCK (ATCC) cells in 6-well plates were washed 3 times with room temperature sterile PBS adding 1 mL of sterile Dulbecco’s modified Eagle’s medium afterwards. Plates were infected with 100 μl of serially diluted IAV-infected lung samples (10-fold dilutions) for 1 hr at 37°C. Plates were washed twice with sterile room temperature PBS. Wells were overlaid with 2 mL of a 1:1 mixture of 2X EMEM and 1.6% agarose containing 1 mg/mL TPCK-trypsin and incubated at 37°C, 5% CO2 for 3 days. Agarose plugs were carefully removed, and monolayers were fixed with 2 mL 70% ethanol for 20 mins at room temperature. Monolayers were stained with 1 mL of 1% crystal violet in methanol for 10 mins at room temperature, and plates were washed in a pool of warm water. Plates were allowed to dry overnight, and plaques were counted the next morning.
Pulmonary function of mice was evaluated using unrestrained whole-body plethysmography. Enhanced pause (Penh) and respiratory minute volume (MVb) were measured using a whole-body plethysmograph (Buxco Electronics, Wilmington, NC) and averaged over a 5 min period. Weight loss was tracked daily following RSV or IAV infection of mice. Mice that were at or below 70% of their starting weight were euthanized.
For IFN-γ and TNF neutralization, mice were treated i.n. with 200 μg of anti-IFN-γ (clone XMG1.2) or anti-TNF (clone MP6-XT22) antibody during RSV challenge. For controls, mice were administered a matching dose of control isotype IgG antibody.
Whole lungs were harvested on day 5 following RSV challenge and fixed in 10% neutral buffered formalin (Fisher Scientific). Lungs were processed as previously described [68] and stained with H&E for routine evaluation. Representative images of lung sections were taken at 20X, 200X, and 400X magnification for each immunization regimen. Tissues were examined and scored in a manner masked to experiment groups [69]. Each sample was assessed for evidence of DAD. Histopathologically, early stages of DAD include alveolar septal injury, such as cellular sloughing, necrosis, hyaline membrane formation, hemorrhage, and early cellular infiltrates. DAD scores were assigned as follows: 1—absence of cellular sloughing and necrosis; 2—Uncommon solitary cell sloughing and necrosis; 3—Multifocal cellular sloughing and necrosis with uncommon septal wall hyalinization; 4—Multifocal cellular sloughing and necrosis with common and prominent hyaline membranes.
Serum was collected and whole lungs were harvested on days 0, 2, and 4 p.i. Lungs were disrupted using a tissue homogenizer (Ultra-Turrax T25; IKA Works, Inc., Wilmington, NC) in Cell Lysis Buffer (eBioscience). Lung homogenates were centrifuged at 2000 rpm for 10 mins, and supernatants were collected. The protein levels of 20 different cytokines and chemokines in the lung and serum were determined using a ProcartaPlex Multiplex Immunoassay kit (eBioscience) according to the manufacturer’s instructions. The assay was run on a BioPlex instrument (Bio-Rad, Hercules, CA). Lung and serum IFN-γ levels were determined by ELISA as previously described (eBioscience) [68]. Lung TNF levels were determined using a mouse TNF ELISA kit (Invitrogen) according to manufacturer’s instructions.
Lung and BAL were harvested from mice as previously described [70, 71]. Spleens and mLN were gently dissociated between the frosted ends of microscope slides. Cells from the lung, BAL, spleen, mLN, and PBL were stained for extracellular surface molecules with antibodies specific to CD11c (clone N418), Siglec F (BD Biosciences, clone E50-2440), F4/80 (clone BM8), Ly6c (clone HK1.4), Ly6g (clone 1A8), CD49b (clone DX5), NKp46 (clone 29A1.4), CD11a (clone M17/4), CD90.2 (clone 53–2.1), CD3ε (clone 145-2C11), CD4 (clone GK1.5), CD8 (clone 53–6.7), CD69 (clone H1.2F3), and CD103 (clone 2E7) for 30 mins at 4°C and fixed with fix/lyse solution (eBioscience) for 10 mins at room temperature. After extracellular staining, cells were stained for FoxP3 (eBioscience clone FJK-16s) with transcription factor staining buffer set (eBioscience) according to manufacturer’s instructions. For intracellular cytokine staining, cells were stimulated for 5 hrs at 37°C with 2 μM M282-90 peptide in 10% FCS-supplemented RPMI. Stimulated cells were stained for surface markers as indicated above and then stained intracellularly with antibodies specific to IFN-γ (clone XMG1.2) and TNF (clone MP6-XT22) in FACS buffer containing 0.5% saponin (Sigma-Aldrich) for 30 mins at 4°C. Total numbers of cytokine producing cells were calculated after subtraction of background staining from BFA-only controls. All monoclonal antibodies were purchased from BioLegend unless otherwise stated. Stained cells were run on LSRFortessa and analyzed with FlowJo (Tree Star, Ashland, OR) software. Cell types were phenotyped as follows: CD8 T cells (CD90.2+CD8+), CD4 T cells (CD90.2+CD4+), Tregs (CD90.2+CD4+FoxP3+), NK cells (CD3ε-CD49b+NKp46+), monocytes (CD11c+F4/80+), eosinophils (SiglecF+CD11clo), and neutrophils (Siglec F-CD11c-Ly6c+Ly6g+).
Mice were injected i.v. with 1 μg CD45-FITC (CD45 labeled with fluorescein isothiocyanate) (clone 30-F11) antibody 3 mins prior to euthanasia. Cells from the lung were processed as previously described [37].
Analysis of IFN-γ-producing cells was performed using in vivo BFA Administration [43]. Mice were injected i.v. with 250 μg BFA (0.5 mg/mL; Sigma) in 500 μl PBS, and lungs, BAL, spleen, mLN, and PBL were harvested 6 hrs later. Leukocytes were stained as indicated above.
All statistical analyses are described in each figure legend and were performed using Prism software (GraphPad Software, San Diego, CA). Data were evaluated using unpaired, two-tailed Student’s t tests between two groups or one-way ANOVA with Tukey-Kramer post-test analyses for more than two groups to determine if there was a statistical significance of at least α = 0.05. Asterisks or pound signs are used to define a difference of statistical significance between the indicated group and its respective control group unless otherwise indicated by a line or stated in the figure legend.
|
10.1371/journal.pntd.0002586 | Impact of Six Rounds of Mass Drug Administration on Brugian Filariasis and Soil-Transmitted Helminth Infections in Eastern Indonesia | The lymphatic filarial parasite Brugia timori occurs only in eastern Indonesia where it causes high morbidity. The absence of an animal reservoir, the inefficient transmission by Anopheles mosquitoes and the high sensitivity to DEC/albendazole treatment make this species a prime candidate for elimination by mass drug administration (MDA).
We evaluated the effect of MDA using DEC and albendazole on B. timori and soil transmitted helminths (STH) in a cross-sectional study of a sentinel village on Alor Island annually over a period of 10 years. Pre-MDA the microfilaria (MF) prevalence was 26% and 80% of the residents had filaria-specific IgG4 antibodies. In 2010, 34 months after the 6th round of MDA, MF and antibody rates were only 0.17% and 6.4%, respectively. The MDA campaign had also a beneficial effect on STH. Baseline prevalence rates for Ascaris, hookworm and Trichuris were 34%, 28%, and 11%, respectively; these rates were reduced to 27%, 4%, and 2% one year after the 5th round of MDA. Unfortunately, STH rates rebounded 34 months after cessation of MDA and approached pre-MDA rates. However, the intensity of STH infection in 2009 was still reduced, and no heavy infections were detected.
MDA with DEC/albendazole has had a major impact on B. timori MF and IgG4 antibody rates, providing a proof of principle that elimination is feasible. We also documented the value of annual DEC/albendazole as a mass de-worming intervention and the importance of continuing some form of STH control after cessation of MDA for filariasis.
| The impact of six annual rounds of mass drug administration (MDA) using DEC combined with albendazole on brugian filariasis and soil-transmitted helminths (STH) was evaluated. Microfilaria rates of B. timori dropped quickly after MDA and were below 1% for 34 months after stopping intervention when the study ended. The prevalence of filarial-specific IgG4 antibodies in the community as measured by the Brugia Rapid test was about 80% before MDA and dropped slowly to about 6% at the end of the study period. MDA had also a beneficial effect on STH, especially on hookworm, but prevalence rates rebounded 34 months after cessation of MDA and quickly approached pre-control levels, while infection intensity was still reduced. Our study indicated that B. timori infection can be eliminated by DEC/albendazole MDA and that the anti-filarial IgG4 rate in the community significantly declines over time. While lymphatic filariasis (LF) MDA can be considered as a powerful general deworming campaign, STH infection rates rebound quickly and a supplementary control strategy is needed after LF MDA is ceased.
| Lymphatic filariasis (LF) has been targeted by the World Health Organization for global elimination by the year 2020 [1]. During the years 2000 to 2009 the Global Program to Eliminate Lymphatic Filariasis (GPELF) has provided >2.8 billion treatments with anti-filarial drugs to a minimum of 885 million individuals living in 53 endemic countries [2], [3]. The recommended oral regimen for use in Asia is annual mass drug administration (MDA) with diethylcarbamazine (DEC, 6 mg/kg body weight) combined with albendazole (alb, fixed dose of 400 mg) [1]. We have previously published a preliminary report on the impact of two annual rounds of MDA on brugian filariasis in Alor Island in Eastern Indonesia [4]. Other studies have shown that BmR1 rapid antibody test [5], [6] is a sensitive marker for detecting brugian filariasis in populations. However, more data are required to validate antibody testing as a tool for monitoring the impact of MDA on filariasis in populations.
Studies on bancroftian filariasis in Egypt concluded that five rounds of MDA may have been sufficient to eliminate the infection in most implementation units in that country [7]. Declines in rates of infection markers such as circulating filarial antigenemia and microfilaremia were accompanied by diminished rates of anti-filarial IgG4 antibodies in school children. Similar findings were reported after three rounds of MDA in Papua New Guinea, where antibody prevalence decreased faster in children than in adults [8]. Anti-filarial antibodies are a marker for past, present or exposure to filarial infections. Antibody clearance tends to occur faster after treatment in children than in adults, since children tend to have shorter term exposure and lighter infections. Therefore, adults may have persistent antibodies years after effective therapy [9], [10]. Less is known about antibody clearance after treatment of brugian filariasis, and research is needed to determine the dynamics of this clearance with different antibody assays.
Mass drug administration of DEC combined with alb has an additional beneficial effect in reducing prevalence and intensity of infection with intestinal helminths such as Ascaris lumbricoides, hookworms and Trichuris trichiura [4], [11], [12]. Unfortunately, most studies have evaluated the effect of MDA on intestinal helminths after only one or two treatment rounds in school-aged children, and data on the impact of multiple rounds after population-based MDA used in filariasis elimination programs are lacking.
Pilot studies in 2001 detected a high prevalence of B. timori infection and filariasis-associated morbidity in the highlands of Alor island [13]. We initiated treatment trials and worked with local health officials to develop an MDA program on the island. We reported the results of the first two rounds of MDA in prior publications [4], [14], [15]. The objective of the present paper is to evaluate the impact of six annual rounds of MDA on brugian filariasis and on soil-transmitted helminths (STH) infections in a sentinel village on Alor and also report the results collected over 3 years following the last round of MDA.
The study was performed in Mainang village (population in 2002 approximately 1,500) on Alor island (East Nusa Tenggara Timor, Indonesia). Details of the study site have been published elsewhere [4], [13]. Conditions in Alor and in Mainang changed over the course of the study. For example, the island received considerable financial support following a major earthquake in 2004 which improved infrastructure and living conditions. Bed net use has increased and the hygienic conditions have improved during the study period. However, Alor district remains one of the poorest districts of Indonesia.
Over the 10 year study period (Fig. 1) the study population of the three study sectors of the village increased from about 1,500 to about 1,800. Annual surveys collected samples from 600–750 residents, which comprised 33%–50% of the eligible population. Children younger than 3 years and severely ill persons were considered not eligible and excluded from the surveys. Almost all residents were examined at least once over the study period, while most individuals were examined twice or three times. However, only about 20% of the population participated in all 10 surveys. The study population and the sample collection procedure were described in detail in earlier reports [4], [13]. Briefly, sex, age and name were noted; after a brief clinical examination 3 ml venous blood was collected between 7.00 p.m. and 11 p.m. One labeled stool container was provided to each individual participating in the blood collection, and these were collected one day later by local health workers.
The study was approved by the ethical board of the Faculty of Medicine, University of Indonesia, Jakarta and the institutional review boards at Bernhard Nocht Institute for Tropical Medicine in Hamburg, Germany, as well as at Washington University School of Medicine in St. Louis, USA. Since written consent is not consistent with cultural norms on Alor island, oral informed consent was obtained from all adults or, in case of children, from their parents. The ethical board of the University of Indonesia and the institutional review boards in Germany and the USA approved the use of oral consent. The participant's oral consent was noted on the survey questionnaire.
Community-based MDA using a single dose of DEC and alb was performed by the local District Health Authority. The local primary health care center (“Puskesmas”) trained a number of villagers (cadres) that were responsible for distributing anthemintics to 100 to 200 residents in their neighbourhood by directly observed treatment. Pregnant or breastfeeding women, children younger than two years of age, or persons suffering from acute or severe illnesses were not eligible for treatment. In most cases cadres knew which community members were ineligible (pregnant or lactating women), but they used a short interview to assess eligibility for others. Medications were dosed based on age rather than on weight, as previously described [4]. The District Health Authority reported coverage rates (number of distributed doses per number of residents) between 75% and 85% for all years of MDA (Paul Manoempil pers. commun.). In addition, an independent survey of compliance in five villages on Alor in the first year of the MDA program documented a compliance rate of about 88% [16]. Compliance with MDA was defined as the percentage of the eligible population that reported that they had swallowed the drugs. MDA was discontinued following the treatment round in 2007 (Fig. 1), because MF rates had fallen below 1% in the previous 2 years and because no funds were available at that time to continue MDA or to perform a more extensive assessment according to the WHO guidelines at that time [17].
Microfilaraemia was determined by membrane filtration of 1 ml of night blood collected in EDTA as previously described [4].
A single stool sample was examined for intestinal helminths from each individual surveyed. The Harada-Mori hatching test was used to detect living hookworm larvae as described previously [4]. Based on larval morphology and PCR, the predominant hookworm species in Alor was found to be Necator americanus (Supali and co-workers, unpublished results). Part of the stool samples was preserved in the field using 4% formaldehyde and examined in the laboratory in Jakarta using the formalin/ether-enrichment method within 2 months of collection.
The Kato-Katz method was used to assess STH infections in the baseline survey in 2002 and again in 2009, 22 months after the 6th (and final) round of MDA (Fig. 1). A single Kato Katz smear of 41 mg of stool was performed to determine STH egg counts and to classify egg densities according to WHO guidelines as light, moderate and heavy intensity infection [18]. Stool samples were examined (Kato Katz, Harada-Mori) or preserved (formalin/ether-enrichment) within 18 h of distribution of stool containers. For tests other than the Kato-Katz, the numbers of helminth eggs or hookworm larvae were arbitrarily scored as follows: low density (1–10 eggs per slide or 1–50 hookworm larvae per plastic bag containing approximately 0.5 g stool), moderate density (11–100 eggs or 51–500 larvae) or high density (more than 100 eggs or 500 larvae).
Specific IgG4 antibodies were detected using the dipstick version of the Brugia Rapid test (Malaysian Bio-Diagnostics Research Sdn Bhd, Bangi, Malaysia). Tests were performed in the field using 25 µl of plasma as previously described [19].
The total number of subjects included in the surveys was calculated using a 95% confidence level and a confidence interval of less than 3%. The chi-square test was used to assess the significance of differences in infection prevalence rates between different groups or between study periods, while the Mann-Whitney U test was used to assess infection intensity data. A p value is regarded as significant if less than 0.05. All statistical analyses were done using the SPSS software package (IBM).
Changes in microfilaremia prevalence are shown in Figure 2A. MF rates fell from a baseline of 26% to less than 1% after round 4, and this decrease was maintained for the duration of the study including the period following cessation of MDA. Only 5 MF-positive individuals were detected among 737 individuals tested one year after the 6th round of MDA. Three females (18, 19 and 47 years old) and two males (19 and 26 years old) had microfilaria densities of 1356, 63, 6, 66, and 311 MF/ml respectively. The two men were long-term visitors from another village who reported that they had never received treatment. The three females had participated in the pre-treatment blood collection, and their MF counts at that time were 88, 2456 and 333MF/ml, respectively. Two of the women had not regularly participated in the surveys, but one was examined annually and was found to be MF-positive during all examinations. In 2009 and 2010, 22 and 34 months after the 6th and last round of MDA, we examined a total of 668 and 598 persons, respectively. Only one woman was found to be MF-positive with 192 MF/ml and 115 MF/ml, respectively. It was the same young lady who had been found to be MF-positive all the years before. At first she claimed to have regularly participated in MDA, but later she admitted that she had not taken the medicines after the first treatment round because she had experienced adverse events that included high fever and severe headache.
We used the Brugia Rapid dipstick test as a marker for ongoing or prior infection, because there is no adult worm antigen detection test for Brugia spp. This test employs a recombinant B. malayi antigen (BmRI) that has been extensively studied [19]. Antibody test results are shown in Figure 2B. Briefly, the antibody rate decreased from 80% at baseline to a rate of 6.4% in 2010. There was no significant difference between antibody prevalence rates in male and female subjects all through the study. As reported previously, antibody rates were high (about 80%) in all age groups in 2001 [6]. In 2010 the antibody rates had strongly declined. The rates varied with age (4.9% for subjects less than 21 years of age versus 8.7% for older subjects) (Fig. 3). However, this difference between the two age groups was not significant. WHO now recommends transmission assessment surveys (TAS) to support decisions to stop MDA. The TAS are designed to systematically sample school-aged children within an evaluation unit [20]. In the absence of a rapid antigen test for Brugia, antibody rates in school-aged children are used as the indicator for TAS. Therefore, we analysed the decline of the filarial antibody rate as determined by Brugia Rapid in pre-school and primary school-aged children (Table 1). While the positive test rate for Brugia Rapid in children aged between 3 and 10 years dropped rapidly following MDA, the prevalence and the upper 95% confidence limit dropped below 5% and 10%, respectively, for the first time in 2009, 22 months after the last round of MDA. In 2009 the majority of children aged between 3 and 10 years were born after the initiation of MDA, but many of these children might have been exposed to B. timori during the early years of the MDA program.
The relationship between antibody and MF rates varied considerably during the study period (compare Fig. 2A and B). The antibody to MF ratio was 3.1 at baseline, but increased dramatically after MDA. This means MF rates fell more rapidly after MDA than antibody rates. For this reason, it is not possible to use antibody rates to estimate MF prevalence rates following MDA.
Ascaris prevalence rates varied considerably during the study period (Fig. 4A). The lowest prevalence of 18% was detected in 2006 after 4 rounds of MDA. In 2010, 34 months after the last round of MDA, the prevalence (45%) was significantly higher than the baseline prevalence (34%) prior to MDA (p<0.01).
In contrast, the hookworm prevalence rates decreased from 28% at baseline to 13% after the first round of MDA and continued to decrease to 0.7% in 2006 (Fig. 4B). The hookworm prevalence rates were determined all through the study period by two independent methods, formalin/ether enrichment and the Harada Mori hatching test, which increases the overall sensitivity. Hookworm rates rebounded to 7% at 34 months after the last MDA.
The first two rounds of MDA had little impact on the Trichuris prevalence rate (baseline 9.4%), but this decreased dramatically to less than 1% after the third round (Fig. 4C). The lowest rate (0.7%) was observed after the last round of MDA in 2006; this slowly increased after MDA ended to 7% in 2010. Interestingly, STH infection rates were similar for male and female subjects (p>0.05%), and there were no significant differences in age specific infection rates in 2002 prior to MDA as well as in 2010 (p>0.5%).
Infection intensities (by Kato-Katz) were assessed before MDA in 2002 and in 2009, 22 months after the 6th round of MDA. As an aside, infection intensities determined by Kato Katz were generally consistent with informally scored infection intensities obtained by the ether enrichment and Harada Mori methods.
MDA has shifted the frequency distributions of STH intensities so that fewer subjects had heavy infections in 2009 than in 2002, and these data are summarized in Figure 5. For example, infected subjects in 5 of the 6 age groups had geometric mean of Ascaris infection intensities in the moderately high range (between 5,000 and 49,000 eggs per gram [epg] according to WHO criteria) in 2002, and the geometric mean egg count in all infected subjects was 15,000 epg. In 2009, three age groups had geometric mean egg counts in the moderately high infection intensity range; the geometric mean count for all infected subjects was 6,500 epg, which represented a 57% reduction from the baseline value (Fig. 5A).
Similar results were obtained for intensities of hookworm infection (Fig. 5B). Adults aged 21 to 50 years with hookworm infections had geometric mean egg counts in the high range (>4,000 epg) in 2002, while children and adolescents had geometric mean counts in the moderately high range (2,000–3,999 epg), while other age groups had moderate mean infection intensities. In 2009, no age group had geometric mean epg in the high range, while moderate range geometric mean counts were only seen in children less than 10 years of age and in adults in the 41–50 years age group. Geometric mean egg counts for T. trichiura in 2009 were not significantly different from those in 2002, and they tended to be low in all age groups at both time points (range 48–164 epg). However, only relatively few Trichuris infections were assessed in 2009 quantitatively.
We have studied the dynamics of helminth infection during and for up to 34 months after six rounds of annual MDA using DEC combined with albendazole in one sentinel village in eastern Indonesia. The MF rate fell to less than 1% in 2006, and this declined further to 0.17% in 2010 when the study ended. MDA rapidly reduced MF counts in the population. We would have detected fewer MF carriers if we had used thick blood smears for MF detection according to WHO guidelines instead of the more sensitive membrane filtration method [21]. In 2009 and 2010 only a single MF positive individual was detected. This individual is an example for systematic non-compliance with treatment because of the fear of adverse events. The contribution of a single MF positive person to the transmissions cycle appeared to be negligible because no new MF positive individuals were found. Anti-filarial antibody rates in the general population fell from 80% in 2001 to 6.4% in 2010. The antibody rate in children and young adults less than 21 years of age in 2010 was 4.9%. These results suggest that MDA interrupted transmission of B. timori infection in Mainang village.
Since the MF rate in 2006 (after the 4th round of MDA) was below 1% and remained below this threshold for the next 2 years, the District Health Authority decided to stop MDA in 2008 due to financial constraints. When external funding became available, a TAS was performed in the entire district (Alor and Pantar islands) in 2009 and in 2011 according to WHO guidelines [17], [20]. The district passed the TAS in both years, and this suggests that the decision to stop MDA in 2008 was valid (Supali and co-workers, unpublished results).
Analysis of anti-filarial IgG4 antibody rates over the 10 year study period of the study showed that this rate has declined more slowly than the MF rate following MDA. Current guidelines recommend testing for antibody in children as a tool for detecting ongoing transmission of brugian filariasis [10], [21]. Our study showed that IgG4 antibody rates as determined by Brugia Rapid declined in children as well as in adults and eventually reached very low levels some nine years after initiation of MDA and after six annual rounds of MDA. The fact that a few children aged ≤10 years in 2010 had positive antibody tests can be explained by the high antibody rate in young children in 2001 and the fact that it takes several rounds of MDA to interrupt transmission of B. timori [6]. The observation that it took 6 rounds of MDA and additional 22 months until the filarial antibody rate as determined by the Brugia Rapid test dropped below 5% (upper 95% confidence limit 10%) in children aged between 3 and 10 years suggests that the antibody rate is a sensitive marker for previous exposure to B. timori. While all examined children in that age group were constantly MF negative following the first two rounds of MDA, the antibody prevalence dropped quickly but remained higher than 2% for the entire study period. A target of 2% antigen prevalence is used for TAS in W. bancrofti areas [20]. It is possible that a higher threshold of antibody prevalence is needed for TAS in Brugia areas. However, Alor district as evaluation unitis endemic for B. timori and W. bancrofti and passed the TAS in 2009 and in 2011 (one year after conclusion of the present study) using antibody and antigen tests (Supali et al, unpublished results) despite the fact that one primary school in Mainang village was included in the surveys,
In order to use antibody rates in the general population as a marker for evaluation of an MDA program, frequent follow-ups are necessary to assess the decline of antibody prevalence over time. It is likely that antibody rates would increase if transmission resumes in the area, but this can only be assessed if longitudinal antibody data are available for the study area. Moreover, the Brugia Rapid test has already been employed to assess the success Brugia malayi elimination from parts of the Republic of Korea [22].
The benefits of MDA using DEC combined with albendazole beyond LF elimination have been extensively discussed previously [23], [24]. However, field data on the effects of MDA on STH are limited since most programs in Africa use ivermectin instead of DEC, and results are often available only for one or two rounds of MDA. In addition results are often not as clear as expected. For example in an area with low STH prevalence and relatively low compliance to MDA in Sri Lanka, it was concluded that four rounds of MDA had little effect on STH infections in school children [25]. In India only a single round of MDA showed reduction of STH infections in school children at several time points following treatment [26]. Prior studies have shown that individuals become rapidly re-infected with STH and that re-infection rates for Ascaris are the highest [27]. Since rapid re-infections are a major problem for STH, evaluation of the effect of MDA on STH should be done probably more frequently or closer to the next treatment date. Modeling studies have predicted that LF elimination can be achieved more rapidly by employing twice yearly MDA, while reducing the overall program costs [28]. Twice yearly MDA could help to prevent re-infection of STH and should, therefore, show a stronger benefit for STH reduction. On the other hand the present study also shows an effect of MDA on STH using annual follow-ups.
Unfortunately, when MDA for LF was stopped in 2008, the district government had no school- or community-based STH deworming program in place to help preserve the beneficial effect that MDA had had on STH. While it was relatively easy for the district government to obtain external funds for LF elimination (since that was an attractive new public health program), it was not possible for the district to obtain support for routine de-worming. At the time the discussion about integration of control and elimination programs for the different NTDs was just starting.
Taken together our study has provided evidence that Anopheles-transmitted B. timori can be eliminated by MDA alone, and that antibody prevalence assessed by Brugia Rapid in the population decreases after MDA but at a slower pace than MF prevalence. Our data also show that MDA for LF with DEC plus albendazole is a mass de-worming program that reduces prevalence rates for hookworm and Trichuris and also reduces STH infection intensities. However, the study also showed STH rates rebounded quickly after MDA was discontinued. Further work is needed to develop strategies and guidelines for controlling STH in communities following suspension of MDA for LF.
|
10.1371/journal.ppat.1002372 | Galactosaminogalactan, a New Immunosuppressive Polysaccharide of Aspergillus fumigatus | A new polysaccharide secreted by the human opportunistic fungal pathogen Aspergillus fumigatus has been characterized. Carbohydrate analysis using specific chemical degradations, mass spectrometry, 1H and 13C nuclear magnetic resonance showed that this polysaccharide is a linear heterogeneous galactosaminogalactan composed of α1-4 linked galactose and α1-4 linked N-acetylgalactosamine residues where both monosacharides are randomly distributed and where the percentage of galactose per chain varied from 15 to 60%. This polysaccharide is antigenic and is recognized by a majority of the human population irrespectively of the occurrence of an Aspergillus infection. GalNAc oligosaccharides are an essential epitope of the galactosaminogalactan that explains the universal antibody reaction due to cross reactivity with other antigenic molecules containing GalNAc stretches such as the N-glycans of Campylobacter jejuni. The galactosaminogalactan has no protective effect during Aspergillus infections. Most importantly, the polysaccharide promotes fungal development in immunocompetent mice due to its immunosuppressive activity associated with disminished neutrophil infiltrates.
| Aspergillus fumigatus is an opportunistic human fungal pathogen that causes a wide range of diseases including allergic reactions and local or systemic infections such as invasive pulmonary aspergillosis that has emerged in the recent years as a leading cause of infection related mortality among immunocompromised patients. Polysaccharides from the fungal cell wall play essential biological functions in the fungal cell biology and in host-pathogen interactions. Indeed, it has been shown that polysaccharides can modulate the human immune response; some of them (β-glucan and α-glucans) having a protective effect against Aspergillus infection. We report here the purification and chemical characterization of a new antigenic polysaccharide (galactosaminogalactan) produced by A. fumigatus. This polymer is secreted during infection. In murine models of aspergillosis, this galactosaminogalactan is not protective but it is immunosuppressive and favors A. fumigatus infection. Particularly it induces the apoptotic death of neutrophils that are the phagocytes playing an essential role in the killing of fungal pathogens.
| Aspergillus fumigatus is an opportunistic human fungal pathogen that causes a wide range of diseases including allergic reactions and local or systemic infections such as invasive pulmonary aspergillosis (IA) that has emerged in recent years as a leading cause of infection-related mortality among immunocompromised patients [1], [2]. The innate immune system provides the first line of defense against A. fumigatus with macrophages and neutrophils that sense, phagocytose and kill conidia and hyphae through the production of anti-microbial agents. Later, antigen presenting cells initiate an adaptative response activating various populations of T-helper cells that impact differently on the evolution of the disease [3], [4]. Because of its external localisation, and specific composition, the cell wall represents a specific target for recognition and specific interaction with the host immune cells. The cell wall of A. fumigatus is mainly composed of branched β1-3glucans, α1-3glucans, chitin, β1-3/1-4 glucan and galactomannan [5]. These constitutive polysaccharides have been shown to induce specific immune responses from the host. For example in murine models of aspergillosis, α1-3glucan and β1-3glucan chains induce a protective response through the activation of Th1 and Th17 or Treg responses [4] whereas galactomannan favours the disease through the activation of the Th2/Th17 response. In other medically important fungi, capsular and cell wall polysaccharides and especially mannan and β-glucans also induce an immune response that either favours or inhibits fungal infection [6], [7], [8], [9].
During growth in vitro in aerial conditions or in vivo in the lung tissues, the mycelium of A. fumigatus is covered by a polysaccharide-rich extracellular matrix (ECM) that because of its outer position, plays a major role in the interaction with the host immune cells [10], [11]. The ECM contains α1-3glucan and galactomannan that are two of the major cell wall polysaccharides, recognised by T cells. A third galactosamine-rich polysaccharide has been now identified in the ECM. Although the presence of such cell wall associated polysaccharide was noticed 20 years ago [12], [13], its structural analysis has not been investigated to date. The present report shows that this polysaccharide is a linear heterogenous chain constituted by α1-4 linked galactose and α1-4 linked N-acetylgalactosamine residues. Most interestingly, the analysis of the immune response towards this polysaccharide shows that it is immunosuppressive and favors A. fumigatus infection.
The culture filtrate of A. fumigatus was precipitated by 70% ethanol. In our experimental conditions, an amount of 80 mg of ethanol precipitate was recovered per g of mycelial dry weight. The incubation of the ethanol precipitate of the culture filtrate of A. fumigatus for 1 h in a 150 mM NaCl aqueous solution resulted in the solubilisation of glycoproteins and galactomannan. The NaCl-insoluble material represented 43+/−8% of the ethanol precipitate. The remaining insoluble material was separated in two fractions based on their solubility in 8 M urea. The urea-soluble material (SGG, urea soluble galactosaminogalactan) accounted for 30+/− 4% of the total ethanol precipitate whereas the urea-insoluble material (PGG, urea insoluble galactosaminogalactan) represented 13+/− 6% of the total ethanol precipitate. Gas liquid chromatography (GC) analysis of both fractions showed that they were exclusively composed of galactosamine and galactose with ratios of 60/40 and 15/85 in SGG and PGG respectively. Nitrous deamination of native polysaccharide did not solubilise the polysaccharide and did not produce anhydrotalose showing that all galactosamine residues were N-acetylated (not shown). GC analysis showed that the galactosaminogalactan was absent in resting conidia but was present in the cell wall of mycelium from both solid and liquid cultures and in different media (not shown). Immunofluorescence with specific anti-GG mAb confirmed that GG was not present on the surface of resting conidia. In contrast, a positive detection was seen in the cell wall as soon as the conidium germinates (Fig. 1). This result indicated that part of the galactosaminogalactan was not secreted and remained strongly associated with the cell wall. The amount of cell wall bound galactosaminogalactan was equivalent to the amount recovered in the culture medium (data not shown).
GC analysis of permethylated GG revealed only two methyl ethers: 2,3,6-tri-O-methyl-galactitol and 3,6-di-O-methyl-N-acetylgalactosaminitol (Fig. S1), indicating the substitution in position 4 of both monosaccharides. The absence of methylether from non-reducing end sugar or disubstituted monosaccharide indicated that the galactosaminogalactan was an unbranched linear polysaccharide. The apparent Mr estimated by gel filtration chromatography after the carboxymethylation of the GG fraction was in agreement with methylation data. The galactosaminogalactan was eluted as a polydisperse homogenous polymer between 10 and 1000 kDa with a median size of 100 kDa (Fig. S2). The 1D 1H and 2D 1H, 13C nuclear magnetic resonance (NMR) spectra of carboxymethylated GG fraction exhibited two main signals in the sugar anomeric region at 5.003/103.07 and 5.287/99.07 ppm compatible with α-anomers (Fig. S3). NMR data showed downfield shifts for the carbone-4 of both sugar residues, indicating their 4-O substitution and their pyranose configuration, which were in agreement with the methylation data.
In order to elucidate the repartition of each monosaccharide on the main polysaccharidic chain, two specific chemical degradations of both galactosaminogalactan fractions (PGG and SGG) were undertaken: periodate oxidation that degraded 4-O-substituted galactose residues and N-de-acetylation/nitrous deamination that degraded hexosamine residues. Solubilised fractions were separated by gel filtration on HW40S column and chemically analysed by methylation, GC-Mass spectrometry (GC-MS), Matrix-assisted laser desorption-Time of flight (MALDI-TOF) and NMR.
The periodate oxidation followed by mild acid hydrolysis solubilised 90% of the SGG. The insoluble product was composed of only N-acetylgalactosamine (GalNAc) residues. Three solubilised fractions were separated by gel filtration on HW40S column (Fig. 2A). GC-MS analyses showed that fraction III corresponded to threitol, resulting from the periodate degradation of 4-O-subsituted galactose residues (not shown). GC-MS analyses of permethylated fraction II showed the presence of compound with a pseudomolecular ion mass [M + H]+ of m/z = 424 and [M + NH4]+ of m/z = 441 corresponding to the permethylated GalNAc-threitol (Fig. S4). NMR data confirmed this analysis and showed that the fraction II contained a compound with an α-GalNAc1-2-threitol arrangement (Table S1). MALDI-TOF analysis of compounds of fraction I indicated a mixture of GalNAc oligosaccharides linked to one threitol residue (Fig. 2B).
Nitrous deamination solubilised 95% of the SGG. Here, only a polygalactan that accounted for 5% of the total polysaccharide was not solubilised. The soluble material was separated on a HW40S gel permeation column. In addition to the anhydrotalose (resulting from the degradation of the galactosamine), a wide peak (fraction I) was eluted from the column (Fig.3). The MALDI-TOF analysis of fraction I revealed the presence of several pseudomolecular ion masses with a regular increase of m/z = 162 and a shift of 18, corresponding to hexose oligosaccharide linked to a non-reduced anhydrotalose in its aldehyde and hydrated forms, respectively [14] (Fig. 3). This result showed that the fraction I was composed of a mixture of galactooligosaccharides of dp 2 to 11 with an anhydrotalose at the reducing end. This result was confirmed by the NMR analysis that indicated the presence of the linkage -4-αGal1-4AHT in this fraction (Table S2).
Carbohydrate structure analyses showed that the galactosaminogalactan from A. fumigatus is a linear heterogeneous polymer of α1-4galactosyl and α1-4N-acetylgalactosaminyl residues. Both SGG and PGG were analyzed and showed similar structures (Table 1). The major differences between these two fractions relied on the degree of polymerization of the galactooligosaccharides and the presence of a higher amount of GalNAc in PGG. The insoluble material after periodate treatment accounted for 25% of the initial material of the PGG indicating that the homogenous linear polyN-acetylgalactosamine was 2 to 3 times higher in PGG. In addition, in contrast to SGG where galactose oligosaccharides of 2 to 10 residues were joined by one GalNAc residue, in PGG GalNAc or polyGalNAc oligosaccharides were mainly joined by a single galactose residue (Fig. S5). These data showed that the galactosaminogalactan of A. fumigatus did not contain a repeat unit and displayed a high heterogeneity in the sequences of oligosaccharides composed of Gal and GalNAc and that this heterogeneity impacted on the physicochemical properties of the polysaccharide.
The antigenicity of the GG was tested first with sera from a blood bank. Surprisingly, antibodies directed against GG were present in most human sera tested: in our experimental conditions, 40% of the 131 tested sera gave by direct ELISA an OD reading >1 at a 1∶500 dilution (Fig. S6). The isotype responsible was mainly IgG2 and full inhibition of the antigen-antibody reaction was obtained with SGG, confirming the specificity of the antibody reaction. Infection with Aspergillus was not associated with an increase in the serum titers against GG. In a similar ELISA format with sera from aspergillosis patients, only 40% of aspergilloma patient gave an OD value higher than 1 by direct ELISA, whereas all these aspergilloma patients had high titers against the galactomannan that is a marker polysaccharide antigen of A. fumigatus. Similarly, only 30% of patient with invasive aspergillosis reacted positively with the galactosaminogalactan (not shown).
The lack of correlation between aspergillosis and the occurrence of high serum titers against GG in healthy patients suggested that the antibody reaction against GG was due to a cross-reactivity with α-GalNAc-containing molecules, since GalNAc has been recognised to be an immunologically reactive hexosamine present in several human or microbial antigens. Among all the molecules tested, the Tn-antigen (α-GalNAc-serine/threonine) or the serotype A marker (α-GalNAc1-3[β-Fuc1-6]β-Gal1-) did not cross react with GG (data not shown). The lack of cross reactivity with human molecules that contained a single GalNAc molecule at their non-reducing end suggested that the presence of several GalNAc molecules was required to form the immunogenic epitope. Accordingly, a high cross reactivity was found between the GG of A. fumigatus and the N-glycan of cell surface glycoproteins of Campylobacter jejuni (AcraA) that is an α1-4 linked GalNAc rich structureα-GalNAc1-4 α-GalNAc1-4 [β-Glc1-3]α-GalNAc1-4 α-GalNAc1-4 α-GalNAc1-3 β-Bac1-Asn(where Bac is 2,4-diacetamido-2,3,6-trideoxy-D-glucose; Asn, asparagine and Glc, glucose, [15]).
A significant positive correlation was calculated for the OD values obtained with the GG and AcraA molecules in 131 blood bank sera. The Spearman's rho correlation coefficient had a value of 0.71 (p<0.0001) (Fig. S6). In addition, ELISA showed that a specific rabbit polyclonal antiserum directed against the N-glycan of surface proteins of C. jejuni reacted positively with the GG of A. fumigatus (not shown). ELISA-inhibition assays were performed on a group of 30 sera with OD >1 for both AcraA and GG antigens. AcraA positive sera with OD>1 were always highly inhibited with 5 µg/ml of SGG (not shown), suggesting that the epitope recognised by these AcraA (and GG) positive sera was a linear α1-4GalNAc oligosaccharide. This result was confirmed by ELISA inhibition studies using the fraction containing exclusively the α1-4GalNAc oligosaccharide obtained after periodate oxidation or acid hydrolysis of SGG (Fig. S7). However, in 20% of serum samples, oligoGalNAc did not completely inhibit the GG recognition, indicating that, in human sera, some IgG could be specifically directed against Gal-GalNAc or Gal-Gal sequences (Fig. S7).
Mice were treated with antigen and CpG (oligonucleotide containing umethylated CpG motifs) as adjuvant to assess the putative protective effect of this antigen against pulmonary aspergillosis in a murine model of vaccine-induced resistance [4]. Figure 4 shows that, in contrast to the protection afforded by conidia, SGG failed to confer resistance to infection and even favored fungal growth (Fig. 4A). No reduced inflammatory pathology was seen in the lung where actual fungal growth was observed in GpG+SGG-treated mice (Fig. 4B) and the cytokine pattern showed that SGG inhibited Ifnγ/Il10 and activated Il4 gene expression in the TLN, thus suggesting inhibition of protective Th1/Treg cells and promotion of Th2 responses (Fig. 4C).
Most interestingly, when the immunomodulatory activity of SGG was assessed in intact mice with primary infection, SGG promoted the infection, as seen by the increased lung CFUs and inflammatory pathology in SGG-treated mice as compared to controls (Fig.5A and B). Figure 5C shows that SGG induced inflammatory cytokine gene expression, such as Tnfα and Il6. Moreover, SGG induced the expression of Il17a genes but suppressed Ifnγ and Il10 expression. These data were in agreement with the expression of the relative Th cell specific transcription factors in the TLN (data not shown). Of interest, SGG appeared to reduce neutrophil infiltrates in the lung during infection, as also seen by the reduced Mpo expression (Fig. 5C). These data suggest that SGG inhibit host defence against A. fumigatus.
Bloodstream neutrophils have a short half-life and prolongation of their lifespan is critical for efficient pathogen destruction. As SGG-treated mice exhibited reduced neutrophil infiltrates in the lung during infection as compared to controls, we investigated the effect of SGG on neutrophil apoptosis. Neutrophils cultured at 37°C died rapidly by apoptosis, about 60% of cells being annexin V+ after 20 h. As previously reported [16], apoptosis was accelerated by cycloheximide and delayed by GM-CSF. The percentage of apoptotic cells in whole-blood samples incubated with SGG (10–20 µg/ml) was significantly higher than in the PBS control. In addition, SGG significantly inhibited GM-CSF-induced PMN survival (Fig. 6).
Since the C-type lectin MGL has been shown to be specific for GalNAc residues, the binding of GG to MGL was investigated. Using SGG and PGG as ligands, ELISA experiment showed a lack of specific interaction of the GG of A. fumigatus and recombinant MGL-Fc. Similarly, immunofluorescence experiments showed that MGL-Fc did not bind to the cell wall of germinated conidia expressing GG on their surface (data not shown). ELISA inhibition using GalNAc coupled to polyacrylamide (GalNAc-PAA) as the ligand showed that GG did not inhibit the binding of GalNAc-PAA to MGL. In contrast, GalNAc oligosaccharides obtained by HCl hydrolysis (Fig. S8) inhibited the interaction with MGL (Fig. 7). Since MGL recognized terminal GalNAc residues and since the average degree of polymerisation of the oligosaccharide fraction used was 7.5, the relative inhibition was similar for GalNAc and the GalNAc oligosaccharide pool when expressed in molar concentration. In contrast to GalNAc monomers that inhibit 100% of the binding at 1 mg/ml in our experimental conditions, no full inhibition was obtained with the oligoGalNAc fraction because at concentrations higher than 500 µg/ml, the GalNAc oligosaccharides precipitated. The lack of binding of MGL to the whole GG was due to the presence of one terminal GalNAc per 700 GalNAc residues in average in the linear 100 kDa GG polysaccharide. Only oligoGalNAc resulting from the degradation of GG can be recognised efficiently by MGL.
Here, we describe the purification and the chemical characterization of a new galactosaminogalactan secreted by the mycelium of A. fumigatus. Cell wall and extracellular polysaccharides containing galactosamine residues have been also identified in other filamentous fungi, such as Neurospora, Rhizopus, Helminthosporium, Penicillium and Aspergillus species [17], [18]. However, the structure of these polysaccharides has been poorly characterized with linkages that can be either α1-4 and/or α1-3 linkages with part of the GalNAc molecules being N-deacetylated [19], [20], [21], [22]. The A. fumigatus galactosaminogalactan is exclusively composed of α1-4linked galactose and α1-4linked N-acetylgalactosamine residues. In our growth conditions, the GG was totally N-acetylated. It was, however, shown that this linear polysaccharide is extremely heterogeneous, with strands of galactose and N-acetylgalactosamine of variable length that impact on the polysaccharide solubility and putatively on biological properties. This heterogeneity is unique to the galactosaminogalactan because the other constitutive cell wall polysaccharides of A. fumigatus are homopolymers (chitin, glucans) or have well defined repeating unit, such as A. fumigatus galactomannan [12], [23]. The main motif is Gal-GalNAc, but the variable Gal/GalNAc ratio inside each polymer chain suggests random synthesis of the polymer, as in some plant polysacharides [24]. The synthesis of polygalactose and polyN-acetylgalactosamine oligosaccharides, as well as the synthesis of repetitive Gal-GalNAc unit is totally unknown. The galactose of GG is present in a pyranose form, whereas the galactose of the galactomannan, which is a major antigen of A. fumigatus, is in a galactofuranose form. A. fumigatus has the ability to synthesise the two isoforms of galactose, like many bacterial, parasite and fungal microorganisms [25], [26], [27], [28], [29]. This was indeed shown in a UDP-Gal epimerase mutant, in which galactofuranose synthesis was abolished, but some galactose was still present in the cell wall, corresponding to the GG [30] (data not shown.).
It was very surprising to see that a majority of the sera from the blood bank had high titers of IgG against GG, with GalNAc residues being the main determinant for the antigenicity. This result suggested that this polysaccharide could be a very potent immunoadjuvant that could be used to induce the production of antibodies against poorly antigenic molecules. The only cross-reacting antigen identified so far was the N-glycans of glycoproteins of C. jejuni. This result suggests that the portal of entry for the GG could be the gut barrier as has been demonstrated for the galactomannan polymer [31], [32]. N-glycoproteins of C. jejuni can bind to human intestinal epithelial cells [33], [34]; Gal/GalNAc rich-polysaccharides are produced by many environmental fungal food contaminants including Aspergillus and Penicillium species suggesting that in both cases α1-4GalNAc oligosaccharides could cross the intestinal epithelium.
This galactosaminogalactan study has confirmed the essential immunological role of the fungal cell wall polysaccharides. This has been seen with all medically important fungi [6], [35], [36], [37]. Some of the cell wall polysaccharides of A. fumigatus, such as α1-3glucan and β1-3glucan chains, have been shown to induce a protective immune response through the activation of Th1, Th17 or Treg responses and the inhibition of the Th2 response [4]. A different immunological function can be conveyed by other cell wall polysaccharides. GG not only is not inducing a protective response but is promoting an immunosuppressive function that can trigger disease in immunocompetent mice. A similar function can probably be attributed to the galactomannan that also has been shown to induce a Th2/Th17 response that was not protective [4]. However, at that time the authors did not investigate the immunosuppressive role of the later polysaccharide in immunocompetent mice. The production of a Th2/Th17 response is in agreement with the presence of anti-GG and anti-Galactomannan IgG2 antibodies in human sera, whereas the level of anti-α1-3glucan and β1-3glucan antibodies in humans is absent or extremely low. In addition, GG-induced PMN death may be involved, at least in part, in the decrease in neutrophil infiltrates in lungs from GG-treated mice despite an increased Th17 response. The GG is the first Aspergillus polysaccharide that induces cell apoptosis. The pathogenic yeast, Cryptococcus neoformans produces a polysaccharide capsule constituted by 2 polymers: glucuronoxylomannan and galactoxylomannan that induce in vitro apoptosis of human macrophages and T-cells [38], [39].
Polysaccharide receptors of mammalian macrophages remain poorly characterized. Besides Dectin1 that recognizes β1-3glucans, receptors able to recognize α1-3glucans or galactan have not been identified yet [40], [41]. The MGL (macrophage galactose-type lectin) was the obvious candidate for GG binding since it recognizes specifically GalNAc residues [42]. This receptor is located at the cell surface of immature dentritic cells and has been shown to be involved in the recognition of pathogens through GalNAc residues and in the retention of immature DCs in peripheral tissue and lymphoid organs [42], [43], [44]. The MGL is able to bind to N-glycoproteins of C. jejuni through α1-4 linked GalNAc residues [45] and the binding of these N-glycans to MGL influences the function of human dentritic cells. However, no specific binding of GG to human MGL-Fc was seen, suggesting that cell surface MGL was not involved in the recognition of A. fumigatus GG. However, the intracellular hydrolysis of GG, as shown for some bacterial polysaccharides [46], may release oligosaccharides that can bind to MGL that has been seen in endocytic compartments. Such intracellular recognition of GG oligosaccharides could then induce the pro-inflammatory response. This hypothesis is currently being investigated.
The A. fumigatus, strain CBS 144–89 was grown in a 15l fermenter in modified Brian medium (2% asparagine, 5% glucose, 2.4 g/l NH4NO3, 10 g/l KH2PO4, 2 g/l MgSO4-7H2O, 26 mg/l ZnSO4-7H2O, 2.6 mg/l CuSO4-5H2O, 1.3 mg/l Co(NO3)2-6H2O, 65 mg/l CaCl2, pH 5.4) for 72 h at 25°C. The mycelium was removed by filtration under vacuum and the supernatant was precipitated with 2.5 vol. of ethanol overnight at 4°C. The pellet was collected by centrifugation (3000g, 10 min). The pellet was washed twice with 2.5 l of 150 mM NaCl and then extracted with 8 M urea (2 h twice at room temperature under shaking). Urea-supernatants (SGG) were pooled and extensively dialyzed against water and freeze-dried. Urea-insoluble pellet (PGG) was washed with water and freeze-dried.
Total hexoses were measured by the phenol-H2SO4 method using galactose as a standard [47]. Total hexosamines were determined with p-(dimethylamino)-benzaldehyde reagent after 4 h of 8N HCl hydrolysis at 100°C using galactosamine as a standard [48]. Monosaccharides were identified by GC as their alditol acetates after total acid hydrolysis (trifluoroacetic acid (TFA) 4N or HCl 4N, 100°C, 4 h) [49]. Threitol resulting from the periodate oxidation of galactose was identified by GC-MS and NMR. In absence of reference spectrum, anhydrotalose resulting from the nitrous deamination of galactosamine was identified by GC-MS by comparison with the mass spectrum of anhydromannitol and by NMR.
Prior to the methylation procedure, polysaccharides were peracetylated as previously described [50]. Dried sample (2 mg) was methylated by the DMSO/lithium methyl sulfinyl carbanion/ICH3 procedure [50]. After hydrolysis of the permethylated sample (4 N TFA 100°C, 4 h), borodeuteride-reduction and peracetylation, methyl ethers were identified by GC-MS. Oligosaccharides were permethylated by the DMSO/NaOH/ICH3 procedure [51].
Polysaccharide fractions (30 mg) were resuspended in 4 ml of 10 mM HCl at 50°C for 24 h and then oxidized with 100 mM sodium m-periodate at 4°C in darkness during 7 days. Excess reagent was destroyed by adding 0.5 ml of ethylene glycol. The solution was dialysed against water and freeze-dried. The material was reduced overnight by 10 mg/ml NaBH4 at room temperature. After neutralisation to destroy the excess of borohydride, reduced oxidized polysaccharide was dialysed against water and freeze-dried. A mild acid hydrolysis was performed by 1.5 ml of 50 mM TFA at 100°C for 1 h. The solubilised fraction was fractionated on a HW40S column (TosoHaas, 90×1.4 cm) equilibrated in 0.25% acetic acid at the flow rate of 0.4 ml/min. Eluted sample were detected by refractometry. The insoluble fraction was washed twice in water.
Polysaccharide fractions (30 mg) were resuspended in 4 ml of 10 mM HCl at 50°C for 24 h and then de-N-acetylated with 40% NaOH (final concentration) at 100°C for 4 h. After neutralisation by addition of acetic acid, samples were dialysed and freeze-dried. Dried samples were resuspended in 600 µl of NaOAc 0.5 M pH 4. The deamination was started by addition of 300 µl of 1 M NaNO3 and performed at 50°C during 3 h with the addition of 300 µl of 1 M NaNO2 each hour. Soluble degraded products were fractionated by gel filtration chromatography through a HW40S column, as described above. Neutral sugars were detected by the phenol-H2SO4 method [47].
10 mg of polysaccharide were treated with 1 ml of 0.1 M HCl for 3 h at 100°C. After neutralisation with 1% Na2CO2, solubilised materials were purified by gel filtration through a Sephadex G25 column (GE Heathcare, 90×1.4 cm) and eluted with 0.25% acetic acid at a flow rate of 9 ml/h.
Due to its insolubility, carboxymethylation of GG was necessary to estimate its molecular size by gel filtration. The polysaccharide (0.2 g) was carboxymethylated by addition of 20 ml of 1.6 M NaOH and 0.3 g of monochloroacetic acid. The mixture was heated at 75°C and stirred magnetically for 8 h. After neutralisation, the solution was dialysed against water and freeze-dried. The carboxymethylated polysaccharide was soluble in 0.5% acetic acid and 10 mg were deposited onto a Sephacryl S400 column (Pharmacia, 90×1.4 cm) at the flow rate of 10 ml/h. Dextrans (Pharmacia, T2000, T500, T70, T40) were used as standards for the column calibration.
GC was performed on a Perichrom PR2100 instrument with a flame ionisation detector using a capillary column (30 m×0.32 mm id) filled with a DB-1 (SGE) under the following conditions: gas vector and pressure, helium 0.7 bar; temperature program 120 to 180°C at 2°C/min and 180 to 240°C at 4°C/min. GC-MS was performed on an EI/CI mass spectrometer detector (model 5975C, Agilent technologies, Massy France) coupled to a chromatograph (model 7890A), using a HP-5MS capillary column (30 m×0.25 mm id, Agilent technologies) under the following conditions: gas vector: helium at 1.2 ml/min; temperature program: 100 to 240°C at 8°C/min and 240°C for 10 min. Ammoniac gas was used for the chemical ionisation.
MALDI-TOF mass spectra were acquired on a Voyager Elite DE-STR mass spectrometer (Perspective Biosystems, Framingham, MA, USA) equipped with a pulsed nitrogen laser (337 nm) and a gridless delayed extraction ion source. The spectrometer was operated in positive reflectron mode by delayed extraction with an accelerating voltage of 20 kV and a pulse delay time of 200 ns and a grid voltage of 66%. Samples were prepared by mixing directly on the target 0.5 µl of oligosaccharide solution in water (10–50 pmol) with 0.5 µl of 2,5-dihydroxybenzoic acid matrix solution (10 mg/ml in CH3OH/H2O, 50∶50, V/V). The samples were dried for about 5 min at room temperature. Between 50 and 100 scans were averaged for every spectrum.
NMR spectra of the polysaccharides were acquired at 318 and/or 343 K on a Varian Inova 500 spectrometer equipped with a triple resonance 1H{13C/15N} PFG (pulsed field gradient) probe whereas spectra of either nitrous deamination or periodate oxidation products were acquired at 298 K on Varian Inova 500 and 600 spectrometers equipped with a triple resonance 1H{13C/15N} PFG and a cryogenically-cooled triple resonance 1H{13C/15N} PFG probe respectively (Agilent technologies, Massy France). Polysaccharidic samples solubilized in acetic acid 0.05%V/V in H2O by warming for one hour at 100°C were freeze dried and redissolved in DCl 0.06 M in D2O (DCl ≥ 99.0% 2H atoms and D2O ≥99.9% 2H atoms, Euriso-top, Saint-Aubin, France). After a second freeze-drying, they were redissolved in D2O and transferred in a 5 mm NMR tube (Wilmad 535-PP, Interchim, Montluçon, France). The final concentration was about 5 mg/mL. Samples were dissolved in D2O and transferred in a 5 mm NMR tube (Shigemi BMS-005 V, Shigemi Inc., Alison Park, United States). 1H chemical shift were referenced to external DSS (2,2-methyl-2-silapentane-5-sulfonate sodium salt hydrate, its methyl resonance was set to 0 ppm). 13C chemical shifts were then calculated from 1H chemical shift and gamma ratio relative to DSS. 13C/1H gamma ratio of 0.251449530 was used [52].
The following strategy was used for assignment of nuclei. First, the non-exchangeable proton resonances of intra glycosidic residue spin systems were assigned using two-dimensional COSY (correlation spectroscopy), relayed COSY (up to two relays) and TOCSY (Total correlation spectroscopy; with mixing time ranging from 30 to 120 ms) experiments [53]. Secondly, 1H-13C edited gHSQC (Gradient selected heteronuclear single-quantum correlation) and gHSQC-TOCSY (mixing time up to 80 ms) experiments allowed the 13C chemical shifts assignment from previously identified 1H resonances [54]. Then, 1H,1H coupling constants for the oligosaccharides were extracted from 1D and/or 2D spectra (1H resolution of 0.1 Hz and 0.6 Hz respectively) and the anomeric configuration was established from the knowledge of 3J1,2 value. Finally, the interglycosidic linkages determination was achieved with 1H-1H NOESY (Nuclear overhauser effect spectroscopy) experiments for the polysaccharides (mixing time of 15 and 50 ms) and with 1H-1H ROESY experiments (mixing time of 250 ms) [55] and/or 1H-13C gHMBC (Gradient selected heteronuclear multiple bond correlation) experiment (long range delay of 60 ms) [54] for the oligosaccharides.
Patient samples were collected according to French Ethical rules. Written informed consent and approval by institutional review Board at the Pitié-Salpêtrière Hospital, at the Etablissement français du sang and at Saint-Louis Hospital were obtained.
Mouse experiments were performed according to the Italian Approved Animal Welfare Assurance A–3143–01. Legislative decree 157/2008-B regarding the animal licence was obtained by the Italian Ministry of Health lasting for three years (2008–2011). Infections were performed under avertin anesthesia and all efforts were made to minimize suffering.
Serum samples from 131 healthy subjects (from Groupe français du sang and Hôpital Saint-Louis, Paris), 25 invasive aspergillosis patients (Hôpital Saint-Louis; kind gift of A. Sulhaian) and 5 aspergilloma patients (CHU Toulouse; kind gift of P. Recco) were used through this study. Blood group was determined by the Etablissement français du sang. The presence of antibodies directed against the A. fumigatus galactosaminogalactan was assessed by a direct enzyme-linked immunosorbent assay method (ELISA). Purified A. fumigatus galactosaminogalactan and AcraA, a recombinant N-glycoprotein from Campylobacter jejuni expressed in E. coli [56], [57] were used as antigens. Wells of microdilution plates (F-form, Greiner, Frickenhausen, Germany) were coated with 100 µl of a suspension of 1 µg/ml galactosaminogalactan (PGG) or 5 µg/ml AcraA diluted in 50 mM Na2CO3 pH 9 and incubated overnight at room temperature. Binding of antibodies to the ELISA-plate was estimated with patient sera diluted 1∶500 and peroxidase-conjugated anti-human immunoglobulin G, as previously described [12]. Cross reactivity between GG and the Tn antigen (α-GalNAc-Serine) was analysed by ELISA with a monoclonal antibody against the Tn antigen (kind gift from Dr R. Lo-Man, Institut Pasteur).
Mice (Balb-C) were immunized subcutaneously with a crude cell wall preparation of A. fumigatus mycelium. Monoclonal antibodies have been produced by F. Nato and P. Beguin (Plateforme technique de protéines recombinantes et anticorps monoclonaux, Institut Pasteur) as previously described [58]. Screening of positive hybridoma was followed by ELISA using the HCl-treated PGG as specific antigen. These mAbs did not react with other Aspergillus polysaccharides, such as galactomannan, β1-3glucan, α1-3glucan. ELISA-inhibition experiments showed that the recognition of mAb-galactosaminogalactan was fully inhibited by oligoGalNAc obtained after partial HCl hydrolysis and gel filtration chromatography on G25 sephadex column as described above (Fig. S8)
Resting conidia and conidia germinated for 8 h in a 2% glucose/1% peptone liquid medium were fixed with 2.5% p-formaldehyde (PFA) overnight at 4°C. After fixation, cells were washed with 0.2 M glycine in PBS for 5 min, then with 5% goat serum in PBS for 1 h. Cells were incubated with the anti-galactosaminogalactan monoclonal antibody at 20 µg Ig/ml in 5% goat serum/PBS for 1 h at room temperature. After washing with PBS-BSA 1%, cells were incubated with a goat FITC-conjugated Ab directed against mouse IgG(H+L) diluted 1∶100 in goat serum/PBS. After washing in PBS, cells were visualized with an inverted fluorescence light microscope. Specificity of labelling was assessed by preincubation of MAb with 50 µg/ml of G25-I fraction (Fig. S8).
Binding assay to the macrophage galactose lectin (MGL) was done by ELISA-inhibition using a recombinant MGL-Fc chimeric protein as previously described [42]. Briefly, α-GalNAc-conjugated polyacrylamide (2 µg/ml, Lectinity) was coated on ELISA plates. Plates were blocked with 1% BSA and the MGL-Fc was added (0.5 µg/ml) for 2 h at room temperature. For inhibition assays, MGL-Fc was previously incubated for 1 h at room temperature in the presence of increasing concentrations of SGG, GalNAc oligosaccharides, Gal or GalNAc or 10 mM EGTA. Binding was quantified using a peroxidase-conjugated secondary antibody directed against human IgG Fc (Jackson).
The putative binding of MGL to mycelium was investigated by immunofluorescence. For that purpose, 0.4×105 conidia were incubated in 200 µl of Brian's medium in wells of chamber slides (Lab-Tek, Nunc) at 37°C for 9 h, washed with PBS and fixed in 2.5% PFA overnight. Cells were washed with 0.2 M glycine in TSM buffer (20 mM TrisHCl; 150 mM NaCl, 2 mM MgCl2, 1 mM CaCl2, pH 7.4) for 5 min, then with 5% goat serum in TSM for 1 h. Cells were incubated with the MGL-Fc at 65 µg/ml in 5% goat serum/TSM for 1 h at room temperature. After washing with TSM, cells were incubated with an anti-human Fc FITC conjugated-goat anti-serum at 15 µg/ml in goat serum/TSM. After washing in TSM, then water, cells were visualized under a fluorescent light microscope.
Female, 8- to 10-week-old inbred C57BL6 (H-2b) mice were obtained from Charles River Breeding Laboratories (Calco, Italy). The vaccination model was as previously described [4]. Briefly, mice were injected intranasally with 2×107 Aspergillus conidia/20 µl saline 14 days before the infection or with 5 µg SGG + 10 nM CpG oligodeoxynucleotide 1862 (CpG)/20 µl saline, administered 14, 7 and 3 days before the intranasal infection. Mice were immunosuppressed with 150 mg/kg/i.p. of cyclophosphamide a day before infection and then intranasally infected with a suspension of 2×107 viable conidia/20 µl saline. In another set of experiments, immunocompetent mice were injected with 250 mg/kg SGG i.n. the day of the infection (2×107 viable conidia/20 µl saline) and on days 1, 2 and 3 post-infection. Mice were monitored for fungal growth (CFU/organ expressed as mean ± SEM) as described [59]. For histology, sections (3–4 µm) of paraffin-embedded tissues were stained with periodic acid-Schiff (PAS). Cytokines were quantified by Real-time PCR, performed using the Stratagene Mx3000P QPCR System, and SyBR Green chemistry (Stratagene Cedar Creek, Texas). Total lung cells were recovered 3 days after the infection. CD4+ T cells (>99% pure on FACS analysis) from thoracic lymph nodes (TLNs) recovered 7 days after the infection, were separated by magnetic cell sorting with MicroBeads and MidiMacs (Miltenyi Biotec). Cells were lysed and total RNA was extracted using RNeasy Mini Kit (Qiagen) and reverse transcribed with Sensiscript Reverse Transcriptase (Qiagen), according to manufacturer's directions. The PCR primers were as described [60], [61]. Amplification efficiencies were validated and normalized against Gapdh. The thermal profile for SYBR Green real-time PCR was at 95°C for 10 min, followed by 40 cycles of denaturation for 30 seconds at 95°C and an annealing/extension step of 30 seconds at 72°C. Each data point was examined for integrity by analysis of the amplification plot. The mRNA-normalized data were expressed as relative cytokine mRNA in treated cells compared with that of mock-infected cells.
Neutrophil apoptosis was quantified by using annexin V and the impermeant nuclear dye 7-amino-actinomycin D (7-AAD) as previously described [16]. Apoptosis was measured after incubation in 24-well tissue culture plates at 37°C with PBS or SGG (1-20 µg/ml) for 20 h. Cycloheximide (Calbiochem, La Jolla, CA) (10 µg/ml) and GM-CSF (R&D Systems) (1000 pg/ml) were used as proapoptotic and antiapoptotic controls, respectively. In some experiments, blood samples were first incubated with SGG for 1 h and then with GM-CSF. Whole-blood samples (100 µl) were then washed twice in PBS, incubated with allophycocyanin (APC)-anti-CD15 mAb (BD Biosciences) for 15 min, and then incubated with fluorescein (FITC)-annexin V (BD Biosciences) for 15 min. After dilution in PBS (500 µl), the samples were incubated with 7-AAD (BD Biosciences) at room temperature for 15 min and analyzed immediately by flow cytometry (GalliosTM, Beckman Coulter). Neutrophils were identified as CD15high cells and 2×105 events were counted per sample. The combination of FITC-annexin V and 7-AAD was used to distinguish early apoptotic cells (annexin V+/7-AAD-), from late apoptotic cells (annexin V+/7-AAD+), necrotic cells (annexin V-/7-AAD+) and viable cells (unstained).
Statistical analyses of the ELISA data were performed using the Spearman's rho test with the JMP software (SAS; Cary, NC).
Data from mouse experiments were analyzed by GraphPad Prism 4.03 program (GraphPad Software, San Diego, CA). Student's t test or analysis of variance (ANOVA) and Bonferroni's test were used to determine the statistical significance (P) of differences in organ clearance and in vitro assays. The data reported are either from one representative experiment out of three to five independent experiments (western blotting and RT–PCR) or pooled from three to five experiments, otherwise. The in vivo groups consisted of 6–8 mice/group.
Data on the measurement of neutroplil apoptosis are reported as means ± SEM. Comparisons were based on ANOVA and Tukey's Post Hoc tests, using Prism 3.0 software (GraphPad software).
|
10.1371/journal.ppat.1004751 | CD169-Mediated Trafficking of HIV to Plasma Membrane Invaginations in Dendritic Cells Attenuates Efficacy of Anti-gp120 Broadly Neutralizing Antibodies | Myeloid dendritic cells (DCs) can capture HIV-1 via the receptor CD169/Siglec-1 that binds to the ganglioside, GM3, in the virus particle membrane. In turn, HIV-1 particles captured by CD169, an I-type lectin, whose expression on DCs is enhanced upon maturation with LPS, are protected from degradation in CD169+ virus-containing compartments (VCCs) and disseminated to CD4+ T cells, a mechanism of DC-mediated HIV-1 trans-infection. In this study, we describe the mechanism of VCC formation and its role in immune evasion mechanisms of HIV-1. We find HIV-1-induced formation of VCCs is restricted to myeloid cells, and that the cytoplasmic tail of CD169 is dispensable for HIV-1 trafficking and retention within VCCs and subsequent trans-infection to CD4+ T cells. Interestingly, introduction of a di-aromatic endocytic motif in the cytoplasmic tail of CD169 that results in endocytosis of HIV-1 particles, suppressed CD169-mediated HIV-1 trans-infection. Furthermore, super-resolution microscopy revealed close association of CD169 and HIV-1 particles in surface-accessible but deep plasma membrane invaginations. Intriguingly, HIV-1 particles in deep VCCs were inefficiently accessed by anti-gp120 broadly neutralizing antibodies, VRC01 and NIH45-46 G54W, and thus were less susceptible to neutralization. Our study suggests that HIV-1 capture by CD169 can provide virus evasion from both innate (phagocytosis) and adaptive immune responses.
| Dendritic cells (DCs) are professional antigen presenting cells, and their sentinel roles are important to elicit a potent antiviral immunity. However, HIV-1 has exploited DCs to spread infection by several mechanisms. One such mechanism is the DC-mediated trans-infection pathway, whereby DCs transmit captured virus to CD4+ T cells. We have recently identified the type I interferon (IFN-I) inducible protein, CD169, as a receptor on DCs which mediates HIV-1 capture and trans-infection. We have also demonstrated extensive co-localization of HIV-1 with CD169 within peripheral non-lysosomal compartments in DCs, although the mechanism and biological importance of the compartment formation remain unclear. Here in this study, we report that a myeloid cell specific co-factor interacts with CD169 following virus capture leading to compartment formation. This co-factor is induced in DCs by an IFN-I-inducing TLR ligand LPS, but not by IFN-I itself. Though the CD169+ HIV-1 containing compartments are surface-accessible, these compartments have considerable depth and are connected to the surface, such that captured virus particles localized within these unique structures are protected from detection by anti-gp120 broadly neutralizing antibodies. Our study suggests that CD169–HIV-1 interaction provides an evasion mechanism from degradation by phagocytosis and neutralization by anti-viral humoral responses.
| Myeloid dendritic cells (DCs) are professional antigen presenting cells that play sentinel roles in sensing pathogens and priming adaptive immunity [1]. HIV has, however, exploited DCs to spread to CD4+ T cells and thus DCs have been suggested to play a role in systemic HIV dissemination from peripheral mucosa to secondary lymphoid tissues [2,3]. While DCs are infected with HIV and DC-derived progeny viruses can infect CD4+ T cells [4–7], productive infection of DCs is limiting for several reasons including low receptor/co-receptor density, presence of cell-intrinsic restriction factors and innate sensing mechanisms eliciting anti-virus immune responses such as type I interferon secretion [8–11]. In contrast, DCs can capture HIV-1 particles and transmit captured virus to CD4+ T cells without establishing productive infection in DCs via a tight cell-to-cell junction called virological synapse [12], a mechanism of DC-mediated HIV-1 trans-infection, that might have evolved to bypass DC-intrinsic anti-viral responses.
Recently, our group and others have identified CD169, also known as Siglec-1, as a predominant receptor for mature DC-mediated capture of HIV-1 and subsequent virus transmission to T cells [13,14]. CD169, a type I transmembrane protein, is the largest member of the sialic-acid-binding immunoglobulin-like lectin (Siglec) family, containing 17 extracellular repeats of immunoglobulin like domain including a N-terminal V-set domain that recognizes α2–3 linked sialic acid residues, a single transmembrane domain and a short cytoplasmic tail (CT) [15]. Upon HIV-1 binding to CD169 on mature DCs, HIV-1 particles accumulate in CD81 tetraspanin+ compartments [13,14]. These compartments are, however, only weakly or poorly stained with endosome/lysosome markers such as CD63 and Lamp1 [16,17]. Whether or not these HIV-1+ compartments are connected to cell surface has been matter of intense debate [reviewed in [18]]. While early studies suggested that endocytosis of HIV-1 particles was important for efficient trans-infection of T cells [19–21], recent studies, however, have called these findings into question and have suggested that surface bound HIV-1 particles present in plasma membrane invaginations were the major source of viruses contributing to efficient DC-mediated HIV-1 trans-infection of T cells [22,23]. Interestingly, the CT of human CD169 contains 44 amino acids, and there are no defined signaling motifs or phosphorylation sites that could contribute to potential virus particle trafficking and internalization upon ligand binding. Therefore, how CD169-bound HIV-1 particles are accumulated and viral infectivity preserved in these compartments remains unclear.
In this study, we have investigated the role of CD169 in trafficking of HIV-1 in mature DCs and facilitating HIV-1 trans-infection of T cells. We found that CD169-mediated HIV-1 trafficking to non-endocytic plasma membrane invaginations is cell-type specific, and that trans-infection could be achieved even in the absence of the CT. Trans-infection efficacy was correlated with the ability of CD169 to retain HIV-1 particles on the cell surface. Interestingly, a single amino acid substitution (Ala to Tyr at position 1683) in the CT of CD169 resulted in the endocytosis of CD169-bound HIV-1 and the mutant CD169 was unable to support trans-infection of T cells, suggesting surface retention by CD169 is critical for HIV-1 to gain access to the trans-infection pathway. Furthermore, using super resolution microscopy, we observed that CD169 and HIV-1 particles were closely associated in LPS-matured DCs in compartments at the cell periphery, approximately 800 nm to 1 μm in depth from the cell surface. These peripheral virus-containing plasma membrane invaginations were not observed in DCs matured by exposure to IFN-α alone, suggesting a requirement for a LPS-inducible host co-factor for formation of the CD169+ HIV-1 containing plasma membrane invaginations. Intriguingly, HIV-1 particles localized within plasma membrane invaginations in LPS-matured DCs were inefficiently accessed by and hence less susceptible to α-gp120 broadly neutralizing antibodies compared to cell free viruses. Our study here, therefore, demonstrates that CD169-mediated capture and trafficking of HIV-1 within DCs can not only provide virus evasion from endocytic mechanisms that can lead to virus particle degradation in lysosomal compartments but also protect HIV-1 from neutralizing antibodies via formation of virus-containing surface-exposed plasma membrane invaginations in LPS-matured DCs.
Previous studies have reported that upon virus capture by mature DCs, HIV-1 particles accumulate in compartments at the cell periphery [23,24]. Furthermore, formation of DC–T cell conjugates results in polarized release of captured virus particles towards T cells for establishment of optimal CD4+ T cell infection [25]. We, as well as others, have recently reported that HIV-1 particles in these compartments are strongly colocalized with CD169 [13,14]. Since CD169 was also colocalized with HIV-1 at the DC–T cell virological synapse [13,14], we wanted to determine the mechanism by which CD169 mediates trafficking of HIV-1 particles in mature DCs.
First, we sought to establish a cell line which could recapitulate the formation of peripheral virus-containing compartments (VCCs) that are observed upon HIV-1 capture by CD169 in mature DCs [13,14]. CD169 was stably transduced into a monocytic cell line THP-1, Raji B cell line and HeLa cells. Cell-surface CD169 expression was tested by flow cytometry and found to be comparable to or higher than that observed on mature DCs (S1 Fig A). Furthermore, induced expression of CD169 on primary cells (LPS treatment of DCs, mature DCs) or engineered expression of CD169 on cell lines (THP-1, Raji or HeLa) resulted in a dramatic enhancement in virus capture (Fig. 1A). Next, we examined if any of the cell lines were able to recapitulate the formation of CD169+ VCCs found in mature DC. Mature DCs, THP-1/CD169, Raji/CD169 and HeLa/CD169 cells were incubated with HIV Gag-mCherry VLPs and stained for total CD169 following membrane permeabilization with TritonX-100 (+Tx100) or without membrane permeabilization to visualize cell surface CD169 (Surface) expression (Fig. 1B). In all the cells tested, VLPs were strongly colocalized with CD169 when stained after membrane permeabilization, as reported previously [14]. In mature DCs, VLPs were often found within compartments at the cell periphery some of which were stained with CD169 without membrane permeabilization (Fig. 1B). In THP-1/CD169 cells, VLPs were strongly colocalized with CD169 in compartments similar to those found in mature DCs (Fig. 1B). Interestingly, similar to mature DCs, CD169+ VLP+ compartments in THP-1/CD169 cells were also partially accessible to surface applied anti-CD169 antibodies. While VLPs captured by Raji/CD169 cells were strongly colocalized with CD169, VLPs remained at the surface in the absence of formation of VCCs. In contrast, captured VLPs were found in intracellular CD169+ compartments in HeLa/CD169 cells, since anti-CD169 antibody was unable to stain VCCs without membrane permeabilization (Fig. 1B).
We next determined if differential localization of HIV-1 particles upon CD169 capture in cell lines could affect CD169-mediated trans-infection. While HIV-1 particles captured by mature DCs, THP-1/CD169 cells or Raji/CD169 cells were transmitted to CD4+ T cells, resulting in robust infection of T cells (Fig. 1C; trans-infection was enhanced more than 10-fold in CD169+ cells compared to CD169low immature DCs or empty vector transduced control cell lines), HeLa/CD169 cells failed to transmit HIV-1 to T cells (Fig. 1C). These findings suggest that retention of HIV-1 particles at the cell surface upon CD169-mediated capture (Fig. 1B) is necessary for virus access to the trans-infection pathway. A corollary of these findings is that endocytosed HIV-1 particles are incompetent for CD169-mediated trans-infection.
CD169 has been reported as a phagocytic receptor on porcine macrophages that can mediate endocytosis of PRSSV [26]. However, to date, no previously defined endocytosis signaling motifs have been described in the CT of human CD169. Since CD169 was trafficked to and colocalized with HIV-1 in surface-accessible compartments in myeloid cells (Fig. 1), we postulated that there was an unidentified trafficking motif in the CT that contributed to colocalization of CD169 and HIV-1 in VCCs. Two CD169 CT truncation mutants were constructed (Fig. 2A), one of which has a stop codon right after the transmembrane domain of CD169 (CD169ΔCT) [15]. Since previous studies have demonstrated severe reduction in cell surface expression of plasma membrane targeted proteins upon deletion of cytoplasmic tails [27,28], we constructed a second CD169 CT mutant that expressed the first four amino acids of CT (CD169ΔCT4R). These CD169 CT mutants were transduced into THP-1 cells and the ability of these stably transduced cell lines expressing CD169 mutants to capture HIV and form VCCs was compared to that observed with THP-1 cells expressing wild type CD169 (THP-1/CD169) (Fig. 1B). Deletion of the cytoplasmic tail (CD169ΔCT) resulted in decreased expression of CD169 in in THP-1 cells (Fig. 2B and S1 Fig B). Furthermore, cell surface expression of CD169ΔCT was further reduced (Fig. 2C and D) and resulted in severe attenuation of HIV-1 capture (Fig. 2E). Interestingly, inclusion of the membrane proximal 4 arginine residues in the cytoplasmic tail resulted in higher expression of CD169 in cells and partial rescue of cell surface expression of CD169 (Fig. 2C, 2D and S1 Fig B), and importantly, capture of HIV-1 particles (Fig. 2E). The efficiency of virus capture by THP-1/CD169ΔCT4R cells was much lower than that exhibited by wt THP-1/CD169 cells (Fig. 2E), in correlation with CD169 expression level on the cell surface (Fig. 2C and D). We next co-cultured CD4+ T cells with THP-1 cells expressing CD169 CT mutants to investigate the role of CD169 CT in mediating HIV-1 trans-infection. Interestingly, THP-1/CD169ΔCT4R but not THP-1/CD169ΔCT cells could transmit HIV-1 to CD4+ T cells (Fig. 2F). Furthermore, there was no significant difference in the efficiency of trans-infection (T cell infection per amount of virus captured by THP-1 cells) mediated by THP-1/CD169 and THP-1/CD169ΔCT4R cells (Fig. 2G). Finally, CD169+ VCCs were also observed in THP-1/CD169ΔCT4R cells (Fig. 2H), suggesting that the CD169 CT sequences downstream of the four arginine residues were dispensable for the formation of VCCs and CD169-mediated HIV-1 trans-infection.
Whether endocytosed HIV-1 particles in DCs remain competent for trans-infection has been a matter of significant debate [6,13,16,18,21–24,29]. Since CT sequences proved dispensable for CD169 mediated trans-infection and HIV-1 particles captured by CD169 remained within surface-accessible VCCs (Fig. 2) we hypothesized that HIV-1 has exploited CD169-dependent trafficking to evade host phagocytic responses that target captured pathogens to degradative compartments. To test this hypothesis, we introduced a single point mutation in the CT of CD169 that introduces a di-aromatic motif (Ala to Tyr at position 1683) such as one known to be essential for mannose receptor-mediated phagocytosis of bacterial pathogens bearing terminal mannosylated proteins in their cell wall [30,31] (Fig. 3A). This mutant CD169, designated as CD169YF, was constitutively expressed in THP-1 cells via retroviral transduction. CD169YF expression was confirmed both by western blotting (Fig. 3B) and flow cytometry (Fig. 3C and S1 Fig B), and was expressed at similar levels at the cell surface as wild type CD169 (Fig. 3D). Interestingly, kinetics of anti-CD169 antibody internalization were enhanced in THP-1/ CD169YF compared to THP-1 cells expressing wild type CD169, suggesting the single amino acid substitution functioned as an internalization signal motif (S2 Fig A).
We next investigated the localization of HIV Gag-mCherry VLPs upon capture by THP-1/CD169YF cells. Both wt CD169 and CD169YF expressing THP-1 cells were challenged with VLPs and stained for CD81, a tetraspanin protein that colocalizes with HIV-1 in VCCs in mature DCs [16,17], or CD63 and Lamp1 (late endosomal compartment markers). In THP-1/CD169 cells, VLPs were colocalized with CD81, but not with CD63 or Lamp1 (Fig. 3E), which is consistent with previous reports on HIV-1 localization in mature DCs [16,17,23,24]. In contrast, colocalization of HIV Gag-mCherry VLPs in THP-1/CD169YF cells was reduced within CD81+ compartments but enhanced within CD63+ or Lamp1+ compartments (Fig. 3E). In addition, VCCs in THP-1/CD169YF were inefficiently accessed by surface-applied antibodies (S2 Fig B and C), suggesting that CD169YF internalized VLPs to late endosomes or lysosomes. These differences in intracellular localization of HIV Gag-mCherry VLPs between THP-1/CD169 and THP-1/CD169YF cells were statistically significant (Fig. 3F and S2 Fig B). While THP-1/CD169YF cells captured HIV-1 particles as efficiently as THP-1/CD169 cells (Fig. 3G), HIV-1 trans-infection of CD4+ T cells by THP-1/CD169YF cells was completely abrogated (Fig. 3H and I). Collectively, these results suggest that endocytosed HIV-1 particles are incompetent for accessing the CD169-dependent HIV-1 trans-infection pathway.
We next sought to characterize the architecture in greater detail of CD169+ VCCs in mature DCs. CD169 expression in DCs is induced upon treatment with TLR ligands such as LPS and polyI:C [14]. As opposed to TLRs, exposure to IFN-α that results in partial maturation of DCs [32,33] can also upregulate CD169 expression [14], though putative differences in IFN-α and TLR-induced maturation phenotypes might alter virus trafficking in differentially matured DCs [34]. Therefore, DCs differentially matured with LPS or IFN-α (referred as LPS-DC or IFN-α-DC, respectively), were used for determining HIV-1 localization in phenotypically divergent CD169-expressing primary cells. While CD169 was highly upregulated on both LPS-DCs and IFN-α-DCs (Fig. 4A), IFN-α-DCs displayed a partial maturation phenotype expressing low levels of the activation antigens, CD86 and HLA-DR consistent with previously published findings [32,33]. HIV-1 capture by both LPS-DCs and IFN-α-DCs and subsequent trans-infection of CD4+ T cells were similarly enhanced over that observed with immature DCs (Fig. 4B and C).
We next investigated the nature of the CD169+ VCCs formed in IFN-α-DCs and LPS-DCs by conventional deconvolution microscopy, electron microscopy and super-resolution microscopy. While HIV-1 particles were strongly colocalized with CD169 at the cell periphery in both cell types, virus-containing pocket-like compartments were only found in LPS-DCs but not in IFN-α-DCs (Fig. 4D). Electron microscopy also revealed virus-containing pocket-like compartments in LPS-DCs as previously reported (Fig. 4E, S3 Fig A to C and [19,20]). In contrast most of HIV-1 particles were found at the surface in IFN-α-DCs (Fig. 4E and S3 Fig D to F) in valleys between dendritic extensions or present in structures presumably formed upon collapse of the dendrites in IFN- α-DCs (Fig. 4E and S3 Fig D to F). This divergent localization of HIV-1 in IFN-α- and LPS-DCs was further investigated by super resolution microscopy. We used a fluorescence photoactivation localization microscopy (FPALM) with bi-plane capture technique [35–37] which allowed us to visualize CD169+ VCCs in mature DCs at 20–40 nm (X-Y) and 50–80 nm (Z) resolution. In agreement with conventional deconvolution and electron microscopy (Fig. 4D and E), HIV-1 and CD169 were accumulated in pocket-like compartments in LPS-DCs, while HIV-1 was found mostly at the cellular edge in IFN-α-DCs (Fig. 4F, top panels, S4 Fig D to F and S1 Movie). Focusing at the cross section of these cells (along the line between a and b in the top panels), the depth of the compartments harboring HIV-1 particles in LPS-DCs was measured at 800 nm-1 μm (Fig. 4F, middle panels). In contrast, HIV-1 particles (p24gag) and CD169 clustered in a long “valley-like” structure that appeared to be on the surface of IFN-α-DCs (Fig. 4F, middle panels). In both cell types, p24gag molecules (green) were closely associated with CD169 (red) (Fig. 4F, bottom panels, S4 Fig and S2 Movie and S3 Movie), implying an important role of CD169 in the formation of VCCs in DCs. All together, these results suggested that LPS and IFN-α treatment of DCs resulted in divergent CD169+ VCCs and that formation of CD169+ HIV-1 containing pocket-like structures in DCs requires a LPS-induced host factor.
We next sought to determine if HIV-1 particles in CD169+ VCCs in LPS-DCs and IFN-α-DCs are exposed to the extracellular milieu. Uninfected or virus-exposed CD169+ LPS-DCs were subjected to extensive proteolytic digestion with either trypsin or pronase (Fig. 5A). Cell-surface CD169 expression and amount of HIV-1 particles that remained associated with LPS-DCs and IFN-α-DCs following protease treatment was determined by FACS and p24gag ELISA, respectively (Fig. 5A). In the absence of HIV-1 binding, CD169 was mostly present at the LPS-DC and IFN-α-DC surface, and remained sensitive to cleavage by pronase but not trypsin (Fig. 5B and C, No virus), suggesting that extracellular domain of CD169 lacks trypsin-recognition sequences. Interestingly, when HIV-1 containing compartments were formed prior to pronase treatment, CD169 was still sensitive to pronase-digestion (Fig. 5C, + Virus). We next investigated if HIV-1 particles associated with CD169 in LPS-DCs and IFN-α-DCs were sensitive to protease digestion. LPS-DC- or IFN-α-DC-associated HIV-1 particles were insensitive to trypsin exposure (Fig. 5D), consistent with the findings that CD169 was trypsin resistant (Fig. 5B and C) and CD169—HIV-1 interaction is a protein (CD169)—lipid (GM3) interaction [13,14,38,39]. In contrast, consistent with the ability of pronase to effectively cleave cell-surface exposed CD169 (Fig. 5B), pronase treatment decreased LPS-DC- or IFN-α-DC-associated HIV-1 content by ~60% (Fig. 5D), suggesting VCCs were accessible to surface-applied pronase. The pronase-resistant cell-associated HIV-1 fraction might be attributed to those virus particles that either remain bound to residual CD169 (~20% of the CD169 molecules remained cell-associated even after pronase treatment; Fig. 5C), or p24gag in the cytoplasm after virus fusion with mature DCs. All together, these results suggest that the majority of CD169+ VCCs in LPS-DCs and IFN-α-DCs remain accessible from the cell surface and thus sensitive to surface-applied pronase digestion.
We next sought to determine if CD169-bound HIV-1 particles in CD169+ VCCs in LPS-DCs or IFN-α-DCs were accessible to surface-applied large molecular probes, such as anti-gp120 broadly neutralizing antibodies (bNAbs) or anti-CD169 mAbs. LPS-DCs or IFN-α-DCs were pulsed with fluorescent HIV-1 Lai-iGFP particles and stained for either HIV-1 gp120 or CD169 prior to fixation and permeabilization such that antigens accessible to surface-applied antibodies would only be visualized. As a comparison, staining for total HIV-1 gp120 or CD169 was performed in parallel after fixation and permeabilization (+ Tx100). Most of the HIV-1 particles and CD169 in IFN-α-DCs were found at the cell surface and could be visualized with surface-applied anti-gp120 (Fig. 6A) or anti-CD169 antibodies (Fig. 6B). To quantify accessibility of captured HIV-1 particles to surface-applied antibodies, the fraction of fluorescent HIV-1 particles overlapping with antibody staining was calculated using Manders' coefficients. Quantification revealed no significant differences between surface-exposed and total molecule staining for both HIV-1 gp120 and CD169 in IFN-α-DCs (Fig. 6C and D). Though some of the Lai-iGFP+ VCCs in LPS-DCs were stained by surface-applied anti-gp120 or anti-CD169 antibodies (Fig. 6A and B), some of the virus particles present at the “bottom” of the pocket-like structures (arrowheads, Fig. 6A and B, LPS) were inaccessible to both anti-gp120 and anti-CD169 antibody suggesting that CD169+ VCCs in LPS-DCs were either closed structures or inaccessible to surface-applied probes due to steric hindrance. Furthermore, we observed statistically significant differences in Manders’ coefficients amongst cells (LPS-DCs) stained by the two distinct staining techniques (Fig. 6C and D).
The differences in surface accessibility of antibodies to CD169+ VCCs between LPS-DCs and IFN-α-DCs prompted us to hypothesize that HIV-1 particles localized within VCCs in LPS-DCs might remain competent for mature DC-mediated trans-infection even in the presence of anti-gp120 bNAbs. To test this hypothesis, neutralization assays were performed using anti-gp120 bNAbs (VRC01 and NIH45–46 G54W) and two-domain sCD4 (sCD4–183). Either LPS-DC- or IFN-α-DC-associated HIV-1 was incubated with increasing concentrations of VRC01, NIH45–46 G54W or sCD4–183 prior to co-culture with CD4+ T cells. In parallel, cell free HIV-1 infection of CD4+ T cells was performed in the presence or absence of VRC01, NIH45–46 G54W or sCD4. While VRC01 and NIH45–46 G54W inhibited cell free CCR5-tropic HIV-1 (pseudotyped with Bal Env) infection of CD4+ T cells efficiently [IC50 (VRC01) = 0.035 ± 0.005 μg/ml (Fig. 6E and H), IC50 (NIH45–46 G54W) = 0.012 ± 0.004 μg/ml (Fig. 6F and I)], transfer of LPS-DC-associated HIV-1 particles was inefficiently neutralized [IC50 (VRC01) = 1.152 ± 0.308 μg/ml (Fig. 6E and H) and IC50 (NIH45–46 G54W) = 0.223 ± 0.062 μg/ml (Fig. 6F and I)]. Interestingly, transfer of HIV-1 particles from IFN-α-DCs to T cells was more susceptible to neutralization by VRC01 and NIH45–46 G54W [IC50 = 0.508 ± 0.155 μg/ml (Fig. 6E and H) and 0.069 ± 0.016 μg/ml (Fig. 6F and I), respectively] than that mediated by LPS-DCs, though efficiency of IFN-α-DC mediated transfer in the presence of VRC01 and NIH45–46 G54W was still greater than that observed for cell-free infection of T cells.
In contrast, sCD4–183, a small gp120-neutralizing reagent (26kD) was able to equally inhibit all three modes of infection, namely cell free, IFN-α-DC-mediated and LPS-DC-mediated HIV-1 infection of CD4+ T cells (Fig. 6G). The IC50 values for cell free, IFN-α-DC or LPS-DC-associated HIV-1 infection of CD4+ T cells were 0.171 ± 0.047, 0.307 ± 0.039 and 0.467 ± 0.117 μg/ml, respectively (Fig. 6J). These results suggest that HIV-1 association with CD169 in VCCs within mature DCs protects viruses from detection by anti-gp120 bNAbs and might provide virus evasion from adaptive immune responses in vivo.
In this study, we have characterized CD169+ VCCs in mature DCs and found that captured HIV-1 particles in LPS-matured DCs were localized within surface-connected plasma membrane invaginations at depths of ~800nm—1μm (Fig. 4). We hypothesize that multivalent association of CD169 and HIV-1 particles or clustering of multiple CD169 molecules (induced upon virus particle binding) might enhance localized concentration of receptor-ligand complexes that are retained at the cell surface because of the inability of CD169 to mediate endocytosis. Recruitment of a LPS-induced myeloid cell-specific co-factor(s) upon virus capture to the localized membrane microdomain might place additional strain and stress on the membrane that is relieved by formation of membrane invaginations (Fig. 7), though the mechanisms that inhibit membrane closure and endosome formation remain to be identified. Interestingly, localization of HIV-1 within VCCs in LPS-matured DCs reduced the accessibility of anti-gp120 bNAb, VRC01, to virus particles and hence, reduced the neutralization efficiency of anti-gp120 bNAbs, VRC01 and NIH45–46 G54W (Fig. 6). Localization of HIV-1 and CD169 in a lattice-like structure in the VCCs might provide steric hindrance only to large molecules such as neutralizing antibodies. HIV-1 particles captured by IFN-α-DC-associated HIV-1 were also less susceptible to VRC01 and NIH45–46 G54W compared to cell free HIV-1, though localization of HIV-1 particles in IFN-α-DCs was in compartments that lacked comparable depth to that observed in LPS-DCs (Fig. 4), presumably because of the lack of recruitment of the co-factor(s) in IFN-α-DCs necessary for formation of membrane invaginations. Since HIV-1 particles were found at the bottom of dendrites and/or surrounded by dendrites (Fig. 4E) forming clusters of CD169 and HIV-1 in a "valley-like" structure (Fig. 4F), this unique localization of HIV-1 in the IFN-DCs might also hinder access of VRC01 and NIH45–46 G54W to HIV-1.
Acute infection of HIV-1 in vivo induces various pro-inflammatory cytokines including type-I interferon [40]. Such inflammatory conditions can differentiate monocytes at the site of infection into inflammatory DCs [41–43]. We have reported previously that inflammatory DCs generated in vitro are CD169+ and efficiently disseminate HIV-1 to T cells [14]. While triggering type I IFN responses induces the expression of number of interferon-stimulated genes, some of which are anti-viral, and restrict virus replication, induction of CD169 might offset ISG-mediated restrictions to virus replication in the peripheral mucosa. Thus, in acute phase of infection, type I IFN-induced CD169 on inflammatory DCs might support establishment of infection in mucosal CD4+ T cells. In addition to IFNs, previous studies have demonstrated increases in serum LPS levels over the course of HIV-1 infection, primarily due to the compromised integrity of gut epithelium [44]. Mucosal damage-associated-translocation of LPS might lead to systemic activation of DCs and upregulation of CD169 that not only enhances virus spread to CD4+ T cells, but also might provide evasion from humoral responses that develop but fail to neutralize cell-to-cell transmission in the mucosal tissues. A great deal of effort currently supports the design of viral vector-based immunoprophylactic regimens that express anti-gp120 bNAbs to induce protection in vivo [45,46]. Since DC-mediated trans-infection of CD4+ T cells has been suggested as an important pathway of HIV-1 dissemination in vivo [2,3], significantly increased antibody titers might be necessary in vivo to achieve neutralization of IFN-α-DC or LPS-DC-mediated HIV-1 dissemination.
CD169 is expressed exclusively on myeloid cells in vivo [15,47], and interestingly formation of CD169+ VCCs upon HIV-1 capture was only recapitulated in DCs and THP-1 monocytoid cell line, but not Raji B cells or HeLa cells, constitutively expressing CD169 (Fig. 1B). While HIV-1 particle associated with CD169 remained at the cell surface in Raji/CD169 cells, virus particles accumulated in intracellular, surface-inaccessible compartments in HeLa/CD169 cells. These results suggest the formation of surface-connected VCCs might require a cofactor specific to myeloid cells. Interestingly, CT sequences of CD169 proved dispensable for VCC formation, since truncation of CT downstream of the four membrane-proximal arginine residues (THP-1/CD169ΔCT4R) preserved cell surface expression of CD169 and importantly resulted in VCC formation in THP-1 cells upon HIV-1 capture (Fig. 2). Previous results from our laboratory and others have also demonstrated that HIV-1 capture by DCs is also reduced upon treatment with β-methyl-cyclodextrin, a cholesterol sequestering reagent [17,48]. Since VCC formation occurred even in the absence of CT, it is possible that lateral association and clustering of CD169 is driven by interaction of transmembrane domain of CD169 with a protein and/or lipid molecule in such cholesterol-rich plasma membrane microdomains. It is of note that expression of such cofactor(s) is regulated by LPS stimulation of DCs but not upon treatment with IFN-α alone (Fig. 4D). Future studies will be needed to identify the nature of this myeloid-cell specific co-factor by comparing TLR4-mediated (TRIF or MyD88-dependent) and IFNAR-mediated (JAK-STAT dependent) signaling pathways in myeloid cells.
CD169 is a pattern recognition receptor that captures diverse bacterial and viral pathogens by recognizing α2,3-sialylated glycoconjugates on the pathogen surface [47]. In addition to HIV-1, capture of other enveloped viruses such as murine leukemia virus, nipah and hendra hemorrhagic fever viruses by CD169 is also dependent on binding α2,3-sialylated GSLs incorporated in the virus particle membranes [14,49]. Though some studies have implicated CD169 as an endocytic receptor that mediates internalization of pathogens into early endosomes [26], CD169, unlike other members of the Siglec protein family, CD169 has no defined endocytic motifs in its CT [15,47]. Furthermore, studies described in this report suggest that HIV-1 particles captured by CD169 are not targeted for endocytosis but rather retained on the myeloid cell-surface in plasma membrane invaginations. Interestingly, exogenous introduction of a di-aromatic endocytic motif in the CT of CD169 resulted in HIV-1 internalization and dramatic attenuation of CD169-mediated HIV-1 trans-infection (Fig. 3). Collectively, these results strongly suggest a requirement for HIV-1 retention at the cell surface for accessing the mature DC/CD169-mediated trans-infection pathway.
We hypothesize that this unique trafficking pattern is beneficial to HIV-1 since it provides virus particles evasion from endocytic pathways in DCs that can result in degradation of virions and/or antigen presentation to T cells to elicit robust adaptive immune responses [50]. It is interesting to speculate that HIV-1 might have evolved to assemble and exit from GM3-enriched plasma membrane microdomains [49] such that GM3-dependent interactions of HIV-1 with CD169 provide virus sanctuary from both myeloid cell-intrinsic phagocytic mechanisms of virus degradation and antibody-dependent detection and neutralization of virus infectivity. Furthermore, as DCs have been proposed to be the first cells to encounter HIV-1 particles in the genital mucosa [2], topical administration of such reagents might prevent sexual transmission of HIV-1 Therefore, development of agents that target HIV-1–CD169 interaction might be an attractive potential anti-viral therapeutic to curtail the HIV-1 pandemic.
This research has been determined to be exempt by the Institutional Review Board of the Boston University Medical Center since it does not meet the definition of human subjects research, since all human samples were collected in an anonymous fashion and no identifiable private information was collected.
Human CD169 was cloned into a retroviral expression vector, LNCX (LNC-CD169) and has been described previously [14]. Truncations in cytoplasmic tail of human CD169 were introduced by PCR using the following primer sets: for CD169/ΔCT, CD169–4188-sense (ATCAGGGACAGGCCATGTCC) and CD169-ΔCT-antisense (TTTTTATCGATCACCAGGTGTAGCAGGCCC CCAGG); for CD169/ΔCT4R, CD169–4188-sense and CD169-ΔCT4R antisense (TTTTTATCGATTAACGCCTCCTTCTCCAGGTGTAGCAGGC). Point mutation in the cytoplasmic tail of CD169 (A1683Y) was introduced by PCR-based site-directed mutagenesis (QuikChange; Agilent Technologies) using the following primers: CD169-YF-sense (CGAGAATTCGGTGGAGATGTATTTTCAGAAAGAGACCACGC) and CD169-YF-antisense (GCGTGGTCTCTTTCTGAAAATACATCTCCACCGAATTCTCG). A SbfI-ClaI fragment containing truncations or mutations in the CT of CD169 was replaced into the corresponding portion of LNC-CD169. All clones were verified by sequencing. Stable expression of CD169 CT mutants in THP-1 monocytic cell line, HeLa cell line and Raji B cell line was accomplished by transduction with VSV-G pseudotyped LNC-CD169 mutant retroviral vectors followed by G418 selection as previously described [14]. CD169 positive cells were further purified either by MACS (Miltenyi Biotec) or FACS (BD AriaIII). Protein expression was confirmed by western blot analysis and flow cytometry (BD Calibur) as described below.
The expression plasmid, pGag-EGFP, that expresses a HIV-1 Gag-eGFP fusion protein, was obtained from the NIAID AIDS Reference and Reagent Program. HIV-1 Gag-mCherry expression plasmid that expresses a red fluorescent Gag-mCherry fusion protein has been described previously [24]. HIV-1 LaiΔenv-luc (Env deficient HIV-1 Lai containing a luciferase reporter gene in place of the nef orf), Lai/Balenv-luc (a CCR5-tropic infectious proviral construct encoding luciferase) and a CCR5-tropic infectious proviral plasmid Lai/YU-2env have been described previously [51–53]. Lai-imCherry, a proviral construct producing red fluorescent infectious virus particles, was derived from Lai-iGFP [14] by replacing the GFP-encoding fragment with that of mCherry. The CCR5-tropic HIV gp160 (Bal env) expression vector was generated from a CXCR4-tropic HIV gp160 (Lai env) expression vector [54] by replacing the entire Lai env gene with the corresponding region of Bal env.
Human dendritic cells (DCs) were derived from CD14+ peripheral blood monocytes, as described previously [14]. DCs were matured with ultrapure E. coli K12 LPS (100 ng/ml; Invivogen) or IFN-α (1000 U/ml; PBL Interferon Source) for 2 days prior to use in the assays. Primary human CD4+ T cells were positively isolated from CD14-depleted PBMCs, using CD4-conjugated magnetic beads and LS MACS cell separation columns (Miltenyi Biotech). Positively isolated CD4+ T cells were activated with 2% PHA (Invitrogen) for 2 days, washed and cultured in IL-2 (50 U/ml) containing RPMI supplemented with 10% FBS. HEK293T (human kidney epithelial cell line), Raji (human B cell line, obtained from the NIH AIDS Research and Reference Reagent Program), THP-1 (human monocytic cell line, clone ATCC, obtained from the NIH AIDS Research and Reference Reagent Program), and HeLa cells have been described previously [39]. Replication competent viruses, Lai/Balenv-luc, Lai/YU-2env and Lai-imCherry, were derived from HEK293T cells via calcium phosphate transfection as described previously [54]. Fluorescent HIV Gag derived virus-like particles (VLPs) were generated via transient transfections of HEK293T cells with HIV Gag-eGFP or HIV Gag-mCherry expression plasmids. HIV-1 vectors pseudotyped with Bal Env were generated from HEK293T cells via co-transfection of HIV-1 LaiΔenv-luc with HIV-1 Bal Env expression plasmid. Viruses or VLP-containing cell supernatants were harvested 2 days post-transfection, cleared of cell debris by centrifugation (300 x g, 5 min), passed through 0.45 μm filters, and stored at—80°C until further use. For some experiments, viruses in the supernatants were concentrated by ultracentrifugation on a 20% sucrose cushion [24,000 rpm and 4°C for 2 hr with a SW32Ti rotor (Beckman Coulter)]. The virus pellets were resuspended in PBS, aliquoted and stored at -80°C. The capsid content of infectious HIV-1 particles or VLPs was determined by a p24gag ELISA [54]. VSV-G pseudotyped LNC-CD169 mutant retroviral vectors were prepared as described elsewhere [14].
Mature DCs (1x105; see above), THP-1 cells (1x105), Raji cells (1x105) or HeLa cells (5x104) were incubated with virus (10–20 ng p24gag) for 2 hr at 37°C in complete RPMI media, washed 4 times with PBS and analyzed for capture using either p24gag ELISA. Virus capture was quantified by measuring p24gag associated with lysed cells using an in-house p24gag ELISA described previously [54]. For transfer of Lai/Balenv-luc infectious viruses, 1x105 of mature DCs, THP-1 cells, Raji cells or 5x104 HeLa cells were incubated with virus (10–20 ng p24gag) for 2 hr at 37°C in complete RPMI media, washed 4x with PBS and co-cultured with autologous or heterologous CD4+ T cells at a 1:1 or 1:2 cell ratio in complete RPMI media with IL-2. The cells were lysed at 48 hours post infection and luciferase activity in the cell lysates was measured using Bright-Glo (Promega). All assays were performed with cells derived from a minimum of three independent donors and each experiment was performed in triplicate.
To assess expression of CD169 CT mutants in THP-1 cells, cell lysates were resolved with SDS-PAGE and transferred onto PVDF membranes. Membranes were probed with mouse anti-CD169 antibody (7D2, Novus Biologicals) or rabbit anti-actin antibody (SIGMA). To measure CD169 expression on the cell surface, cells were stained with Alexa488-conjugated mouse anti-CD169 (AbD Serotec) and analyzed with a FACS Calibur (BD), as detailed in supporting methods (see S1 Text).
To determine the extent of cell-surface exposure of CD169 bound HIV-1 particles on mature DC surface, cells incubated in the presence or absence of 10 ng (p24gag) of HIV-1 for 2 hours were washed extensively with cold-PBS and chilled at 4°C for 30 min, prior to incubation with pronase (4 mg/ml, in Ca2+ containing PBS, Roche) for 30 min at 4°C. Alternatively, virus-exposed cells were incubated with 0.25% trypsin (Invitrogen) for 5 min at 37°C. After the treatment, cells were washed extensively with cold-PBS. The amount of cell-associated HIV-1 particles was measured by p24 ELISA (described above) and CD169 expression was measured by FACS as described above. The values were normalized to those of untreated samples.
To investigate sensitivity of DC-associated HIV-1 to gp120-targeting neutralizing reagents, VRC01, NIH45–46 G54W and sCD4–183 (obtained from the NIH AIDS Reagent Program), 5x104 mature DCs were incubated with HIV-1 Bal Env pseudotyped LaiΔenv-luc particles (10 ng p24gag) for 2 hours at 37°C, washed 4 times with PBS and chilled at 4°C for 15 min. Serially diluted VRC01, NIH45–46 G54W or sCD4–183 starting at 10 or 20 μg/ml in final was added to HIV-1 pulsed DCs or cell free virus (50 ng of p24gag) and incubated for 1 hour at 4°C. Cells were washed twice with cold-PBS and autologous or heterologous CD4+ T cells were added at 1:2 ratio to monitor trans-infection of CD4+ T cells as described above. Cell free HIV-1 was added directly to CD4+ T cells. These experiments were performed in triplicates with DCs from at least nine independent donors. Cell free infections of CD4+ T cells were performed with cells derived from at least five independent donors in triplicates. Nonlinear regression was used to estimate a fitted curve and IC50 values were calculated in GraphPad Prism 5.
To investigate structure of HIV-1 containing CD169+ compartments in DCs, super resolution FPALM (fluorescence photoactivated localization microscopy) was used. LPS- or IFN-α-stimulated DCs (1x106 cells) were incubated with 2.5 μg p24gag Lai/YU-2env for 2 hours at 37°C and washed extensively to remove unbound viruses. Cells were fixed with 4% PFA, permeabilized, blocked with normal donkey serum and stained for HIV-1 with mouse monoclonal anti-p24 antibody (AG3.0, obtained from the NIH AIDS Reagent Program), followed by secondary donkey anti-mouse IgG-Cy3B. Cells were blocked with 20% normal mouse serum, and CD169 expression was visualized with Alexa647-conjugated anti-CD169 mAb (AbD Serotec). Cells were attached onto a grass coverslip and subjected to microscopy analysis. Images were recorded with a Vutara 200 super-resolution microscope (Bruker Nano Surfaces, Salt Lake City, UT) based on the Biplane FPALM approach [36]. Samples were imaged using a 647 nm and 488 nm excitation lasers, respectively, and 405 nm activation laser in photoswitching buffer comprising of 20 mM cysteamine, 1% betamercaptoethanol and oxygen scavengers (glucose oxidase and catalase) in 50mM Tris buffer at pH 8.0. Images were recorded using a 60x/1.2 NA Olympus water immersion objective and Photometrics Evolve 512 EMCCD camera with gain set at 50, frame rate at 50 Hz and maximal powers of 647 nm, 488 nm and 405 lasers set at 8, 10, and 0.05 kW/cm2 respectively. Data was analyzed by the Vutara SRX software (version 4.09). Briefly, particles were identified by their brightness from the combined images taken in both planes and two color channels simultaneously. If a particle was identified in multiple subsequent camera frames, data from these frames was combined for the specific identified particle. Identified particles were then localized in three dimensions by fitting the raw data in a customizable region of interest (typically 16x16 pixels) centered around each particle in each plane with a 3D model function which was obtained from recorded bead data sets. The four-recorded fields were aligned automatically by computing the affine transformation between each pair of planes. Sample drift was corrected by cross-correlation of the determined localized particles [55] or tracking of fiduciary markers. Fit results were stored as data lists for further analyses.
Structure of HIV-1 containing CD169+ compartments in DCs was visualized by electron microscopy. 4.5 x106 LPS- or IFN-α-stimulated DCs were incubated with 9 μg p24gag Lai/YU-2env for 2 hours at 37°C and washed extensively to remove unbound viruses. Cells were fixed with 4% PFA and 1% glutaraldehyde in 0.1 M PHEM buffer (60 mM PIPES, 25 mM HEPES, 2 mM MgCl2 and 10 mM EGTA). Cells were further fixed with 2% osmium tetroxide, dehydrated in ethanol and embedded in epoxy resin as previously reported [29]. Ultra-thin sections (60–80 nm) of embedded cells were stained with 3% uranyl acetate and 1% lead citrate and subjected to imaging with a Philips CM-12 electron microscope at 100kV.
To determine if VLP containing compartments remained connected to the cell surface, HeLa/CD169 cells (seeded on coverslips in a 24-well tissue culture plate on the day before), THP-1/CD169 cells, Raji/CD169 cells or mature DCs (2x105 cells) were incubated with 10 ng p24gag of VLP Gag-mCherry for 2 hours at 37°C, washed extensively to remove unbound VLPs, chilled to 4°C and stained with Alexa488-conjugated mouse anti-CD169 mAb (AbD Serotec) on ice for 1 hour, prior to fixation with 4% PFA. For total CD169 staining, virus-exposed cells were fixed, permeabilized and stained with Alexa488-conjugated mouse anti-CD169 for 1 hour at RT. To determine intracellular localization of VLPs, THP-1/CD169 (2x105) cells were incubated with 10 ng p24gag Gag-mCherry VLPs for 2 hours at 37°C, washed and fixed with 4% PFA. Virus-containing compartments were visualized by staining with anti-human CD81 (BD), anti-human CD63 (Santa Cruz) or anti-human Lamp1 (Santa Cruz) at 10 μg/ml followed by secondary Alexa594-conjugated goat anti-mouse IgG (Invitrogen) at 10 μg/ml. To visualize HIV-1 containing compartments in DCs, LPS- or IFN-α-stimulated DCs (4x105 cells) were incubated with 1 μg Lai-iGFP for 2 hours at 37°C and washed extensively to remove unbound viruses. For staining of surface-exposed HIV-1 particles or CD169, cells were chilled and stained with anti-gp120 antibody (VRC01) or anti-CD169 mAb (7D2, Novus Biologicals), respectively, at 10 μg/ml on ice for 1 hour, prior to fixation with 4% PFA. Alternatively, virus-exposed cells were fixed with 4% PFA, permeabilized with TritonX-100 and then stained with anti-gp120 antibody (VRC01) or anti-CD169 mAb (7D2, Novus Biologicals). Cells were then stained with Alexa594-conjugated goat anti-human IgG (for visualizing gp120 staining, Invitrogen) or Alexa594-conjugated goat anti-mouse IgG (for CD169 staining, Invitrogen). Nuclear staining was visualized with DAPI (Sigma) and cells were mounted on a glass slide with Fluoromount G (Southern Biotech). Images were acquired using a Olympus IX70 microscope equipped for DeltaVision deconvolution (Applied Precision). Images were deconvolved using the SoftWoRx software (Applied Precision), processed with ImageJ and pseudocolored for data presentation. For the colocalization study on THP-1 cells, images were acquired for at least 20 cells, deconvoluted, flattened for maximum intensity in order to avoid selection bias inherent in analysis of single focal plane images and analyzed for Pearson’s coefficient of correlation (R) with ImageJ. For the quantification of accessibility of antibodies to HIV-1 particles in CD169+ VCCs in differentially stimulated mature DCs, images were acquired on 10–15 cells, deconvolved, and flattened for maximum intensity. To specifically quantify the fraction of fluorescent HIV-1 particle overlapping with antibody signals (acquisition of red (antibody) on green (HIV-1)), Manders’ coefficients were calculated using ImageJ. Thresholds were set as the mean ± standard deviation of intensity at each channel.
|
10.1371/journal.pgen.1005372 | Arabidopsis PCH2 Mediates Meiotic Chromosome Remodeling and Maturation of Crossovers | Meiotic chromosomes are organized into linear looped chromatin arrays by a protein axis localized along the loop-bases. Programmed remodelling of the axis occurs during prophase I of meiosis. Structured illumination microscopy (SIM) has revealed dynamic changes in the chromosome axis in Arabidopsis thaliana and Brassica oleracea. We show that the axis associated protein ASY1 is depleted during zygotene concomitant with synaptonemal complex (SC) formation. Study of an Atpch2 mutant demonstrates this requires the conserved AAA+ ATPase, PCH2, which localizes to the sites of axis remodelling. Loss of PCH2 leads to a failure to deplete ASY1 from the axes and compromizes SC polymerisation. Immunolocalization of recombination proteins in Atpch2 indicates that recombination initiation and CO designation during early prophase I occur normally. Evidence suggests that CO interference is initially functional in the mutant but there is a defect in CO maturation following designation. This leads to a reduction in COs and a failure to form COs between some homologous chromosome pairs leading to univalent chromosomes at metaphase I. Genetic analysis reveals that CO distribution is also affected in some chromosome regions. Together these data indicate that the axis remodelling defect in Atpch2 disrupts normal patterned formation of COs.
| In the reproductive cells of many eukaryotes, a process called meiosis generates haploid gametes. During meiosis, homologous parental chromosomes (homologs) recombine forming crossovers (CO) that provide genetic variation. CO formation generates physical links called chiasmata, which are essential for accurate homolog segregation. CO control designates a sub-set of recombination precursors that will mature to form at least one chiasma between each homolog pair. Recombination is accompanied by extensive chromosome reorganization. Formation of a proteinaceous axis organizes the pairs of sister chromatids of each homolog into conjoined linear looped chromatin arrays. Pairs of homologs then align and synapse becoming closely associated along their length by a protein structure, the synaptonemal complex (SC). The SC is disassembled at the end of prophase I and recombination is completed. We have investigated the link between recombination and chromosome remodelling by analysing the role of a protein, PCH2, which we show is required for remodelling of the chromosome axis during SC formation. In wild type, immunolocalization reveals depletion of the axis-associated signal of the axis component, ASY1, along synapsed regions of the chromosomes. In the absence of PCH2, the ASY1 signal is not depleted from the chromosome axis and the SC does not form normally. Although this defect in chromosome remodelling has no obvious effect on CO designation, CO maturation is perturbed such that the formation of at least one CO per homolog pair no longer occurs.
| During meiosis genetic crossovers (COs), the products of homologous recombination, in conjunction with sister chromatid cohesion establish physical links, referred to cytologically as chiasmata, between homologous chromosome pairs (homologs) to ensure accurate chromosome segregation at the first nuclear division that follows prophase I. In the absence of crossing over the homologs segregate randomly. This leads to the formation of aneuploid gametes following the second meiotic division [1]. Recombination is initiated by the programmed formation of DNA double-strand breaks (DSBs), catalysed by the topoisomerase type II related protein Spo11 [2,3]. In Saccharomyces cerevisiae (budding yeast) around 40% of DSBs are repaired as non-CO (NCO) products with the remainder progressing to form COs [4]. In Arabidopsis thaliana and other multicellular organisms the proportion of COs is substantially less, typically 5–10% [5]. Importantly, the CO/NCO balance is highly controlled. This control is manifested in several ways. First, each pair of homologs acquires at least one CO. Second, CO interference ensures that multiple COs are well spaced along the chromosomes. Finally, CO homeostasis maintains CO numbers in the face of perturbations that may affect the number of earlier recombinational interactions [6–10]. It is hypothesized that a CO patterning phenomenon, that can be simulated by the beam-film model, underlies these three features of CO control [11,12].
In budding yeast, DSBs form in early leptotene coincident with the elaboration of a proteinaceous chromosome axis that organizes each pair of sister chromatids into linear looped chromatin arrays conjoined by a shared axis. DSBs occur in the context of the chromosome axis [13–15]. At the transition from leptotene to zygotene, formation of the synaptonemal complex (SC), a tripartite structure comprising the chromosome axes linked by overlapping transverse filaments (TFs), is initiated at multiple synapsis initiation sites [1,16,17]. Synapsis continues throughout zygotene bringing the axes into close apposition and is completed at the onset of pachytene when the SC is fully formed. This programmed morphogenesis of the chromosome axes and SC is critical for the coordination of recombination, playing important roles in the meiosis-specific bias that favours inter-homolog recombination and the maturation of CO designated recombination intermediates [18–26].
In budding yeast mutation of the PCH2 gene, which encodes a member of the conserved AAA+ ATPase protein family, disrupts remodelling of the chromosome axis during prophase I of meiosis [27,28]. In wild type cells the chromosome axis protein, Hop1, and the SC transverse filament protein, Zip1, appear to load uniformly at a basal level along the chromosomes. Superimposed on this, each forms a series of non-overlapping, alternating hyper-abundant domains. In a pch2 mutant this domainal loading is disrupted to give a uniform overlapping signal for each protein along the chromosomes [27,28]. Pch2 may modulate inter-homolog bias by remodelling the chromosome structure in the vicinity of DSBs and have a role in a recombination checkpoint [29,30]. Loss of the protein also affects CO formation. In one study, a pch2∆ mutant had increased COs on larger chromosomes, while CO frequency on the small chromosome III was unaffected [31]. Genetic data suggested the mutant also exhibited a defect in CO interference. A link with CO interference was also established in a parallel study, although in this instance no effect on overall CO number was observed [28]. However, further analysis based on the distribution of foci of the E3 ligase Zip3, which arise at CO designated intermediates and so provide an early marker for CO interference, reported that interference is not affected in a pch2 deletion mutant [32].
Orthologs of PCH2 have been identified in a variety of organisms. In mouse, analysis of a weak hypomorphic allele of TRIP13 (PCH2) indicated that the protein was required for the efficient repair of DSBs that enter the NCO pathway but not COdesignated intermediates, which were processed normally. Despite the presence of unrepaired DSBs synapsis was normal in these mice [33]. Subsequently, a study of a more severe Trip13 mutant reported a defect in CO formation and synapsis [34]. Similar to Pch2 in budding yeast, TRIP13 is required for the depletion of the Hop1 orthologs HORMAD1 and HORMAD2 along synapsed regions of the chromosome axes [35]. In Drosophila, PCH2 acts in a checkpoint to monitor defects in recombination and chromosome structure [36]. In Caenorhabditis elegans it is reported to maintain the fidelity of recombination and synapsis during prophase I by acting to constrain these processes [37]. A PCH2 ortholog, referred to as CRC1 (CENTRAL REGION COMPONENT1) has also been identified in rice (Oryza sativa) [38]. The CRC1 protein is 43.8% identical to TRIP13 and 23.1% identical to Pch2 from budding yeast. The crc1 mutant is completely asynaptic and forms univalents at metaphase I due to a failure to make DSBs [39].
Here we describe the identification and analysis of the PCH2 orthologs from Brassica oleracea and its close relative Arabidopsis thaliana. Using super-resolution structured illumination microscopy (SIM) we reveal dynamic changes in localization of PCH2 in relation to chromosome axis and SC morphogenesis during meiotic prophase I. Analysis of Arabidopsis mutants lacking PCH2 reveals a meiotic role that is markedly different to that reported for the rice CRC1 protein. Loss of PCH2 results in a failure to deplete ASY1 from the chromosome axes during zygotene coupled with a synaptic defect. Although recombination initiation and CO designation appears to occur normally during early prophase I, the defects in remodelling of the chromosome axes which influence SC formation are associated with a disruption of the patterned formation of COs along the homologous chromosomes.
Protein complexes were immunoprecipitated from Brassica oleracea var. alboglabra A12DH pollen mother cells (PMCs) in meiotic prophase I using an anti-ASY1 antibody as previously described [40]. Co-precipitating proteins were analysed by mass-spectrometry and identified using the Brassica rapa sequence [41]. Up to 10 unique peptides corresponding to 25% sequence coverage (124/490 amino acids) of the Bra013827 predicted gene product were detected in three independent experiments and were absent from control samples (S1A Fig). The protein was identified as a P-loop containing nucleoside triphosphate hydrolase superfamily member. BLAST searches revealed that the protein is 87% identical to the Arabidopsis At4g24710 predicted gene product. ClustalW2 analysis (http://www.ebi.ac.uk) showed that Bra013827 and At4g24710 are members of a sub-family of the AAA+ ATPase super-family that contains the budding yeast PCH2 and mouse TRIP13 genes (S1B Fig).
To determine whether At4g24710 encodes a functional ortholog of Pch2/TRIP13 we obtained three T-DNA insertion lines: SAIL_1187_C06, SALK_031449 and SALK_130138, hereafter referred to as Atpch2-1, Atpch2-2 and Atpch2-3 respectively. For all lines, the T-DNA insertion site was confirmed by DNA sequencing and the absence of a full-length AtPCH2 transcript confirmed by RT-PCR (S2 and S3 Figs). The vegetative phenotype of each line was indistinguishable from wild type Arabidopsis, Col-0, but their fertility was reduced (S4A and S4B Fig). Quantification of the fertility defect in Atpch2-1 revealed a slight, yet significant reduction in mean silique length from 1.66 ± 0.05 cm in wild type to 1.41 ± 0.06 cm (n = 50; P<0.05) in Atpch2-1 (n = 50). This was accompanied by numerous gaps between the seeds within the siliques such that overall the mean seed-set was significantly reduced from 67.8 per silique in wild type to 34.6 in Atpch2-1 (n = 50; P< 0.01). Analysis of Atpch2-2 and Atpch2-3 revealed very similar fertility defects (S4C Fig).
The Atpch2 reduced fertility phenotype suggested a meiotic defect. Cytogenetic analysis of DAPI stained chromosome spreads from Atpch2-1 PMCs at leptotene revealed the threadlike chromosomes with no obvious differences to the wild type controls (Fig 1A and 1B). In wild type PMCs, the homologs achieved full synapsis at pachytene with the threadlike signals visibly paired along their lengths, giving a thicker appearance than at leptotene (Fig 1C). However, in Atpch2-1 pachytene stage cells were not observed, instead the majority of the chromosomes remained as single threadlike signals with some limited regions where paired axes were visible (Fig 1D). During diplotene both Atpch2-1 and wild type chromosomes desynapsed and began to condense, such that by diakinesis chiasmata linking the homologs were visible. At metaphase I, following further condensation, distinct bivalents were observed. Five bivalents were invariably present in wild type, but some Atpch2-1 nuclei contained a mixture of bivalent and univalent chromosomes (Fig 1E and 1F). To quantify this we counted chiasmata in Atpch2-1 in metaphase I chromosome spreads using fluorescent in situ hybridization (FISH) with 45S and 5S rDNA probes to identify individual chromosomes [42] (Fig 1G and 1H). This revealed a significant reduction in the mean chiasma frequency in Atpch2-1 compared to wild type (6.9; n = 50 versus 9.6; n = 50; P < 0.001). No univalents were observed in the wild type sample, whereas they were present at a frequency of 10.0% in Atpch2-1 with all chromosomes affected. Similar results were obtained for Atpch2-2 (6.9; n = 37; P < 0.001; univalent frequency 14.6%) and Atpch2-3 (6.2; n = 26; P < 0.001; univalent frequency 7.7%). As a consequence of this, in contrast to wild type, mis-segregation of the chromosomes was observed at the first meiotic division in Atpch2-1 (Fig 1I and 1J) leading to unbalanced tetrads (Fig 1K and 1L). No precocious sister chromatid separation was observed suggesting that there was no cohesion defect. Analyses of Atpch2-2 and Atpch2-3 revealed that the meiotic defect in the three mutants is essentially identical (S5A–S5H Fig). To confirm that the observed phenotype was due to a loss of AtPCH2 function, an allelism test was conducted by crossing Atpch2-1 with Atpch2-2. Phenotypic and cytological analysis of the Atpch2-1/Atpch2-2 progeny revealed the same defects as in the parental lines confirming that these arose due to the loss of AtPCH2 function (S5I–S5M Fig).
We next investigated if the reduction in chiasmata in Atpch2-1 reflected a defect in CO formation. Approximately 85% of COs in Arabidopsis exhibit CO interference [43]. Formation of these, so-called Class I COs, require a group of proteins known as ZMMs (Zip1, Zip2, Zip3/Hei10, Zip4, Mer3, Msh4 and Msh5) [5,18]. The remainder (Class II) are insensitive to CO interference and dependent on the structure-specific endonuclease AtMUS81 [44]. To determine if the loss of AtPCH2 affected one or both classes of COs we generated an Atmsh5-1/Atpch2-1 double mutant. In Atmsh5-1, the number of chiasmata per PMC ranged between 0 and 4 with a mean chiasma frequency of 1.2 (n = 50) (S6B Fig). In comparison, the mean chiasma frequency in the Atmsh5-1/Atpch2-1 double mutant was significantly reduced to 0.3 (n = 50; P < 0.001), with the number of chiasmata per nucleus ranging between 0 and 2 (S6C Fig).
Thus, overall the cytological analysis suggests that the reduction in chiasmata in Atpch2-1 arises from a recombination defect that impacts on both interference sensitive and insensitive CO formation, rather than through an effect on sister chromatid cohesion.
The failure to observe pachytene stage PMCs in Atpch2-1 suggested a defect in formation of the SC. To investigate further, we examined chromosome axis reorganization during early to mid-prophase I using immunocytochemistry combined with fluorescence light microscopy and SIM. At leptotene in wild type Arabidopsis, the HORMA domain protein ASY1 is detected in chromosome spreads of PMCs as a linear axis-associated signal. This appears to be comprised of a series of alternating regions of higher and lower signal intensity, suggestive of a domainal organization of ASY1 abundance along the chromosome axis [19,45] (Fig 2A and 2C). Analysis of Atpch2-1 PMCs at leptotene did not reveal any obvious differences, with localization of ASY1 appearing normal (Fig 2B and 2D). Similarly, the cohesin complex protein SYN1 [46,47] (S7A and S7B Fig) and the chromosome axis protein ASY3 [19] (S7C and S7D Fig) appeared unaffected in Atpch2-1, with both forming a linear axis-associated signal from leptotene through to mid-prophase I. That SYN1 localization was normal supported the cytological observation that there was no evidence of a sister chromatid cohesion defect. Consistent with these observations, comparison of the mean total axis length per PMC at leptotene was not significantly different to wild type (Atpch2-1: 229 μm versus wt: 220 μm, n = 10, P = 0.53).
Previous immunolocalization studies show that the Arabidopsis SC TF protein ZYP1 begins to polymerize between the aligned homologous chromosomes from multiple sites of synapsis initiation at the onset of zygotene. Polymerization continues throughout zygotene until completion of SC formation at pachytene [20]. Dual-localization of ZYP1 and ASY1 in wild type Arabidopsis revealed that SC formation is accompanied by a reduction in the intensity of ASY1 signal which appeared less continuous and appeared to be associated with the chromatin loops rather than the axis along synapsed regions (Fig 2E–2J: compare synapsed segment with unsynapsed region in 2G and 2J; S8A–S8C Fig). Quantification of the relative intensity of the ASY1 signal (S9 Fig) indicated a significant reduction of 67.0% (n = 23; P = <0.001) on the synapsed region compared to the unsynapsed axes (S9C and S9D Fig; S1 Table). Analysis of Atpch2-1 PMCs at mid/late-prophase I suggested that unlike wild type, the ASY1 signal intensity along the synapsed compared to unsynapsed regions remained unchanged (n = 22; P = 0.25) (S9E and S9F Fig; S1 Table). However, the differentiation of the ASY1 signal into putative domains of high and low intensity appeared enhanced in the mutant, possibly a consequence of the delayed synapsis and increased axis compaction relative to leptotene (S9F Fig). SC polymerization was compromised in the mutant (Fig 2M–2P). Stretches of ZYP1 were detected but varied in number and length from cell to cell. On average the SC length at late prophase I in Atpch2-1 was 32% that of wild type (57 μM, n = 16 versus 179 μM, n = 8), although this ranged from 13% to 57%.
To gain further insight into the relationship between PCH2 and the components of the chromosome axes, we conducted immunolocalization studies using an anti-PCH2 antibody on chromosome spreads of wild type PMCs from Arabidopsis and B. oleracea (Figs 3 and 4). Analysis of Arabidopsis using SIM revealed numerous chromatin-associated PCH2 foci (mean 165; n = 10) in G2 coinciding with the appearance of foci and short stretches of ASY1 (Fig 3A). Most PCH2 foci remained distinct from the ASY1 signal (Fig 3A, inset 3C). As the chromosome axis formed in leptotene, the ASY1 signal became more linear. At this stage the proteins appeared associated, with 51.2% (n = 12) of the PCH2 foci overlapping the ASY1 signal to some extent (Fig 3B, inset 3D), possibly a consequence of the chromosome reorganization that occurs at leptotene. As the SC formed during zygotene PCH2 distribution changed. ASY1 associated foci were no longer apparent. Instead PCH2 now tracked the depleted ASY1 signal along the synapsed region, forming a linear array of foci that tended to coalesce (Fig 3E and 3I). Dual-immunolocalization of PCH2 and ZYP1 confirmed that the PCH2 signal localized to the regions where SC nucleates and was present as foci along the SC during zygotene through pachytene (Fig 3J–3P and S10A–S10C Fig). Analysis in Atasy1 and Atasy3 mutants where SC formation is severely compromised, such that only short stretches or accumulations of ZYP1 are formed [19,23], also revealed colocalization of the PCH2 and ZYP1 signals (S10D–S10I Fig). No PCH2 signal was detected in any of the three Atpch2 mutant lines (S11 Fig).
Immunolocalization in B. oleracea PMCs revealed that similar to Arabidopsis, numerous PCH2 foci were detected in late G2/early leptotene (S12A and S12B Fig). At late leptotene/early zygotene PCH2 formed fewer, large foci (mean number per nucleus = 14.2; range = 10–20; n = 18) (Fig 4A–4C). Dual localization of ZYP1 and PCH2 at this stage indicated that these foci correspond to sites of SC nucleation at the leptotene/zygotene transition (Fig 4D–4F) and SIM analysis revealed that the ZYP1 signal at the nucleation site often appeared to form a ‘arrowhead-like’ shape to which PCH2 co-localized (Fig 4G). From the SIM images the arrowhead-like foci were estimated to have a mean length of 602 nm (range = 560–640 nm; n = 40) and a mean maximum width of 419 nm (range 400–480 nm; n = 40) and seemed quite consistent in number (mean 15 per nucleus; range = 12–19; n = 5). In a larger sample, analysed using fluorescence microscopy, a mean of 12.2 foci per nucleus was observed (n = 50). Although the range (5–22) was greater than in the SIM sample, most nuclei (76.0%) contained 10 or more foci. The slight variation in the number of PCH2 foci observed in these experiments probably reflected the dynamics of the process and the increased resolution afforded by SIM relative to fluorescence microscopy. In addition to the large foci, slightly more numerous smaller ZYP1 foci were also observed at early zygotene (mean number per nucleus = 17.0; n = 50). These also co-localized with PCH2 (Fig 4D–4F). As the SC began to extend, SIM revealed extensive overlap between ZYP1 and PCH2 signals each appearing to be comprised of multiple smaller foci (Fig 4H). At zygotene, the ASY1 signal appeared to be reduced along synapsed regions of the chromosomes (S12C Fig) (64.3% reduction relative to unsynapsed axes; n = 14), which were decorated with numerous small PCH2 foci. At pachytene, PCH2 foci were still detected along the entire length of the ZYP1-stained SC as well as in the surrounding chromatin (S12D Fig).
In budding yeast deletion of PCH2 results in an accumulation of nuclei in pachytene and a delay in progression through meiosis I [27]. We therefore investigated if the protein has a role in prophase I progression in Arabidopsis. 5-ethynyl-2’-deoxyuridine (EdU) was used to pulse-label Atpch2-1 PMCs during meiotic S-phase [48]. Progression through meiosis was then monitored (S13A Fig). In wild type and Atpch2-1, EdU labelled leptotene nuclei were detected 10h post S-phase. By 25h all labelled wild type PMCs were at zygotene or pachytene and at zygotene in Atpch2-1. At 32h the wild type PMCs had exited pachytene and were at diplotene/diakinesis and by 36h were at the dyad stage, whereas Atpch2-1 PMCs were still at zygotene suggesting a delay of 5-8h (S13A and S13B Fig).
The cytological analysis (see earlier) suggested a defect in CO formation in Atpch2 mutants. To further examine the recombination phenotype of Atpch2-1 we used the fluorescent-tagged-line (FTL) system [49,50] which relies on the segregation of three genetically linked transgenic markers, each encoding a distinct pollen-specific fluorescent protein expressed post-meiotically. The FTLs are in a qrt1-2 mutant background which prevents the separation of the gametes and facilitates the visualisation of the meiotic recombination events that have occurred between the transgenic markers in the tetrad pollen [51,52]. Three pairs of adjacent genetic intervals, one on each arm of chromosome 5 and another on chromosome 2 were examined (S14 Fig). This revealed that the genetic map distance determined using the Perkins mapping equation [53] in the adjacent intervals I5c and I5d was not significantly affected by the Atpch2-1 mutation (I5c wild type 6.1 cM v Atpch2-1 6.8 cM; P = 0.17; I5d wild type 5.5 cM v Atpch2-1 6.0 cM; P = 0.28) (Fig 5A). However, interval I5a showed a significant decrease in recombination frequency in the presence of Atpch2-1 (15.1 cM) compared to wild type (27.7 cM; P < 0.001), whereas the map distance in interval I5b exhibited a significant increase from 17.3 cM in wild type to 22.3 cM in the mutant (P < 0.001) (Fig 5A). A significant increase in map distance was observed in intervals l2f and l2g in the presence of the Atpch2-1 mutation (l2f wild type 6.1 cM / Atpch2-1 8.0 cM P < 0.001; l2g wild type 5.1 cM / Atpch2-1 7.1 cM P < 0.001) (Fig 5A).
We used the FTL data to obtain a genetic estimate for CO interference in adjacent intervals by calculating the Interference Ratio (IR). This method, developed by Malkova et al. [54], uses the ratio of the genetic map distance in an interval with and without the presence of a CO in an adjacent interval to provide an estimate of the strength of CO interference. When COs in adjacent intervals are entirely independent of each other the IR is 1, indicating no interference. Values less than 1 indicate increasing levels of (positive) interference with a value of 0 indicating complete interference. IR ratios greater than 1 are indicative of negative interference. The CO interference ratio was 0.412 for I5ab in wild type. In Atpch2-1, the genetic map distance of I5a was similar with and without the presence of a CO in the interval I5b (14.8 cM with a CO in interval I5b vs 15.2 cM without a CO in interval I5b). The CO interference ratio is 0.976 and is statistically higher than wild type (Z-score = 5.40; P < 0.001) (Fig 5B). This suggests that CO interference is reduced in the interval I5ab in Atpch2-1. In contrast, the CO interference ratio of I5cd is similar in wild type (0.568) and in Atpch2-1 (0.552; Z-score = 0.01; P = 0.92) (Fig 5B). The interference ratio for interval l2fg is also increased in the Atpch2-1 mutant. In wild type the ratio is 0.113 whereas in Atpch2-1 it is 0.315 (P = 0.021).
We also used the FTL data to estimate the coefficient of coincidence (CoC) for the three pairs of intervals. The CoC is calculated by dividing the observed frequency of double COs in two adjacent intervals by the expected frequency assuming no interference [55]. When interference is absent the CoC is 1 and where it is complete the CoC is 0. The overall result was similar to that obtained for the IR (S2 Table). For l5a/b interference appeared reduced (CoC wild type = 0.46 v CoC Atpch2-1 = 0.99), for l5c/d it was unchanged (CoC wild type = 0.60 v CoC Atpch2-1 = 0.60) and for l2fg there was an apparent decrease (CoC wild type = 0.13 v CoC Atpch2-1 = 0.37).
In the absence of CO control the numerical distribution of chiasmata between cells is predicted to fit a Poisson distribution [56]. This expectation is borne out in ZMM mutants such as Atmsh4 and Atmer3, whereas in wild type the distribution is non-Poissonian. [43,57,58]. We analysed the chiasma distribution in the sample of Atpch2-1 cells described above. The number of chiasmata per nucleus ranged between 4 and 10 in Atpch2-1 and between 7 and 12 in wild type. Further inspection revealed that the proportion of Atpch2-1 PMCs with a chiasma frequency close to the mean of 6.9 was over-represented in the sample analysed, with 74% having between 6 and 8 chiasmata per cell (vs 42.8% if the numerical distribution of chiasmata was random) (Fig 5C). Over-distribution of chiasma around the mean is also a feature of wild type [43]. The chiasma distribution in Atpch2-1 differed significantly from a Poisson distribution (X(11)2 = 45.2; P < 0.001). This was also confirmed in Atpch2-2 (X(11)2 = 37.2; P < 0.001) and Atpch2-3 (X(11)2 = 40.00; P < 0.001).
We investigated the basis for the reduction in chiasmata in Atpch2-1 using immunolocalization of recombination pathway proteins on prophase I chromosome spreads from Atpch2-1 PMCs. Immunolocalization of the strand-exchange proteins RAD51 and DMC1 which are recruited to DSBs at leptotene was used to monitor early recombination and immunolocalization of the ZMM protein AtMSH4 was used to detect later recombination progress [43,59,60]. There were no significant differences between wild type (Fig 6A, 6C and 6E) and Atpch2-1 (Fig 6B, 6D and 6F) PMCs. At mid-leptotene the mean number of RAD51 foci in Atpch2-1 was 144 versus 146 in wild type (n = 12; P = 0.37) (Fig 6A and 6B). For DMC1 the corresponding values were 167 versus 173 (n = 12, P = 0.56) (Fig 6C and 6D). In PMCs at the leptotene/zygotene transition the mean number of MSH4 foci was 150 in Atpch2-1 versus 152 in wild type (n = 12; P = 0.63) (Fig 6E and 6F).
HEI10 (Human enhancer of invasion-10) is a member of the Zip3/Hei10 family of proteins which are thought to possess SUMO/ubiquitin E3 ligase activity [61]. Studies reveal that Zip3/Hei10 marks the sites of future type I COs [61,62]. In Sordaria macrospora Hei10 foci that mark COs are ~300 nm in size and emerge from a much larger population of small axis-associated foci during early/mid-prophase I [63]. Dual localisation of ASY1 and HEI10 on chromosome spreads of Arabidopsis wild type and Atpch2-1 PMCs at late leptotene showed that in both cases HEI10 formed a very similar large number of foci (166 versus 165 respectively, n = 10; P = 0.81) along the chromosome axes (Fig 6G–6L). As prophase I progressed the foci decreased in number and disappeared by pachytene. Most foci were small (~175 nm) but in addition, a number of larger (>250 nm) HEI10 foci were observed in both sets of PMCs. In wild type at the leptotene/zygotene transition we observed 7 to 15 large HEI10 foci (mean 10.6, n = 33). This remained constant through late pachytene (mean 9.9 range 9–12, n = 14) (Fig 7A, 7C and 7I). During early prophase I the number and distribution of large HEI10 foci in Atpch2-1 PMCs was not significantly different to wild type (mean 10.6 versus 10.2; P = 0.20; n = 21). However, at mid/late prophase the mean number of large HEI10 foci was significantly reduced to 6.9 (n = 27) compared to wild type nuclei (P < 0.001) (Fig 7B, 7D and 7I). HEI10 foci were mostly found as singletons on the stretches of SC in Atpch2-1 (83.4% n = 185) (Fig 7B and 7D), with two or three HEI10 foci observed in 14.6% and 2.0% of cases respectively.
Dual localization of HEI10 and ZYP1 in B. oleracea PMCs at the leptotene/zygotene transition revealed that most of the large ZYP1 foci at SC nucleation sites that had been shown to co-localize with PCH2 at this stage (see earlier), also co-localized with HEI10 (86.0% foci; n = 30 nuclei) (S15 Fig).
To confirm that the reduction in large HEI10 foci in Atpch2-1 reflects a reduced number of mature CO intermediates we analysed the distribution of the late recombination protein MLH1 which marks the sites of Type I COs/chiasmata [64]. Dual immunolocalization of MLH1 and ZYP1 (N-terminus Ab, see Materials and Methods) on chromosome spreads of wild type PMCs, showed the number of MLH1 foci per nucleus varied between 9 and 11 with a mean count of 9.9 (n = 12) at pachytene (Fig 7E, 7G and 7I). In Atpch2-1 the mean number of MLH1 foci per nucleus was 7.1 (n = 12), a significant reduction compared to wild type (P < 0.002) (Fig 7F, 7H and 7I). We noted that in both cases the MLH1 foci were often adjacent to the ZYP1 signal rather than directly over the SC central region. Similar to the distribution of HEI10 foci, MLH1 foci were mostly observed as singletons on stretches of SC in Atpch2-1 PMCs (61.2% n = 85) with two or three foci occurring in 28.2% and 10.6% cases respectively (Fig 7F and 7H).
Formation of the chromosome axis in early prophase I appears unaffected by loss of PCH2 based on immunolocalization of the axis proteins and axis length measurements. This differs from rice where the PCH2 ortholog, CRC1, is required for recruitment of the ASY1 ortholog PAIR2 onto the chromosome axes at leptotene [38]. This difference between the two plant species is perhaps surprising but it is not the first example where the phenotype of a rice meiotic mutant is different to that in other plants. For instance, loss of ZYP1 in Arabidopsis and barley results in a reduction of CO formation whereas mutation of the corresponding rice gene, ZEP1, leads to increased COs [20,65,66].
At mid-prophase I in Arabidopsis and B. oleracea, PCH2 forms foci along the SC which correlate with regions of ASY1 signal depletion on the axes. The overall distribution of PCH2, together with the fact that ASY1 signal intensity is not reduced along the synapsed axes in Atpch2 mutants, suggests that PCH2 participates in the depletion of ASY1 from the axis at the leptotene/zygotene transition. This could be a direct effect since biochemical studies in budding yeast show that Pch2 can bind to Hop1 in vitro and binding is strongly enhanced if its ATP hydrolysis activity is blocked [67]. In addition, Pch2 was shown to displace Hop1 from double-stranded DNA. Direct interaction in vivo has not been established as it is argued that this would be transient in the presence of ATP [67]. Based on the number of peptides recovered, PCH2 is found as an abundant component of a complex that is co-precipitated with ASY1 from Brassica PMCs. Although this could reflect a direct interaction, PCH2 may be co-precipitated as part of a larger chromosome axis-protein complex. Thus, an alternative possibility is that the reduction in the ASY1 signal is an indirect consequence of PCH2-dependent reorganization of the chromosome axis at the onset of zygotene.
In rice, loss of the PCH2 ortholog, CRC1 leads to a failure to form SC. This is unsurprising given that DSBs are not formed in a crc1 mutant [38]. Nevertheless, studies indicate that CRC1 localizes to the central region of the SC at pachytene and interacts with the SC transverse filament protein ZEP1 in a yeast two-hybrid assay, suggesting it is a component of the SC [38]. Analysis of the Atpch2 mutants indicates that PCH2 plays a critical role in formation of the SC, since loss of the protein results in a substantial defect in polymerization of the SC transverse filament protein ZYP1. An average reduction in SC length of 68% was observed but this was quite variable ranging from 43% to 87%. Similar to rice, co-localization between PCH2 and ZYP1 is also observed from the beginning of zygotene through pachytene. In Atasy1 and Atasy3 mutants, where SC polymerization is compromised, PCH2 is associated with the residual ZYP1 signal. Association of ZYP1 and PCH2 is also supported by SIM analysis of the B. oleracea SC as it begins to extend, although this suggests that they are not forming a homogeneous complex. This could reflect that any interaction between the proteins is transient. Since ASY1 appears to be the target for PCH2, it is conceivable that ZYP1 or another component of the SC central region acts to couple/guide the PCH2/ASY1 interaction. A precedent for this is seen in the bacterial P1 plasmid partitioning system in which the ParA ATPase is functionally coupled by the ParB protein to move its plasmid DNA cargo via a diffusion-rachet mechanism [68]. Furthermore the interaction between the C. elegans PCH2 ortholog, PCH-2, and the HORMAD spindle checkpoint protein Mad2, has been shown to involve an adaptor protein, p31 [69].
The SC nucleations seen in B. oleracea were consistent in size and their arrowhead-like shape is likely a consequence of the convergence of the homolog axes at the synaptic site. It is noteworthy that in most nuclei examined the number of large foci is broadly similar to the chiasma frequency (13–15) in B. oleracea [70,71]. Moreover most of the arrowhead-like ZYP1 foci (86.0%) co-localized with HEI10, a further indication that they occur at designated CO sites. Nuclei with fewer large foci may have been at a slightly earlier stage and reflect the dynamic nature of the initial appearance of foci. The smaller, slightly more numerous ZYP1 foci that were also present at early zygotene are likely to be additional synapsis initiation sites. The apparent existence of two classes of SC nucleation structures is reminiscent of observations in S. macrospora [72]. These have revealed distinct types of designations, one defining SC nucleation sites that correspond to CO designated recombination events and another that defines a similar number of sites where SC nucleation alone occurs. Importantly, the distribution of both classes exhibit interference and fits the prediction of the ‘beam-film’ model [11,12]. This posits that mechanical stress arises within a chromatin-axis meshwork as a result of global chromatin expansion during leptotene. Subsequent bi-directional relief of this stress results in a set of CO designations (and SC nucleations) that are spatially separated along the chromosomes. Further studies will be required to establish if the observed SC nucleations in B. oleracea also reflect a corresponding underlying interference-dependent distribution.
In other species loss of PCH2 orthologs leads to a variety of different effects on synapsis. In mouse, mutation of the Pch2 ortholog TRIP13 also results in a synaptic defect, albeit less severe than in Arabidopsis, with the unsynapsed regions accounting for just under 30% of the total axis length [34]. Loss of Pch2 in budding yeast does not appear to affect SC formation [27]. However, budding yeast forms high levels of COs and each designated CO site is thought to nucleate SC formation [73]. Hence, loss of Pch2 may not impact on SC installation to the degree observed in Arabidopsis where the relative CO rates are far lower. PCH2 also impacts on SC formation in C. elegans but in this case SC formation occurs more quickly than in wild type [37]. Interestingly, this defect was suppressed at lower temperatures. Thus loss of Pch2 has differing effects on the extent of SC polymerization in different species but in each case is associated with a recombination defect. Together, these observations suggest that PCH2 is not an integral structural component of the SC and more likely, regulates the coordination of synapsis with the controlled formation of COs.
The controlled formation of COs via homologous recombination is an essential feature of meiosis. Studies of PCH2 in several species have linked loss of the protein to a variety of recombination defects. In the most severe case, loss of the rice PCH2 ortholog, CRC1, is reported to result in a failure to form DSBs [38]. In budding yeast, a minor role for Pch2 in DSB formation has been reported [74]. It is also involved in processing of early occurring, low abundance DSBs and loss of the protein leads to a coordinate delay in the repair of DSBs to form both CO and NCO products [27,28].
In mouse, studies suggest that a severe reduction in TRIP13 expression does not compromise DSB formation but loading of RAD51 onto the resected DSBs is reduced [34]. In Arabidopsis, immunolocalization of RAD51 and DMC1 in Atpch2-1 PMCs indicated that early stages in recombination occur normally. As there is no evidence of chromosome fragmentation, it seems DSBs are also repaired, albeit with a reduction in CO formation, but progression through prophase I is delayed by 5-8h. This is reminiscent of that seen in some meiotic mutants and is indicative of an underlying defect in the recombination pathway [20,43,64]. Since a significant reduction in CO frequency was observed in an Atpch2/Atmsh5 double mutant relative to an Atmsh5 mutant it appears that loss of PCH2 impacts on the formation of both Class I and Class II COs.
Studies in different species have reported a CO interference defect associated with mutation of Pch2/TRIP13. Genetic analysis in budding yeast using intervals across a range of chromosomes of different sizes has revealed an increased frequency of closely spaced double-CO events in the absence of Pch2 [28,31]. In a Trip13 hypomorphic mutant mouse, despite an overall reduction in MLH1 foci at pachytene, a small, yet significant, reduction in the mean inter-focus distance between pairs of foci was observed. This implies that although the COs remain subject to interference, there has been some weakening in its effect, although a subtle change in the positioning of the DSB complexes cannot be excluded [34]. Despite these observations recent evidence from budding yeast has found that inter-focus distance of Zip3 foci that mark future COs is not affected by loss of Pch2, indicating CO interference is normal [32]. Why the discrepancy? Zip3 foci are the earliest known marker of CO designation, appearing in late leptotene. However, maturation of designated intermediates to form COs is dependent on additional later events during the remainder of prophase I [5]. Other analyses of pch2 mutants have used genetic markers or MLH1 foci, which mark mature CO sites. Hence it is conceivable that while loss of Pch2/Trip13 affects the final CO patterning, CO designation occurs and hence interference is initially established. Analysis of Atpch2-1 is consistent with this possibility. The localization of HEI10 foci in wild type and Atpch2-1 at early prophase I was identical. Numerous small axis-associated foci were observed together with around 10 large (~250 nm) foci. At present, it is not technically possible to measure inter-focus distance at early prophase I in Arabidopsis, nevertheless inspection of the nuclei reveals these large HEI10 foci are usually spatially well separated. These are still observed at mid/late prophase I when a similar number of MLH1 foci, which mark interference sensitive CO sites, are also observed. By analogy with budding yeast and S. macrospora where the appearance of Zip3/Hei10 foci are indicative of CO designation in early prophase I, it seems likely this is also the case in Arabidopsis as the number of HEI10 foci at the leptotene/zygotene transition appeared normal in the absence of PCH2. However, the maturation of the CO designated intermediates is compromised by the defect in remodelling of the chromosomes axes in Atpch2-1, leading to a deficit in COs. Overall, our data imply that in Arabidopsis, as in budding yeast and S. macrospora, CO designation and interference arise, and are complete, during zygotene.
We noted that in both wild type and Atpch2-1 some MLH1 foci appeared adjacent to the ZYP1 signal rather than directly over it. This has not been previously recorded in Arabidopsis. It may be a consequence of the spreading procedure but it is worth noting that in this study, the anti-ZYP1 antibody was raised to the N-terminus of the protein which is predicted to mark the central region of the SC. This could suggest the MLH1 containing complexes are not in direct contact with the SC central region. However the basis and significance of this remains unclear.
Other features of the CO distribution in the mutants indicate that CO interference is established. A predicted outcome of CO interference is that the numerical distribution of interference-sensitive COs between nuclei does not fit a Poisson distribution, whereas the converse applies for non-interfering COs [56]. The distribution of COs in Atpch2-1 does not fit a Poisson distribution, suggesting that COs remain subject to spatial patterning and do not arise by the random maturation of a proportion of the recombination initiations into COs. Also, there was a strong tendency for any HEI10 or MLH1 foci that were found associated with stretches of SC in Atpch2-1 to occur as single foci.
Although the data indicate that CO designation occurs normally, it seems that precursor maturation to form CO products is perturbed in Atpch2. This is manifested in several ways. Most obviously, the mean chiasma frequency in Atpch2 mutants is ~7, a reduction of around 30% relative to wild type. This is accompanied by the presence of univalents at metaphase I at a frequency of ~10%. A global reduction in CO formation has also been reported in TRIP13/Pch2-deficient mice and PCH2-deficient C. elegans [34,37]. In budding yeast, an increase in COs has been reported for some genetic intervals whereas in others wild type levels were recorded [31]. Also, the distribution of MLH1 foci in the mouse Trip13 mutants suggests there are chromosomal regions which show an increase in CO frequency [34]. This could suggest variation between different species but it is worth noting that despite the global reduction in COs in the absence of PCH2 an increase in recombination frequency was observed in 3 out of 6 intervals (l2f, I2g and 15b) in the Arabidopsis FTL lines used in this study. However, this comes with the caveat that this approach scores only viable tetrads which could influence the analysis. Estimation of genetic CO interference using the FTL lines suggested its effect may be diminished in at least some chromosomal regions, since a reduction in strength was detected over regions of chromosome 2 and chromosome 5. This apparent contradiction with the cytological evidence can perhaps be reconciled by data from a study of CO patterning in S. macrospora applying the beam-film model to experimental data. This showed that under some circumstances a normal interference signal is established and remains, yet CO interference as measured using CoC as a metric appears to be reduced [7,72].
An altered pattern of COs combined with a synaptic defect and a reduction in genetic interference has been reported in a kinesin mutant, Atpss1, and Ataxr1, a mutant in the E1 enzyme Arabidopsis neddylation complex [55,75]. Both mutants are strongly defective in synapsis with univalents observed at metaphase I. In each case HEI10 and MLH1 foci are observed in late prophase I in approximately wild type numbers but, in contrast to Atpch2-1, often clustered along the limited stretches of SC that have formed. An effect on the distribution of MLH1 foci has also been reported in as1, an asynaptic mutant of tomato [76]. The genetic basis of the as1 mutation is unknown but it is associated with changes in compaction of the chromosome axes. Relative to wild type, the average SC length in as1 was reduced by 81% with MLH1 inter-focus distance decreased by 71%. However the median number of MLH1 foci was unchanged, although the range was more variable. It is hypothesized that the tendency of these plant mutants to maintain CO numbers may reflect a homeostatic mechanism [76]. This is not so obvious in Atpch2-1 but it was notable that the mean reduction in CO frequency (~30%) was not-coordinate with that in SC length (~68%).
This study demonstrates that in the absence of PCH2, remodelling of the chromosome axis at zygotene and the normal patterned maturation of CO designated intermediates in Arabidopsis are aberrant. This further emphasises the functional inter-relationship between the chromosome axis and the controlled formation of COs.
A. thaliana ecotype Columbia (0) was used for wild type analysis. T-DNA insertion lines Atpch2-1: SAIL_1187_C06, Atpch2-2: SALK_031449 and Atpch2-3: SALK_130138 were obtained from NASC for mutant analysis. Plants were grown, material harvested and nucleic acid extractions were performed as previously described by Higgins et al. [43].
AtPCH2 peptides were identified by mass spectrometry in protein extracts from Brassica oleracea var. alboglabra A12DHd PMCs following co-immunoprecipitation with affinity purified anti-ASY1 antibody as previously described [40].
The T-DNA insertion site of the mutant lines was confirmed as previously described [43]. Details of the primers used are presented in S3 Table.
RNA extraction and RT-PCR was carried out as previously described [43]. Details of the primers are given in S3 Table.
Nucleotide sequencing was carried out by the Genomics and Proteomics Unit, School of Biosciences, University of Birmingham, UK.
An anti-PCH2 antibody was raised in rabbit against a 15-residue peptide from the C-terminus of Arabidopsis PCH2 (Abmart Inc., Shanghai, China). Due to the high level of sequence identity between the PCH2 proteins in Arabidopsis and Brassica the antibody was also effective for immunolocalization in Brassica.
Cytological studies were carried out as previously described [43]. The following antibodies were used: anti-AtPCH2 (rat 1/200 dilution), anti-AtASY3 (rabbit, 1/200 dilution) [19], anti-AtASY1 (rabbit/rat, 1/1000 dilution) [45], anti-AtMSH4 (rabbit, 1/500 dilution) [43], anti-AtZYP1 (N-terminus Ab aa residues 1–415; C-terminus Ab aa residues 422–845; rabbit/rat, 1/500 dilution), anti-AtRAD51 (rabbit 1/500 dilution), anti-AtSYN1 (rabbit 1/500 dilution), anti-AtDMC1 (rabbit 1/500 dilution) [20,23], anti-AtMLH1 (rabbit/rat, 1/200 dilution) [64], anti-AtHEI10 (rabbit 1/500 dilution) and anti-γH2AX (ser 139, catalog no. 07–164 Upstate Biotechnology; rabbit, 1/100 dilution). Microscopy was carried out using a Nikon 90i Fluorescence Microscope (Tokyo, Japan). Image capture, image analysis and processing were conducted using NIS-Elements-F software (Nikon, Tokyo, Japan) as previously described [19]. Image deconvolution was carried out using the function “Mexican hat”. This allows better discrimination of the signals. This function performs filtration on the intensity component (or on every selected component—when working with multichannel images) of an image using convolution with 5x5 kernel. Mexican Hat kernel is defined as a combination of Laplacian kernel and Gaussian kernel it marks edges and also reduces noise. SIM was carried out using the OMX facility at the University of Dundee (http://microscopy.lifesci.dundee.ac.uk/omx/).
In Arabidopsis, ASY1 intensity analysis was conducted on chromosome spread preparations stained with anti-ASY1 antibody (rat, 1 in 5000 dilution) and anti-ZYP1 (rabbit, 1 in 500 dilution). 5μl of 6μm, 0.3% relative intensity InSpeck Red microspheres (Life Technologies), were added to slides before coverslips. PMCs and microspheres were imaged using specific exposure times. Randomly selected, non-overlapping sections of axis, ~2–4μm in length, were defined as regions of interest and were analysed for mean signal intensity using Nikon NIS-elements software. Intensities were normalised based on mean intensity of the microspheres. Intensity raw data is shown in grey-scale values. For B. oleracea, ASY1 intensity was determined in on PMC chromosome spreads at zygotene comparing non-overlapping segments of unsynapsed and synapsed sections of axis ~2–4μm in length.
Chiasma counts were carried out as previously described [42]. Chromosome spread preparations from PMCs at metaphase I were examined by light microscopy after fluorescence in situ hybridization (FISH) using 45S and 5S rDNA probes. The use of FISH enabled the identification of individual chromosomes. The overall shape of individual bivalents allowed the number and position of individual chiasmata to be determined and this was also informed by the position of the FISH signals.
The time course of progress through prophase I in wild type and Atpch2-1 was determined as previously described [48] except that 5-ethynyl-2’-deoxyuridine (EdU) was used to label the PMCs which were analyzed at 5 h intervals from 0–30 h and at 2 h intervals thereafter up to 36 h.
Fluorescent tetrad analysis was carried out as described Berchowitz and Copenhaver [49] using genetic intervals I2f and I2g on chromosome 2 (FTL coordinates for the I2fg interval: FTL#800 18286716 bp DsRed2; FTL#3411 18957093 bp YFP; FTL#3263 19373634 bp AmCyan) and I5a, I5b, I5c, and I5d on chromosome 5 as described [49]. Pollen was scored through eCFP, eYFP and DsRed2 filters using an Olympus BX-61 epifluorescence microscope. The Stahl Lab Online Tools (http://molbio.uoregon.edu/~fstahl/) was used for statistical analyses of the data.
The statistical procedures were carried out as described previously [43]. Chi-squared (Χ2) tests were used to determine agreement between the observed chiasma counts and those expected from a Poisson distribution. Numbers of foci in wild type and mutant PMCs were compared using the Wilcoxon signed-rank test. Mean intensities between synapsed and unsynapsed sections of axis/SC were analysed using a 2 tail paired T-test.
|
10.1371/journal.ppat.1005774 | Elevated Basal Pre-infection CXCL10 in Plasma and in the Small Intestine after Infection Are Associated with More Rapid HIV/SIV Disease Onset | Elevated blood CXCL10/IP-10 levels during primary HIV-1 infection (PHI) were described as an independent marker of rapid disease onset, more robust than peak viremia or CD4 cell nadir. IP-10 enhances the recruitment of CXCR3+ cells, which include major HIV-target cells, raising the question if it promotes the establishment of viral reservoirs. We analyzed data from four cohorts of HIV+ patients, allowing us to study IP-10 levels before infection (Amsterdam cohort), as well as during controlled and uncontrolled viremia (ANRS cohorts). We also addressed IP-10 expression levels with regards to lymphoid tissues (LT) and blood viral reservoirs in patients and non-human primates. Pre-existing elevated IP-10 levels but not sCD63 associated with rapid CD4 T-cell loss upon HIV-1 infection. During PHI, IP-10 levels and to a lesser level IL-18 correlated with cell-associated HIV DNA, while 26 other inflammatory soluble markers did not. IP-10 levels tended to differ between HIV controllers with detectable and undetectable viremia. IP-10 was increased in SIV-exposed aviremic macaques with detectable SIV DNA in tissues. IP-10 mRNA was produced at higher levels in the small intestine than in colon or rectum. Jejunal IP-10+ cells corresponded to numerous small and round CD68neg cells as well as to macrophages. Blood IP-10 response negatively correlated with RORC (Th17 marker) gene expression in the small intestine. CXCR3 expression was higher on memory CD4+ T cells than any other immune cells. CD4 T cells from chronically infected animals expressed extremely high levels of intra-cellular CXCR3 suggesting internalization after ligand recognition. Elevated systemic IP-10 levels before infection associated with rapid disease progression. Systemic IP-10 during PHI correlated with HIV DNA. IP-10 production was regionalized in the intestine during early SIV infection and CD68+ and CD68neg haematopoietic cells in the small intestine appeared to be the major source of IP-10.
| Chronic immune activation is a hallmark of HIV infection and contributes in multiple ways to HIV persistence. Here, we gained insights on the association between a pro-inflammatory chemokine, CXCL10/IP-10 and HIV infection in four cohorts of HIV+ individuals, studied at distinct stages of infection (before, primary and chronic stage with spontaneous- and treatment-controlled infection). We further analyzed pathogenic and non-pathogenic SIV infections to address IP-10 levels and the presence of infected cells in tissues (lymph nodes, small and large intestine). We found that elevated systemic IP-10 levels before HIV-1 infection associate with a more rapid disease progression. During primary infection, IP-10 in blood strongly correlated with the amount of infected cells in blood. The animal model showed that IP-10 expression was regionalized in the intestine and highest in the small intestine. Studies of aviremic animals suggest that high IP-10 is indicative of viral replication in lymphoid tissues. Haematopoietic cells rather than epithelial/endothelial cells mainly contributed to the IP-10 production in small intestine (jejunum). The receptor of IP-10 was highly expressed on memory CD4+ T cells, i.e. major target cells for the virus. This study contributes to our understanding of the establishment of HIV reservoirs and why IP-10 associates with HIV/AIDS.
| Chronic immune activation is a hallmark of HIV infection [1]. Effective combined-antiretroviral therapy (cART) reduces HIV viremia to undetectable levels, but milder chronic immune activation nonetheless persists and is associated with onset of both AIDS and non-AIDS illnesses [2, 3]. The mechanisms fuelling chronic inflammation in HIV infection are poorly understood and probably multifactorial. Translocation of microbial products from the gastrointestinal tract may be an important driving factor [4–6, 7, 8].
Studies of SIV+ non-human primates (NHP) such as Asian macaques (MAC) and natural hosts of SIV such as African green monkeys (AGM) support a role of immune activation and microbial translocation in HIV pathogenesis [1, 5, 6, 7, 8–15]. SIV infection in natural hosts is characterized by high viral load but does not result in chronic inflammation [1, 15, 16]. Strong inflammatory responses are only transient in natural hosts and by the end of the primary phase of infection, they are dampened to pre-infection levels [1, 9–15]. We thus asked whether HIV-infected individuals with only weak inflammation near the end of primary HIV infection (PHI) have better outcomes [17]. High inflammation level at Fiebig stages III and IV of PHI was indeed associated with rapid loss of CD4+ T-cells. Among 28 pro-inflammatory factors tested, CXCL10/IP-10 was a strong and independent predictive marker of rapid CD4+ T-cell loss [17]. During PHI, IP-10 was even a more robust predictive marker than viremia or the CD4+ T-cell nadir.
Many cells can produce and release IP-10 [18]. During HIV-1 infection, circulating myeloid cells are the main source of IP-10 in blood [19]. In secondary lymphoid organs from SIV+ macaques, IP-10 is mainly produced by CD3+ T-cells, but also by CD14+ and CD3-CD14- cells [18, 20, 21]. IP-10 is a pro-inflammatory chemokine and a ligand of CXCR3. As CXCR3+ CD4+ T-cells are the main cellular targets of HIV [22, 23], it is conceivable that IP-10 enhances the trafficking of HIV target cells to lymphoid tissues, thereby promoting new rounds of infection and helping to establish viral reservoirs [24].
Here we attempted to gain further insights into the role of IP-10 during HIV-induced inflammation and its relationship with the levels of infected cells.
We raise the hypothesis that IP-10 attracts target cells for HIV. First, we tested if IP-10 pre-infection levels impact on infection outcome. We quantified IP-10 in blood from 136 patients before HIV infection, during PHI, 3 months (M3) after seroconversion (SC) and/or 6 months (M6) after SC in the Amsterdam cohort (ACS) (characteristics described in Table A in S1 Text). Plasma IP-10 levels were higher during PHI (p<0.05) than before infection, and remained elevated at M3 and M6, albeit at lower levels (p<0.01) (Fig 1A). Similar IP-10 profiles were observed in the subset of 16 patients from whom samples were available at every time point (Fig 1B). IP-10 levels correlated negatively with the CD4 T-cells count [r = -0.19 (p = 0.04) and r = -0.39 (p<0.001) at M3 and M6, respectively], and positively with viremia in PHI (Fig 1C), at M3 (Fig 1D).
Strikingly, elevated IP-10 levels before infection were associated with rapid progression (OR = 3.24 p = 0.01) (Table B in S1 Text), CD4+ T-cell counts falling below 350/mm3 more rapidly in individuals with pre-infection IP-10 levels above the median (Fig 1E). In contrast, CD4 T-cell counts before infection had no impact on the rate of progression. Of note, only IP-10 levels measured less than 24 months before infection had such an impact: IP-10 levels measured between 24 and 60 months before infection did not influence the rate of CD4 T-cells loss. IP-10 concentrations in blood before infection were not correlated to canonical immune activation markers (Table C in S1 Text).
Elevated IP-10 levels at M3 post-SC were also associated with rapid CD4+ T-cell decline to below 350/mm3 (Fig 1F) and with an increased risk of rapid progression toward AIDS (OR = 2.54 p = 0.02) before any treatment (Table B in S1 Text). However, at M3, the CD4 T-cell count was a more robust predictor of rapid progression than were IP-10 levels, while both the IP-10 level and the CD4 T-cell counts were more robust predictors than viremia (Table B in S1 Text).
In order to compare the robustness of IP-10 in predicting rapid disease onset to other systemic markers of HIV-induced immune activation, we also measured soluble CD163 (sCD163), a monocyte-macrophage activation marker associated with disease progression in HIV+ individuals [25, 26]. sCD163 concentrations in blood were strongly correlated to IP-10 concentrations at all times studied here (Fig 2). However, sCD163 concentrations before and after infection failed to be associated with rapid disease progression (Table B in S1 Text), in contrast to IP-10.
Thus, by analyzing the time course of IP-10 levels, starting before infection, we found that they rose markedly upon HIV infection. Strikingly, pre-existing elevated IP-10 levels were associated with rapid CD4 T-cells loss upon HIV-1 acquisition.
To further test the hypothesis that IP-10, by promoting CXCR3+ cell trafficking, attracts HIV target cells and thereby increases the number of infected cells, we examined the relationship between the IP-10 level and the number of infected cells. We measured cell-associated total viral DNA in order to include both latently infected cells and cells supporting active viral replication. These HIV DNA levels in PBMC from 134 subjects with PHI from the ANRS PRIMO cohort have previously been reported [17]. IP-10 levels strongly correlated with cell-associated HIV-1 DNA (Fig 3).
We also compared the levels of 27 additional inflammatory molecules with the amount of infected cells in the same patients. These markers had been previously analyzed [17]. Among the 27 markers, only IL-18 also positively correlated with total HIV DNA in PBMC (r = 0.3 p = 0.045; Table D in S1 Text), but to a lesser extent than IP-10 (r = 0.35 p<0.0001, Fig 3B). RANTES was the only factor, which had a negative association with total HIV DNA (r = -0.4 p = 0.007; Table D in S1 Text). Altogether Blood IP-10 levels positively correlate with cell-associated viral DNA during in PHI.
We then examined IP-10 levels in various groups of patients with controlled viremia (Fig 4A and 4B). We first studied HIV controllers (n = 82), divided into two groups: individuals with strong viral control at the time of IP-10 assay (<50 copies of viral RNA/ml) and individuals displaying a moderate viral “blip” at this time point (>50 copies). IP-10 levels in all 82 controllers together were similar to those in the patients on successful cART. However, IP-10 levels tended to be lower in the HIV controllers with <50 copies/ml than in the HIV controllers with >50 copies/ml and cART-treated patients.
Next, we evaluated IP-10 before and during cART (n = 41) (Table E in S1 Text). IP-10 levels fell significantly during cART as compared to pre-treatment levels, but remained higher than in uninfected controls (Fig 4A). Before treatment, IP-10 levels correlated positively with viremia (r = 0.32 p<0.001) and negatively with the CD4 T-cell counts (r = -0.26 p = 0.003). During cART, IP-10 levels correlated negatively with the CD4 T-cell counts (Fig 4C), as strongly as in viremic patients. The recent TEMPRANO and START trials demonstrated that AIDS and non-AIDS events can occur even in patients with high CD4 counts (>500/mm3) [27]. When we considered only patients with CD4 T-cells counts above 500/mm3, there was still a trend towards a negative correlation between IP-10 and CD4 T-cell counts (Fig 4D).
To further determine the relationship between IP-10 and viral replication in a context of viral control, we quantified IP-10 in SIVmac-exposed aviremic macaques. We chose a non-human primate model because it allowed us to investigate viral load in tissues. We studied 34 MAC exposed to moderate doses of SIVmac251. Twenty-seven MAC displayed an expected viremic phenotype. From 3 of these viremic animals, we FACS-sorted the LN cells into four subsets: CD4+ and CD4- T cells expressing or not CXCR3 (Figure A in S1 Text). The viral DNA copy numbers in CD4+ LN cells were high as expected (Figure A in S1 Text). No significant difference could be observed between CXCR3+ and CXCR3neg CD4+ T cells. The samples that were positive for viral DNA were also those positive for IP-10 gene expression (Figure A in S1 Text).
Surprisingly, 7 of these 34 animals remained aviremic (< 12 copies of SIVmac RNA/ml) during the chronic phase of infection (follow up until 1 year p.i.), despite the fact that the viral doses used resulted in 100% of infection in previous experiments [28]. None of the aviremic animals seroconverted during follow-up. We quantified total SIV DNA in total LN cells from the SIV-exposed aviremic macaques. PCR amplification of cell-associated DNA on total LN cells was negative in 5 of the 7 aviremic animals, while positive (4–35 copies/106 cells) in 2 animals (AX414 and 30845) on day 14 p.i.
We then compared plasma IP-10 dynamics in three groups of animals (viremic, aviremic with or without detectable viral DNA in tissues). IP-10 levels were elevated in viremic animals, as expected (Fig 5A). Viremia and IP-10 levels correlated positively with one another (r = 0.69, p<0.0001). In the 5 aviremic animals with no detectable viral DNA in tissues (LN), IP-10 levels remained low during primary infection (Fig 5B). In contrast, IP-10 levels increased in the 2 animals (AX414 and 30845) with detectable SIV DNA during primary infection (Fig 5B). We then compared levels in the aviremic animals with those in 3 MAC in which cART was initiated 4 h after infectious challenge [29]. The animals were treated until day 14 p.i. and then sacrificed. None of these animals had detectable viral DNA in lymphoid tissues during follow-up [29]. These cART-treated animals displayed a weaker induction of IP-10 during follow-up as compared to viremic animals (Fig 5C).
Together, these observations support a strong association between blood IP-10 levels and active viral reservoirs in lymphoid tissues.
Since IP-10 triggers the trafficking of cells expressing its ligand, CXCR3, we sought to address the dynamics of CXCR3+ CD4+ T cells in vivo in parallel to the IP-10 response upon SIV infection. We first determined the levels of CXCR3 expression by flow cytometry on distinct haematopoietic cells in blood before SIV infection in rhesus macaques and AGMs. Strikingly, CXCR3 was mostly expressed on CD4+ T cells when compared to other immune cells (Fig 6A). CXCR3 expression was also significantly higher on CD4+ T cells than on CD8+ T cells.
We then assessed the dynamics of CXCR3+ within the naïve (CD28+ CD95neg) and memory (CD28+CD95+ and CD28-CD95+) CD4+ T cells in LN. This memory cell fraction typically contains the vast majority central memory CD4+ T cells. We found that the frequency of CXCR3+ cells was higher within the memory subset than within naïve CD4+ T cells. Both in macaques and AGM, these cell subsets quickly exhibited changes in their frequencies upon infection (Fig 6C and 6D). The frequencies of CXCR3+CD4+ T cells tended to increase at day 2 p.i. in macaques (p = 0.0625). In the contrary, AGMs exhibited decreased frequencies of CXCR3-expressing cells in memory CD4+ T cells at days 2 and 9 p.i. (p = 0.0313) (Fig 6C and 6D).
All AGMs and one macaque displayed a peak of IP-10 at days 2 and 9 p.i. (Fig 6B). We therefore wondered whether decreases in CXCR3 expression could be related to internalization of the receptor as described for the IL-7R in HIV/SIV infections [30]. To address this hypothesis we evaluated the rate of CXCR3 internalization in LN CD4 T cells from 3 SIVmac-infected macaques and 3 SIVagm-infected AGMs (Fig 6E and 6F). In both species, we observed in the infected animals high proportion of LN CD4 T cells having internalized CXCR3. The levels of CXCR3 in the permeabilized cells varied between 72 and 92% in macaques and between 45 and 60% in AGMs.
Altogether, CXCR3 was mostly expressed on CD4+ T cells when compared to other immune cells and within CD4+ T cells, more frequently expressed on memory than on naïve cells. Modulations in the frequencies of CXCR3+ CD4+ T cells were observed upon infection. Significant decreases might in part be explained by internalization of the CXCR3 receptor in response to the increased production of its ligands (IP-10 or other CXCR3 ligands) during infection.
To determine the origin of IP-10 in blood, we measured IP-10 expression in the largest lymphoid tissue: the intestine. The latter is also the largest site of HIV/SIV replication. We analyzed intestinal IP-10 production in 5 SIVmac-infected rhesus macaques. We also compared the IP-10 expression pattern in a non-pathogenic model (5 SIVagm-infected AGMs). AGMs were used because they are known to display a high level of viral replication in the gut, in the absence of chronic inflammation [7, 31]. Furthermore, as the intestine is regionalized into several sections with specific functions [32], and as we found higher infection rate of the small intestine than in the large intestine in SIVmac-infected cynomolgus macaques (Figure B in S1 Text), we analyzed the IP-10 gene expression in 4 distinct sections, i.e. jejunum, ileum, colon and rectum.
During SIVmac infection, IP-10 was strongly expressed by CD4+ cells in the small intestine (jejunum/ileum) (Fig 7A), and more strongly than by CD4- cells in the small intestine or both CD4+ and CD4- cells in the large intestine. No such regionalization of IP-10 was seen in AGMs. Likewise, CXCR3 expression was the strongest in CD4+ cells of the small intestine during SIVmac infection, whereas this profile was not observed in AGMs (Fig 7B). IP-10 expression correlated with CXCR3 expression (r = 0.66, p = 0.0009).
IP-10 expression levels in CD4+ cells of the small intestine correlated with plasma levels when the two species were grouped for analysis (Fig 7C). When we considered only MAC, which reduced the statistical power, there still was a trend towards a positive correlation (Fig 7C). No such correlation was found in the colon/rectum.
Thus, the strongest IP-10 expression was detected in the CD4+ fraction in the small intestine in SIVmac-infected macaques. Moreover, IP-10 levels in plasma tended to correlate with those in the small intestine.
We then sought to determine which CD4+ cell subset was responsible for elevated IP-10 expression in the small intestine. It has been described that IP-10 is produced by monocytes and macrophages during HIV/SIVmac infection [19, 33–35]. We therefore quantified the expression of genes associated with macrophages (CD14, CD68 and CD163) in the same samples as those analyzed above for IP-10 mRNA. All three macrophage markers (CD14, CD68 and CD163) were more strongly expressed in small-intestinal CD4+ leukocytes from MAC than from AGM, or in any other gut section from MAC (Fig 7D–7F). IP-10 expression levels correlated strongly with these gene expression levels (Fig 7G–7I).
To address whether IP-10 is produced at the protein level in gut macrophages, we performed immunohistochemistral stainings in jejunum fragments of macaques harvested at necropsy (day 240 p.i). IP-10 expression was detected throughout the jejunum in cells that had a shape and localization distinct from epithelial/endothelial cells (Fig 8A). We found a massive expression of IP-10 from small and round CD68neg cells, which ressembled in their shape lymphocytes (Fig 8B) and to a lesser extent from large CD68+ cells (Fig 8C). Very often CD68+IP-10+ cells were found in clusters at top of villi (Fig 8D). These data demonstrate that IP-10+ is produced by macrophages in the jejunum. However many smaller, CD68neg cells produced IP-10 as well.
To further analyze the link between IP-10, macrophages and intestinal inflammation, we quantified the expression of ISGs such as MX1 and IFI30. IP-10 expression was positively correlated to MX1 and IFI30 gene expression. (Figure C in S1 Text). Although these ISGs were also correlated with the macrophage markers, these correlations were less robust than with IP-10 (Figure C in S1 Text).
During HIV-1/SIVmac infections, mucosal immunity in the intestine is generally gradually impaired, notably with a characteristic loss of Th17 cells, which is absent in natural hosts of SIV [5, 36, 37]. As a read-out for gut damage, we quantified the expression of RORC, the master transcription factor for Th17 cells, in enriched CD4+ leukocytes from the intestinal sections. We detected stronger RORC expression in CD4+ leukocytes from AGM than MAC (Fig 7J).
RORC and IP-10 expressions correlated negatively with one another in small-intestinal CD4+ leukocytes (r = -0.74, p = 0.0034) from all the studied NHP. There was also a trend towards a negative correlation when we considered only MAC (Fig 7K). A negative correlation between RORC in the small intestine and plasma IP-10 (Fig 7L) was observed.
Pathogenic SIVmac infection is characterized by a skewed Th response towards a Th1 phenotype in the gut at the detriment of Th17 cells, in contrast to natural hosts [38–40]. To determine if this negative correlation between IP-10 and RORC were associated with a particular profile of Th differentiation in the analyzed CD4+ cell population, we went on evaluating the IP-10 expression in Th subsets. We sorted circulating primary human Th subsets. We observed that primary human circulating Th17 cells are devoid of a strong IP-10 gene expression in sharp contrast to primary Th1-like cells which express the highest levels of transcriptional IP-10 gene activity (Figure D in S1 Text). Altogether these data indicate that IP-10 is associated with inflammation in the small intestine and suggest a negative association between the levels of IP-10 and Th17 cells.
Here, we studied the relevance of IP-10 as a marker of disease progression before infection, and examined why IP-10 is so strongly associated with HIV pathogenesis.
We found that pre-existing elevated IP-10 levels were associated with an increased risk of rapid CD4+T-cells loss upon HIV infection in the ACS. Elevated levels of IP-10 prior to infection may be multifactorial: (i) resulting from co-infections with viruses although we excluded co-infections with HIV-2, HBV and HCV but we couldn’t exclude a co-infection with TB [41]. Indeed this latter report showed that during reactivation of latent TB infection IP-10 is found at elevated concentrations in blood. We couldn’t completely rule out (ii) elevated immune activation [42]. However, when we analyzed additional immune activation/inflammation markers, such as Ki-67, CD8+DR+CD38+, CD8+CD70+, CD4+DR+CD38+, CD4+CD70+, before infection, we couldn’t see any significant impact of such markers on the rate of CD4+ T cells upon HIV-infection and none of them were significantly correlated to IP-10 pre-infection levels in blood (Table C in S1 Text). The caveat was the very low number of patients with documented immune activation markers before infection.
In addition, we analyzed sCD163 levels. They were significantly correlated to those of IP-10 but sCD163 levels failed to predict rapid disease onset in contrast to IP-10 pre-infection levels.
Lower IP-10 levels had been reported in the genital mucosa of highly HIV-exposed-seronegative women than in HIV-seronegative and -seropositive women [43, 44]. Blood IP-10 levels were recently reported to be higher in transmitting HIV-1–infected individuals and in their HIV-1–seroconverting partners than in HIV-1–infected and uninfected partners [45]. This suggested that elevated IP-10 increased the risk of HIV-1 acquisition. It is possible that this increased risk of HIV-1 acquisition, and the more rapid CD4+ T-cells loss observed here in individuals with higher IP-10 levels before infection, is due to IP-10 enhancement of the infection. We indeed found a strong correlation between IP-10 and the amount of cell-associated HIV DNA. Recently we have analyzed plasma IP-10 levels in the ANRS OPTIPRIM study, where patients received mega-ART therapy starting from PHI [46]. IP-10, but not IL-6, sCD14 nor sCD163 was positively correlated with blood HIV-DNA at inclusion (r = 0.53, p = 0.018) and only IP-10 levels among 5 inflammatory markers assessed in plasma correlated with total HIV DNA in semen (A. Cheret, personal communication). In our animal models, IP-10 levels helped to discriminate between SIV-exposed aviremic animals with and without detectable tissue infection. Several studies have shown that elevated IP-10 levels contribute to excessive recruitment of CXCR3+ T-cells into lymphoid tissues during pathogenic SIVmac infection [47, 48]. This was also described in important sites of viral entry, notably the foreskin of sexually active men with a high risk of acquiring HIV [49]. CXCR3 expression is highest in activated memory CD4+ T-cells [22]. It is possible that IP-10 attracts not only CXCR3+ immune cells with potential antiviral activity but also major HIV target cells, indirectly enhancing viral dissemination and the establishment of viral reservoirs. Recent studies have shown a high infection rate among CXCR3+ CD4+ T-cells [22], which are also preferentially enriched for HIV DNA in HIV-infected individuals on cART [23]. IP-10 has also been shown to enhance the susceptibility of resting naïve CD4 T-cells to HIV infection [50]. Here we observed distinct dynamics of CXCR3+ CD4 T cells on LN from SIV-infected macaques and natural hosts. These increased frequencies of CXCR3+ memory CD4 T cells in macaque LN versus reduced frequencies of CXCR3+ memory CD4 T cells in AGM LN during the acute phase of infection may be due to mobilization of these cells in the context of a strong pro-inflammatory context (CXCR3-ligands) in macaques. We could not see such dramatic increases in AGM and we can’t exclude that the CXCR3 was either internalized or that the CXCR3+ memory CD4+ T cells are migrating to another tissue in natural hosts. In macaques, we didn’t detect a higher rate of infection in CXCR3+ versus CXCR3neg CD4+ cells. As CXCR3 might be dramatically internalized upon recognition of its ligands during infection (Fig 6E and 6F), the interpretation of the viral distribution in the context of increased production of IP-10 (and other CXCR3 ligands) in vivo is complex. More animals and more studies need to be done at critical time points and in animals with distinct levels of CXCR3 ligands. Though, when we looked at the intestinal mucosa, we found a significant higher rate of infection in the small intestine where the expression of IP-10 and CXCR3 were the highest. Overall, our findings further demonstrate that elevated IP-10 levels are a strong predictive marker of disease progression but raise the question whether it might, directly and/or indirectly, also promote the establishment of viral reservoirs.
Our data in animals suggest that IP-10 in plasma derives in large part from the gut, the largest lymphoid tissue. Unexpectedly, the IP-10 response was regionalized in macaques, being higher in the small intestine than the large intestine. CXCR3 expression showed the same regionalization. Previous studies have shown marked regional variations in the abundance of infected cells in the gut, but most focused on specific compartments and not the entire gut [51, 52]. In cART-treated patients, the small intestine also seems to harbor more active viral reservoirs than the periphery or the rectum [53, 54]. Here we also observed a higher infection rate in the small intestine than in the large intestine, correlating with the IP-10/CXCR3 regionalized expression. Further, IP-10 mRNA expression in MAC small intestine positively and strongly correlated with the expression of macrophage-associated markers (CD14, CD68 and CD163 mRNA). This association can have several explanations. It might have an indirect cause as both IP-10 and macrophage levels in the gut might be the consequence of higher inflammation. It might also be explained by a higher number of macrophages in the gut or finally by an increase in production of IP-10 in the gut. IP-10 levels correlate with expression of the activation markers CD11b and CD38 on monocytes [55]. In the latter report, the sCD14 levels did not correlate with any of these molecules. The same authors suggest that IP-10, but not sCD14, is a robust and easier tool to measure monocyte activation [55]. Further, an accumulation of CD68+ and CD163+ macrophages in the duodenal mucosa of HIV-infected patients was reported in parallel to increased levels of pro-inflammatory molecules such as IP-10 [20]. By studying patients in the COPANA cohort, we found that plasma IP-10 correlated with the concentrations of monocytic activation markers sCD163 (r = 0.57 p<0.001), sCD14 (r = 0.35 p = 0.008) and TFN-α (r = 0.43 p<0.001), but not with CRP, IL-6, MCP1, sTNFR1 or sTNFR2 (Figure E in S1 Text). Infiltration of the small intestine (duodenum) of HIV-infected individuals by inflammatory CD68+/CD163+ macrophages has been reported in HIV+ individuals in absence of cART [20]. Thus, elevated IP-10 production in the small intestine might derive from infiltrating activated macrophages during progressive HIV/SIV infections. The IP-10, CD14, CD68 and CD163 gene expression in the small intestine of SIVmac-infected rhesus macaques seem to be in line with these observations in HIV+ individuals. We show here that macrophages are indeed one important source of IP-10 production in the jejunum SIV-infected rhesus macaques (Fig 7). This is consistent with previous reports in humans. During acute HIV-1 infection, IP-10 production in blood was found to be associated with circulating myeloid cells [19]. However, many other cells produced IP-10 as well, consistent with data in lymph nodes [18, 20, 21]. Indeed, IP-10 was also more often confined to CD68neg small and rounded-shape cells (Fig 8). These might include predominantly CD4+ lymphocytes since IP-10 mRNA expression levels where highest in the CD4+ cell fraction. However, we cannot exclude differences in the profiles of mRNA and cellular sources due to fact that the PCR and IHC analyses were performed in distinct animals and distinct time points (Day 65 p.i and D240 p.i., respectively). Finally IP-10 increases might result from both increases in cell numbers as well as from intra-cellular upregulations. Altogether, we determined the cellular source of IP-10 and demonstrate a regionalization of IP-10 production in the gut.
The small intestine in particular is remarkably enriched in IL-17 producing T cells [56]. A gradient of T-cell IL-17 production has indeed been reported along the intestinal tract, with the small intestine being enriched in such lymphocytes [56]. We found that IP-10 levels in both blood and the small intestine correlated negatively with the presence of Th17 in the small intestine, suggesting that blood IP-10 levels mirror the extent of gut damage. RORC expression in the small intestine was stronger in AGMs than in MAC, which supports the relevance of our model. We found that IP-10 mRNA expression in MAC small intestine negatively correlated with the presence of gut Th17 cells (Fig 7). IP-10 expression might be indirectly associated to intestinal inflammation and gut damage. Our observations from primary human Th subsets clearly confirm a confinement of IP-10 gene expression in genuine Th1 cells rather than Th17. Thus the negative correlation between IP-10 and RORC in our study could be due to the biased Th1 response at the detriment of Th17 response, which has been described in pathogenic SIVmac infection in contrast to natural hosts of SIVs [38–40].
Persistently moderate IP-10 levels were observed in HIC, as reported elsewhere [57, 58]. IP-10 seems to distinguish between HIC who experience viral blips (Fig 3B) and rapid loss of CD4+ T-cells [57, 58], and could prove useful for identifying those patients with controlled viremia, including HIC, who need therapeutic interventions to further delay disease progression.
In summary, this study of IP-10 in four cohorts of HIV-infected patients and in two non-human primate models provides new information on the tissue source of this pro-inflammatory mediator, reveals its regionalization in gut and indicates an association with the cell infection rate during HIV-1 infection.
Studies were conducted with ethical agreements and with the informed consent of each patient. Patient enrollment respects European guidelines and established guidance promulgated by the World Medical Association in its declaration of Helsinki. All patients were adult subjects. The scientific board of the Amsterdam cohort studies and of the French ANRS cohorts PRIMO C06, COPANA C09 and CODEX C21 approved this study. Institut Pasteur “Comité de recherché Clinique” CoRC #2013–05 approved this study.
Animals were housed in the facilities of the CEA (“Commissariat à l'Energie Atomique”, Fontenay-aux-Roses, France) IDMIT facilities (Center for Infectious Disease Models and Innovative Therapies), Fontenay-aux-Roses, France (permit number A 92-032-02) or the Pasteur Institute, Paris, France (permit number A 78-100-3). All experimental procedures were conducted in strict accordance with the European guideline 2010/63/UE for the protection of animals used for experimentation and other scientific purposes (French decree 2013–118) and with the recommendations of the Weatherall report. The monitoring of the animals was under the supervision of the veterinarians in charge of the animal facilities. All efforts were made to minimize suffering, including efforts to improve housing conditions and to provide enrichment opportunities (e.g., 12∶12 light dark schedule, provision of monkey biscuits supplemented with fresh fruit and constant water access, objects to manipulate, interaction with caregivers and research staff). All procedures were performed under anesthesia using 10 mg of ketamine per kg body weight. For deeper anesthesia required for lymph node removal a mixture of ketamine and xylazine was used. Paracetamol was given after the procedure. Euthanasia was performed prior to the development of any symptoms of disease (e.g., for macaques when the biological markers indicated progression towards disease, such as significant CD4+ T cell decline and increases of viremia). Euthanasia was done by IV injection of a lethal dose of pentobarbital.
A large serum library derived from cART-naïve patients is available, including pre-infection samples from many patients enrolled in the Amsterdam Cohort Studies (ACS) of HIV infection and AIDS. None of the 136 subjects studied here had started antiretroviral therapy when samples were collected. These patients had an estimated date of seroconversion (SC) defined as the midpoint between the date of the last visit with a negative HIV test and the first visit with a positive HIV test (complete or incomplete western blot) [59]. None of the 136 subjects had started antiretroviral therapy when samples were collected. Patients were categorized as rapid progressors (RP) or slow/normal progressors as previously described [17]. Basically, rapid progressors had CD4 T cell counts below 350/ml 12 months post-SC. PHI (M0) was defined by an incomplete WB, with detectable HIV RNA load and/or p24 (Fiebig stage III/IV). Samples collected before infection were obtained at least 3 months before the estimated date of SC. Patients co-infected with other bloodborn pathogens (HIV-2, HBV, HCV) were excluded.
The viremic cART-naive subjects (VIR, n = 121) were part of the French ANRS C09 COPANA cohort (See supplementary materials and Table C in S1 Text). A subgroup of 41 patients was submitted to cART (>24 months on cART and >12 months with VL < 50 copies/mL).
The HIV controllers (HIC, n = 82) were enrolled in the French ANRS C021 CODEX (See supplementary materials and Table C in S1 Text). The samples collected at PHI M0 (n = 126) and PHI M6 (n = 35) were from subjects enrolled in the French ANRS C06 PRIMO cohort, who are thoroughly described in [17].
The viremic cART-naive subjects (VIR, n = 121) were part of the French ANRS C09 COPANA cohort. The main objective of this ongoing cohort created in 2004 is to prospectively evaluate the impact of HIV infection and ART on morbidity and mortality in recently diagnosed (<1 year) HIV-1-infected cART-naive adults in France. The Paris-Cochin Ethics Committee approved the study protocol and all the participants give their written informed consent. Among the 800 patients enrolled in the COPANA cohort, 214 joined the metabolic sub-study [60]. The 121 VIR patients corresponded to patients with available CRP, IL-6, MCP1, TNFα, sCD14, sCD163, sTNFR1 and TNFR2 values and who initiated cART at enrollment or during follow-up. The 121 VIR patients were enrolled within a median of 6 months (range 3.8–9.1) after HIV-1 diagnosis. IP-10 was measured at cART initiation. A subgroup of 41 patients on sustained successful cART (>24 months on cART and >12 months with VL < 50 copies/mL) was identified. They were enrolled in the COPANA cohort less than 6 months after diagnosis of HIV infection. None of the subjects enrolled in our study received immunosuppressive drugs, IFN therapy or chemotherapy, and none had cancer, autoimmune diseases or HIV-unrelated chronic inflammatory metabolic disorders at enrollment. The HIV controllers (HIC, n = 82) were enrolled in the French ANRS C021 CODEX cohort with their written informed consent. This French multicenter cohort was created in 2009. To be enrolled, patients had to be diagnosed as HIV-infected for more than 5 years, to remain cART-naive with viremia below 400 copies/ml in five consecutive assays, regardless of their CD4 cell count. After enrollment, some patients started to exhibit viral blips and/or a slight loss of CD4 T cells [57, 58].
The samples collected at PHI M0 (n = 126) and PHI M6 (n = 35) were from subjects enrolled in the French ANRS C06 PRIMO cohort, who are thoroughly described in [17]. We used plasma leftovers for this study. We used data from the ANRS PRIMO cohort to address the relationship between IP-10 and the amount of infected cells, because viral reservoir size has been extensively studied in these patients from PHI onwards [61–64].
Frozen sera collected on EDTA were obtained from the ACS study, while frozen plasma collected on EDTA was obtained from the ANRS cohorts.
EDTA plasma from HIV/HBV/HCV-seronegative individuals (n = 87) were obtained from Etablissement Français du Sang (EFS, Paris, France) for research purposes.
All animals were housed in the CEA IDMIT facilities (Center for Infectious Disease Models and Innovative Therapies), Fontenay-aux-Roses, France (permit number A 92-032-02) or the Pasteur Institute, Paris, France (permit number A 78-100-3). All experimental procedures were conducted in strict accordance with the European guideline 2010/63/UE for the protection of animals used for experimentation and other scientific purposes (French decree 2013–118). CEA complies with Standards for Human Care and Use of Laboratory Animals of the U.S. Office for Laboratory Animal Welfare under OLAW assurance number A5826-01. The animal experimentation ethics committee approved all experimental protocols (CETEA-DSV, IDF, France; notification numbers 10-051b and 12–006).
Twenty-three (Chlorocebus sabaeus) were infected by intravenous inoculation with 250 TCID50 of purified SIVagm.sab92018 [13, 14, 15].
Nineteen rhesus macaques (Macaca mulatta) and eleven cynomolgus macaques (Macaca fascicularis) were infected i.v. with 50 AID50 of an uncloned SIVmac251 isolate (provided by A. M. Aubertin, Université Louis Pasteur, Strasbourg, France). Twenty-nine cynomolgus macaques were inoculated intra-rectally with 5 (n = 10) or 50 AID50 (n = 13) of the same uncloned SIVmac251 isolate.
In addition, 3 cynomolgus macaques were treated with AZT (4.5 mg/kg) and 3TC (2.5 mg/kg) subcutaneously twice daily and oral indinavir (60 mg/kg) twice daily. The treatment was initiated as early as 4 hours post-challenge and continued until day 14, when the animals were killed. These animals are described in [29]. We used plasma leftovers collected on D9 and D14.
Blood and intestinal samples were obtained from AGM. Blood and lymph node samples were obtained from Cynomolgus macaques. The latter tissues were used for viral load quantification. Blood and intestinal samples were collected from Rhesus macaques, and the tissues were used to measure cellular gene expression.
Whole blood collected on EDTA was used to prepare plasma or serum. IP-10 and sCD163 concentrations were determined in stored plasma or sera samples (−80°C) by specific enzyme-linked immunosorbent assay, human Quantikine CXCL10 and human CD163 Duoset (R&D Systems, Minneapolis, Minnesota) according to the manufacturers' instruction as previously performed.
At necropsy (day 65 post-infection), a fragment (5/7cm in length) was collected from each of 4 sections of the intestine (jejunum, ileon, colon and rectum) of 5 SIVmac251-infected rhesus macaques and 5 SIVagm.sab92018-infected AGM. The ileum fragment was not collected from 2 rhesus macaques and 2 AGMs. The fragments were enzymatically dissociated in RPMI culture medium (Life Technologies) containing collagenase (Collagenase II-S, Sigma-Aldrich) and DNAse (Sigma-Aldrich) for 1 h with agitation (80 rpm) at 37°C. Total leukocytes were separated from epithelial/endothelial cells through a Percoll gradient. CD4-positive and -negative leukocytes were purified on Miltenyi columns and with a CD4 cell purification kit.
Cryopreserved lymph node cell samples were stained with LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (Thermo Fisher), then labeled with anti-CD45-PerCP, anti-CD4 Pacific Blue and anti-CXCR3-PE-Cy7. Viable CD45+CD4+CXCR3-, CD45+CD4+CXCR3+, CD45+CD4-CXCR3- and CD45+CD4-CXCR3+ cells were sorted using a FACS Aria cell sorter (BD Biosciences) equipped to handle biohazardous material. Human primary Th1 (CD25neg, CXCR3high, CCR4neg, CCR6neg), Th1/Th17 (CD25neg, CXCR3high, CD161+, CCR6+), Th17 CD161+ (CD25neg, CXCR3neg, CD161+, CCR4+ CCR6+) and Th17 (CD25neg, CXCR3neg, CD161neg, CCR4+ CCR6+) were isolated as described [65, 66] from HIV negative blood cytapheresis.
Total RNA was extracted and reverse-transcribed as previously described [15]. qPCR (Taqman chemistry) and commercial kits were used to quantify the expression levels of genes of interest (CXCL10 Rh02788358_m1, CD14 Rh03648680_s1, CD68 Hs02836816_g1, CD163 Hs00174705_m1). The expression of each gene was normalized to that of 18S rRNA, and relative expression levels were calculated using the ΔΔCT method. The relative gene expression levels were determined by using as the internal reference the raw value for each gene in rectal CD4+ leukocytes from one rhesus macaque, allowing direct comparison between each species (Fig 7A, 7B, 7D–7F and 7J). Alternatively, relative expression levels were determined by normalizing each value against the raw value of each gene in CD4+ leukocytes enriched from the rectum of each animal. This reduces inter-individual differences and highlights differences between the small and large intestine (Fig 7C, 7G–7I, 7K and 7L).
HIV-1 DNA load in PBMC was measured in the laboratory of Prof. C. Rouzioux [67]. SIV viremia and SIV DNA load in lymphoid tissues were determined as previously described [15, 28]. The cut-offs for cynomolgus macaques were 12 copies/ml of plasma and 12 copies/million cells, respectively.
Fresh jejunum fragments were obtained from SIVmac251-infected cynomolgus macaques within 30 min of necropsy (Day 240 p.i.). These fragments were embedded and snap frozen at optimum cold temperature compound (OCT) and 10 μm frozen sections were stained using unconjugated primary antibodies (CD68 clone KP1 from Santa Cruz, IP-10 clone ab47045 from Abcam) followed by appropriate secondary antibodies conjugated to Alexa 488 (green), Alexa 568 (red) (Molecular Probes, Eugene, OR). Prior to staining, slides were incubated with 100–200 μL of ice cold methanol and 5% acetic acid, allowed to rest at -20°C for 10 min then washed 3 times with PBS. Confocal microscopy acquisition was performed using a Leica TCS SP8 confocal microscope (Leica Microsystems, Exton, PA). Individual optical slices were collected at 512 × 512 pixel resolution. Image J software were used to assign colors to the channels collected.
(See supplementary materials)
|
10.1371/journal.ppat.1006060 | Disrupting Mosquito Reproduction and Parasite Development for Malaria Control | The control of mosquito populations with insecticide treated bed nets and indoor residual sprays remains the cornerstone of malaria reduction and elimination programs. In light of widespread insecticide resistance in mosquitoes, however, alternative strategies for reducing transmission by the mosquito vector are urgently needed, including the identification of safe compounds that affect vectorial capacity via mechanisms that differ from fast-acting insecticides. Here, we show that compounds targeting steroid hormone signaling disrupt multiple biological processes that are key to the ability of mosquitoes to transmit malaria. When an agonist of the steroid hormone 20-hydroxyecdysone (20E) is applied to Anopheles gambiae females, which are the dominant malaria mosquito vector in Sub Saharan Africa, it substantially shortens lifespan, prevents insemination and egg production, and significantly blocks Plasmodium falciparum development, three components that are crucial to malaria transmission. Modeling the impact of these effects on Anopheles population dynamics and Plasmodium transmission predicts that disrupting steroid hormone signaling using 20E agonists would affect malaria transmission to a similar extent as insecticides. Manipulating 20E pathways therefore provides a powerful new approach to tackle malaria transmission by the mosquito vector, particularly in areas affected by the spread of insecticide resistance.
| Mosquito control is the only intervention that can reduce malaria transmission from very high levels to close to zero. However, current mosquito control methods are severely threatened by the rapid spread of insecticide resistance in anopheline mosquito populations that transmit the malaria-causing Plasmodium parasites. Here we show that when steroid hormone signaling is interrupted in female Anopheles mosquitoes, various aspects of their lifecycle are disrupted—females produce and lay fewer eggs, do not mate successfully and die more rapidly. Furthermore, they become less likely to be infected by malaria parasites. When we model the impact of steroid hormone agonists on malaria transmission, we predict that these compounds would provide an important new tool against malaria, particularly in regions of widespread insecticide resistance.
| Despite recent progress in combating the malaria parasite, nearly 200 million infections and around 500,000 deaths are caused by malaria annually, mostly in young children in sub-Saharan Africa [1, 2]. Even with new drugs and vaccines in the research pipeline [3], control of the Anopheles species that transmit human malaria remains the cornerstone of prevention and transmission reduction efforts [2, 4]. Of the four classes of insecticides available for malaria control, pyrethroids are the only compounds approved for use on long-lasting insecticide-impregnated bed nets (LLINs), due to their relatively low toxicity, and they are heavily used in indoor residual spray (IRS) programs [5]. This is a major limitation, as the increased application of both interventions over the last decade has inevitably led to the emergence and spread of insecticide resistance in natural mosquito populations. Indeed, resistance to pyrethroids has been observed in most Anopheles populations from sub-Saharan Africa [6], making the identification of alternative non-toxic compounds that can reduce parasite transmission a high priority in the malaria control agenda [1].
Insecticide-based interventions impact malaria transmission by increasing the mortality rate of exposed female mosquitoes and, in the case of LLINs, by preventing them from biting humans. Mathematical models developed to aid in the design of malaria elimination programs during the first global eradication campaign showed the importance of increasing mosquito mortality [7, 8], which reduces the probability that mosquitoes will survive for the 12–14 day incubation period of the malaria parasite [9]. However, other aspects of adult mosquito biology that determine vectorial capacity, such as host preferences for blood-feeding, susceptibility to parasite development, and reproductive fitness, have not yet been fully exploited for malaria control.
Anopheles population densities are driven by the complex mosquito lifecycle involving multiple gonotrophic cycles in fertilized females. Following a single insemination event, a female stores sperm for her lifetime, using it to fertilize each egg batch produced after a blood meal [10–12]. Many of the processes characterizing this reproductive cycle are regulated by 20-hydroxyecdysone (20E), a steroid hormone originally studied in insects for its fundamental role in larval molting [13]. Besides an essential function of female-produced 20E in triggering egg development after blood feeding [14–16], in Anopheles gambiae, as well as in other important anopheline vector species, sexual transfer of this hormone by the male induces a dramatic series of molecular events that culminate in increased oogenesis, induction of egg laying, and loss of the female’s susceptibility to further mating [17–20].
Based on its multiple physiological effects, it is reasonable to speculate that 20E signaling pathways in the female mosquito could be exploited to manipulate reproductive success and possibly other aspects of mosquito biology that are relevant for vectorial capacity. To this end, synthetic 20E non-steroidal agonists such as dibenzoylhydrazines (DBHs), which mimic the action of 20E by competitively binding to the ecdysteroid receptor, resulting in high ecdysteroid activity [21, 22], could be utilized. When provided to larvae of some Lepidopteran and Dipteran species, DBHs induce precocious and incomplete molting, ultimately leading to death [21–25]. These compounds have extremely low toxicity to mammals and are non-carcinogenic [26, 27], and although reduced fitness of adult stages following DBH exposure has been documented in agricultural lepidopteran species [24], their potential use against adult stages of malaria vectors has not been tested.
Here we show that topical application of the non-steroidal ecdysone agonist methoxyfenozide (a DBH compound, herein referred to as DBH) significantly limits the reproductive success of adult An. gambiae females and greatly increases their mortality. Furthermore, females exposed to DBH are significantly less susceptible to infection with the human malaria parasite Plasmodium falciparum. We incorporate our experimental findings into a mathematical model of the mosquito life cycle to determine the potential impact of these multiple biological effects on mosquito population dynamics and malaria transmission. Our results suggest that the application of compounds targeting ecdysteroid pathways on impregnated bed nets or in indoor spray programs would significantly reduce malaria transmission, achieving results comparable to those from the use of insecticides. Manipulating 20E signaling in Anopheles mosquitoes therefore provides a new strategy for malaria control, especially needed in areas of widespread resistance to insecticides.
To determine if steroid hormone signaling could be disrupted to manipulate entomological parameters key to malaria transmission, we treated An. gambiae females with the 20E agonist DBH and assessed its effects on egg laying and mating success, two reproductive traits that can impact mosquito population size and hence the frequency of encounters with the human host. While 20E is essential for egg development in insects, it has been shown that levels above a critical threshold can induce apoptosis of ovarian follicles [28, 29]. We therefore reasoned that topical application of DBH to the female’s thorax might disrupt egg development and thus reduce the number of eggs laid. Moreover, as 20E injections in virgin females completely abolish insemination in a number of Anopheles species [20], we tested whether exposure to DBH might achieve the same result. We also expected an effect of DBH application on female mortality, as, in Drosophila melanogaster, 20E is known to regulate longevity [30, 31], a crucial parameter of malaria transmission.
Egg laying, mating, and lifespan were all significantly altered in females treated with this 20E agonist (Fig 1). We first tested mated females for their ability to develop and lay eggs when exposed to DBH 24h prior to blood feeding at 5 different doses (ranging in a 2-fold dilution series from 2 μg– 0.125 μg per mosquito, respectively). While 100% of the control females oviposited after a blood meal, females exposed to DBH showed a dose-dependent reduction in oviposition, with 47.4% of individuals laying eggs at the intermediate dose (0.5 μg) and only 10.9% at the highest dose (2 μg) (Fig 1A). An ED50 dose of 0.5 μg (0.08–0.13 95% CI) was determined from the dose-response curve (slope: 2.32, R = 0.993). Moreover, even in cases where females oviposited, we detected at the three highest doses a significant reduction in the number of eggs laid, with a median egg number of zero compared to a median of 91.5 eggs in the control group (Fig 1A). Upon dissection of individuals in all DBH-exposed groups, we found that 98.7% of females had no follicular growth or yolk deposition in the ovaries despite having fully engorged on blood, demonstrating that the reduction in oviposition rates was due to impaired oogenesis. We did not observe induction of autogenous egg development after exposure contrary to what is reported in culicine mosquitoes after DBH treatment in larvae [32]. Microscopic analysis of ovaries after terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assays detected extensive fragmentation of chromatin, indicative of apoptosis, limited to the primary ovarian follicles of treated females 24h after DBH exposure, while control females had no observable apoptosis and had otherwise normal follicular morphology (S1 Fig). Similar apoptotic follicles were observed after 20E injections, showing that the phenotype induced by DBH application recapitulates the effects induced by 20E (S1 Fig). It is worth noting that previous experiments where 20E injections induced egg laying [19] were performed in blood fed females that had completed egg development.
Mating and female longevity were also significantly reduced in a dose-dependent manner compared to controls as a result of DBH exposure (Fig 1B and 1C). After treating virgin females with 3 DBH doses (0.125 μg (low), 0.5 μg (intermediate), and 2 μg (high), corresponding approximately to the ED10, ED50 and ED90 from the oviposition data) and placing them with males for 2 days, we observed a 25–65% reduction in insemination rates, as determined by the presence of sperm in the spermatheca (Fig 1B). Moreover, DBH-treated females showed an eight-day reduction in median survival time at the highest dose compared to controls (median survival time in 2 μg DBH: 11 days; Control: 19 days) (Fig 1C), and lifespan was reduced even at the lowest dose (median survival time in 0.125 μg DBH: 16 days). Overall these results demonstrate a strong, dose-dependent effect of DBH on important determinants of vectorial capacity.
As 20E signaling is an important modulator of the female’s post-blood feeding physiology, we determined whether manipulating steroid hormone pathways via topical application of DBH also impacted the establishment of P. falciparum infection in the mosquito vector, using the three doses utilized in the mating and longevity assays. P. falciparum prevalence (NF54 strain) was significantly reduced 7 days post-infectious blood meal at the two higher DBH doses relative to the control. At the highest dose, only 7.5% of females who fully engorged on an infectious blood meal were positive for oocysts, corresponding to an 87% reduction in infection prevalence relative to controls (Fig 1D). When using the intermediate dose, 25.8% of females failed to develop an infection, providing a 56% reduction in prevalence. Limited effects were observed in the low dose treatment group, where oocyst prevalence (50.7%) was similar to the control group (58.5%). In those females that developed oocysts, the intensity of infection was not significantly affected (S2 Fig).
We developed a discrete-time deterministic mathematical model of the mosquito life cycle (Fig 2, S1 Table) to predict the effect of manipulating steroid hormone signaling via DBH on Anopheles population dynamics. We modeled mosquitoes through their life cycle, starting with juvenile aquatic stages (eggs, larvae, and pupae), followed by mating and up to six gonotrophic cycles consisting of blood feeding, resting, and ovipositing. We used our experimental findings to define the efficacy of DBH in our model (S1 and S2 Tables, Methods). Based on our experimental findings, at the highest dose the efficacy was defined to be a 95% reduction in egg batch size, 65% reduction in mating success, 87% reduction in Plasmodium infection risk, and an enhanced age-dependent mortality based on the experimental survival curves, with an 8-day reduction in median survival time (Fig 1, S2 Table). The efficacies defined for application of lower DBH doses can be found in S1 Table. Note that the purpose of this modeling approach was to elucidate the qualitative impacts of the multiple effects of DBH on the non-linear mosquito lifecycle, rather than to make quantitative predictions about its impact in field conditions.
We explicitly modeled the impact of DBH either delivered on impregnated bed nets to target females as they attempt to blood feed, or in indoor sprays to target females as they rest after feeding. Since our experimental results provide insight into a single exposure to DBH, and in order to compare with insecticide efficacy, we took a conservative approach and only considered possible exposure to DBH or insecticide to occur once, on the first feeding day in the case of treated bed nets or on the first indoor resting day for IRS. The insecticide was modeled at 100%, 80%, and 60% efficacy to reflect a situation of partial resistance emerging in the mosquito population [33]. In addition, we examined varying levels of coverage, i.e. the proportion of mosquitoes exposed in their first feed or first day of indoor rest (S3A Fig, Methods). To qualitatively examine the relative efficacy and general mechanisms of transmission reduction following a DBH-based intervention, we used a single well-mixed mosquito population without spatial structure (Methods).
The modeled mosquito population and its age structure were significantly altered by the use of DBH in bed nets and indoor residual sprays (Fig 3, S4 Fig). The relationship between DBH exposure and mosquito population size was non-linear. At the strongest dose and highest levels of coverage the mosquito population was driven to extinction, but for most levels of coverage the total adult mosquito population actually increased with DBH exposure due to reduced density-dependent larval mortality (Fig 3A, S5A Fig), consistent with previous models of vector control interventions [34–36]. However, this effect was accompanied by a shift in the age distribution of adult females towards younger individuals (Fig 3B). At the low and high DBH doses, for all levels of coverage, this shift alone reduced the proportion of females that lived long enough to transmit malaria, i.e. females surviving at least 12 days following their first blood meal (S5B Fig). Since our experimental results show that DBH also blocks P. falciparum infection, we combined this shift in age structure with the dose-dependent reduction in susceptibility to estimate the overall impact on the fraction of potentially infectious mosquitoes (Fig 3C). In comparison to insecticide exposure at the same coverage, low and intermediate doses of DBH showed a similar reduction in adult mosquitoes able to transmit malaria to that of 60% and 80% insecticide efficacy, respectively, while high DBH dose performed comparably if not better than a 100% effective insecticide (Fig 3C). Delivery of DBH through indoor spraying showed similar effects on the mosquito population (S4 Fig).
To examine the impact of DBH on malaria, we extended our mathematical framework to include malaria transmission with feedback between infectious human and mosquito populations (S6 Fig; S1 Table). In the model, mosquitoes required at least 12 days following an infectious blood meal to become infectious to humans [9]. Given the non-linear relationships between components of the model, transmission intensity prior to interventions may influence their relative impacts. We therefore considered three transmission settings, with high (85%), moderate (45%), and low (5%) malaria prevalence pre-intervention, and different DBH doses (2 μg, 0.5 μg, and 0.125 μg). Importantly, our results are not intended to quantitatively reproduce field conditions, but rather to reflect the relative reductions for a range of pre-intervention transmission settings and in comparison to the use of insecticides. As mentioned above, for simplicity we consider the effectiveness of each intervention to be the reduction in malaria prevalence after a single exposure relative to the pre-intervention prevalence, and in our model mosquitoes become exposed to DBH or insecticides only at the time of their first blood feeding (or their first day of indoor rest for IRS).
Application of 2 μg DBH via bed nets showed strong effectiveness against malaria prevalence, outcompeting the impact of any insecticide efficacy in all transmission settings (Fig 4). The effects of low and intermediate DBH doses were comparable to those of 60% and 80% insecticide efficacy respectively, regardless of initial malaria prevalence. Effectiveness increased significantly with increasing coverage for all doses of DBH. Similar dynamics were observed when modeling DBH use in indoor sprays (S7 Fig). Regardless of the method of application or dose considered, a single exposure to DBH led to a reduction in malaria prevalence in all transmission settings at all coverage levels. We therefore expect that repeated exposures over the course of a female’s lifespan would have a larger impact on malaria.
The development of non-toxic compounds that target the mosquito vector in novel ways will be essential for achieving malaria elimination goals [1]. Our study identifies steroid hormone signaling as a promising new target that can provide an effective and complementary approach to existing tools for vector control. Although our data are based on topical application, and therefore cannot directly be extrapolated to effectiveness in field settings, we observed a robust, dose-dependent impact of the steroid hormone agonist DBH on egg development, insemination rates, and adult female longevity in experimental applications. Moreover, DBH strongly prevented the development of the deadliest human malaria parasite, P. falciparum, a highly desirable feature for any compound used in malaria control strategies. While the effects on fecundity, insemination rates, and longevity are in agreement with previous studies linking high 20E activity to apoptosis of ovarian follicles [28, 29], reduced mating receptivity [19] and premature aging [30, 31], the observed reduction in P. falciparum infections was completely unexpected. In future studies it will be important to test steroid hormone agonists under field conditions, using different mosquito strains and parasite isolates, and to determine whether the effects on oogenesis and parasite development are linked, for example, via the induction of immune pathways.
Interestingly, recent studies targeting another insect hormone, the sesquiterpene juvenile hormone, have shown life shortening and sterilizing activity against adult Anopheles species [37–40]. Pyriproxyfen (PPF, a juvenile hormone analog) is currently available in LLINs for its combined efficacy with insecticides [37, 41, 42] although, to our knowledge, the effects of juvenile hormone agonism on Plasmodium development within the mosquito have not been tested. Upon optimization and testing of DBH or other chemistries targeting steroidal pathways for tarsal uptake, alternating the use of steroid hormone and juvenile hormone agonists in combination with insecticides may provide the key to preventing the spread of resistance to conventional insecticides and extending their effectiveness.
In our experimental setting, DBH had biological efficacy by topical application at higher but still comparable doses to permethrin in mosquitoes exhibiting pyrethroid target-site resistance [43], which is an important factor when considering both the cost effectiveness and safety of using these substances in the fields. We explicitly considered only the acute lethality of insecticide exposure in our model as the overall toll that insecticide resistance and sub-lethal doses of insecticide may impose on mosquitoes and parasites is only now being investigated [44]. As the incidence and severity of resistance increases in wild populations, understanding the biological effects of sub-lethal insecticide exposure will be increasingly important.
Our modeling approach illustrates how the different actions of steroid hormone agonists interact in non-intuitive ways to change mosquito populations and malaria prevalence in comparison to insecticides. For simplicity we measured effectiveness after a single DBH exposure, and for consistency in our comparison we modeled contact with insecticide only during the first blood feeding cycle. Thus our model results provide a conservative estimate of the possible outcomes of DBH and insecticide exposure. Despite this, they clearly indicate that steroid hormone agonists would perform comparably to insecticides in all malaria scenarios tested. Even at lower doses, when the individual effects on reproduction, longevity, and parasite development become less striking, the combination of these effects is powerful enough to achieve a substantial reduction in malaria, mostly due to a decrease in the potentially infectious adult female population.
As observed in other modeling studies, the release of density-dependent mortality plays an important role in the predicted impact of interventions [34–36]. For generalizability, we did not include fluctuations in carrying capacity, although variations in climatic conditions (e.g. rainfall, temperature, humidity) that determine the availability and quality of larval breeding sites are clearly important factors in the field that prevent populations from reaching equilibrium [45–48]. It is therefore likely that our framework is additionally conservative in that it over-estimates the increases in mosquito population size following DBH application. The counter-intuitive finding that malaria prevalence can be reduced even when mosquito population size increases highlights the importance of qualitative models of this kind that can identify potential multifactorial and non-linear effects of interventions. It will be important to validate the model and test the arising hypotheses in the field in future studies, since our results are qualitative and mechanistic, rather than providing quantitative predictions.
In addition to the physiological effects we have examined here, compounds belonging to the DBH class have additional characteristics that make them promising tools for malaria vector control. DBHs are not toxic to mammals (LD50 > 5000 mg/kg by ingestion for methoxyfenozide, which compares favorably to the LD50 = 430 mg/kg by ingestion of the pyrethroid permethrin commonly used on LLINs) and would therefore be ideally suitable for bed net-based strategies, where low toxicity is essential. Moreover, although resistance to DBH has been experimentally induced in larvae from a number of insect agricultural pests and has been described in wild populations with varying reports on the possible biochemical and genetic basis [24, 49–51], it has rapidly reversed when the intervention was withdrawn [24, 50]. Although resistance to DBH or any other active compound will inevitably occur at some point, different strategies such as rotation, mosaics, or combination with insecticide could be utilized to delay its emergence. Interestingly, interventions like DBH that have a combination of physiological effects and/or alter mosquito age structure, rather than causing instant lethality, may exert less selection pressure on targeted mosquitoes relative to conventional insecticidal approaches, whilst impacting malaria transmission to a similar degree. However, Plasmodium parasites could develop independent mechanisms of resistance against the as yet uncharacterized anti-parasitic activity exerted by DBH. In this event, this could potentially uncouple the DBH effects on mosquito physiology from those on parasite development. Based on the conserved role of 20E pathways in regulating female physiology in multiple anopheline species [20], we expect that compounds interfering with steroid signaling will be biologically effective against other important malaria vectors such as An. arabiensis, An. funestus, and An. stephensi. Approaches targeting these hormonal pathways could therefore be a potent addition to the limited toolbox of vector interventions for successful malaria control in Africa and other regions of the world affected by this disease.
|
10.1371/journal.ppat.1004302 | The Murine Gammaherpesvirus Immediate-Early Rta Synergizes with IRF4, Targeting Expression of the Viral M1 Superantigen to Plasma Cells | MHV68 is a murine gammaherpesvirus that infects laboratory mice and thus provides a tractable small animal model for characterizing critical aspects of gammaherpesvirus pathogenesis. Having evolved with their natural host, herpesviruses encode numerous gene products that are involved in modulating host immune responses to facilitate the establishment and maintenance of lifelong chronic infection. One such protein, MHV68 M1, is a secreted protein that has no known homologs, but has been shown to play a critical role in controlling virus reactivation from latently infected macrophages. We have previous demonstrated that M1 drives the activation and expansion of Vβ4+ CD8+ T cells, which are thought to be involved in controlling MHV68 reactivation through the secretion of interferon gamma. The mechanism of action and regulation of M1 expression are poorly understood. To gain insights into the function of M1, we set out to evaluate the site of expression and transcriptional regulation of the M1 gene. Here, using a recombinant virus expressing a fluorescent protein driven by the M1 gene promoter, we identify plasma cells as the major cell type expressing M1 at the peak of infection in the spleen. In addition, we show that M1 gene transcription is regulated by both the essential viral immediate-early transcriptional activator Rta and cellular interferon regulatory factor 4 (IRF4), which together potently synergize to drive M1 gene expression. Finally, we show that IRF4, a cellular transcription factor essential for plasma cell differentiation, can directly interact with Rta. The latter observation raises the possibility that the interaction of Rta and IRF4 may be involved in regulating a number of viral and cellular genes during MHV68 reactivation linked to plasma cell differentiation.
| Through coevolution with their hosts, gammaherpesviruses have acquired unique genes that aid in infection of a particular host. Here we study the regulation of the MHV68 M1 gene, which encodes a protein that modulates the host immune response. Using a strategy that allowed us to identify MHV68 infected cells in mice, we have determined that M1 expression is largely limited to the antibody producing plasma cells. In addition, we show that M1 gene expression is regulated by both cellular and viral factors, which allow the virus to fine-tune gene expression in response to environmental signals. These findings provide insights into M1 function through a better understanding of how M1 expression is regulated.
| MHV68 is a naturally occurring murid gammaherpesvirus that has significant genetic and functional homology to the human gammaherpesviruses Epstein-Barr virus (EBV) and Kaposi's sarcoma-associated herpesvirus (KSHV). Among herpesviruses, there are a large number of genes involved in virus replication that are conserved – both in sequence and spatial arrangement in the viral genome. However, every herpesvirus, having co-evolved with its host during speciation, has acquired unique genes - many of which function to modulate and/or evade the host immune response. Coevolution of with their hosts has led to some divergence of host-pathogen interactions; however, unique genes may reveal homologous functions required for chronic infection of the host. One such gene is the MHV68 M1, which is found in a cluster of unique genes at the left end of the MHV68 genome.
Initial functional studies of M1, utilizing an M1-null virus revealed a hyper-reactivation phenotype from latently infected peritoneal exudate cells (PEC) [1]. Subsequent studies found that this hyper-reactivation phenotype was strain specific – occurring in C57Bl/6 mice, but not Balb/c mice [2]. In addition to the strain specific reactivation phenotype, a strain specific expansion of Vβ4+CD8+ T cells had previously been observed in response to MHV68 infection [3]. This pronounced T cell expansion and activation is a hallmark of MHV68 infection in many inbred mouse strains and is observed in peripheral lymphoid organs, as well as the blood, reaching peak levels after the virus has established latency [3], [4]. Notably, the Vβ4+CD8+ T cells remain elevated during the course of chronic MHV68 infection, and do not adopt an exhausted phenotype [3]. Analysis of M1-null mutants revealed that a functional M1 gene is required for the Vβ4+CD8+ T cell expansion [2]. Furthermore, M1 was shown to be a secreted protein capable of stimulating Vβ4+CD8+ T cells to produce IFNγ and TNFα [2]. These analyses suggested that M1 may exert control over MHV68 reactivation from peritoneal macrophages through the induction of IFNγ from Vβ4+CD8+ T cells [2], this is supported by the observations that: (i) IFNγ−/− mice exhibit hyper-reactivation from PECS [5]; and (ii) the demonstration that IFNγ can suppress MHV68 replication in macrophages [2], [6], [7].
Early experiments to evaluate the expansion in thymectomized mice suggested that Vβ4+CD8+ T cells are maintained through continued stimulation by a stimulatory ligand, which is now known to be M1 [8]. Interestingly, B cells appear to play a critical role in the expansion of Vβ4+CD8+ T cells, as no expansion is observed upon MHV68 infection of mice lacking B cells [9], [10]. Other studies provide some clues to the timing and site of M1 expression during MHV68 infection, where B220+ splenocytes at 14 days post-infection were found to be capable of stimulating Vβ4+CD8+ T cell hybridomas [11].
Though no homolog to M1 has been found in other gammaherpseviruses, HVS has been shown to encode a viral superantigen, immediate early gene ie14/vsag [12]. Like M1, ie14/vsag, is not essential for viral replication; and interestingly, ie14/vsag expression is elevated in phorbol ester treated cells, indicating a link with viral reactivation. In EBV, structural protein gp350, as well as latent membrane proteins LMP-1 and LMP-2A have been shown to activate expression of an endogenous human retroviral superantigen, HERV-K18, which results in a Vβ13+ T cell expansion [13]–[15]. Due to limitations in study of non-human primate and human patients it has been difficult to assess the role of these superantigens and the consequence of their resulting T cell expansion. We are therefore left to speculate what benefit they provide to their host. Do they aid in infection or the establishment of latency? Do they divert the immune response? Are they involved in control of infection? We hope that a better understanding of the expression and role of M1 in MHV68 infection may shed light into the conserved use of viral superantigens by gammaherpseviruses.
Though numerous studies to define the transcriptional program of MHV68 in vitro have identified M1 as an early through late gene [16]–[18] relatively little is known about when and where M1 is expressed during infection. Furthermore, while a number of transcriptome based analyses have detected transcripts extending through the M1 locus during in vivo infection, much of this data relies on methods that are not strand specific and therefore not definitive [19]–[22]. Due to the dearth of information about M1expression in vivo, we set out to characterize M1 expression using a novel approach wherein a fluorescent reporter virus would allow detection of M1 promoter activity during infection. This approach led to the identification of splenic plasma cells as the primary cell type expressing M1 in vivo. Furthermore, factors regulating M1 transcription were previously uncharacterized. The current studies have elucidated key cis-elements and transcription factors controlling the expression of M1 in plasma cells. Overall, these findings provide insights into the role of M1-mediated regulation of MHV68 pathogenesis. Moreover, we reveal a novel and potentially conserved mechanism which controls the timing and site of viral gene expression in response to reactivation in the B cell.
To identify cellular reservoirs in which the M1 gene is expressed in vivo, we generated a series of recombinant viruses that express yellow fluorescent protein (YFP) to mark infected cells. For detection of M1 promoter activity, the M1 coding sequence was replaced with that encoding YFP, creating a M1 promoter-driven YFP mutant (Figure 1A). This strategy allows detection of the cellular reservoirs in which M1 is expressed during infection. Additionally, two important controls were used: MHV68-YFP, in which the YFP transgene under the control of the HCMV IE promoter was cloned into a neutral locus in the viral genome (efficiently marking MHV68 infected B cells and plasma cells) [23]; and (ii) MHV68-M1st.YFP, which contains the M1 translational stop mutation (M1-null virus) in the context of the YFP transgene cloned into the neutral locus (Figure 1B). As M1 has previously been identified as a non-essential for both virus replication and for the establishment of latency in vivo [24], we did not anticipate that the M1pYFP recombinant would change the cellular reservoirs infected by MHV68. However, to formally address this issue, we have included analyses of the MHV68-M1st.YFP virus – which like the M1pYFP lacks a functional M1 gene.
Analysis of MHV68 infection of splenocytes at day 14 post-infection revealed robust marking of splenocytes by both the MHV68-YFP and MHV68-M1stYFP viruses (Figure 2). We have previously noted that there is significant mouse to mouse variation in the frequency of infected splenocytes for a given virus [25], and have recently determined that this directly correlates with the frequency of the CD4+ T follicular helper (TFH) response [26]. For these analyses we observed on average ca. 0.5% and 1.0% of splenocytes were YFP+ for the MHV68-YFP and MHV68-M1stYFP viruses, respectively (Figure 2). The latter result confirms that M1 function is dispensable for the establishment of latency in splenocytes. In contrast, only ca. 0.04% of splenocytes were YFP+ with the M1pYFP virus, indicating that the M1 promoter is active in only ca. 5–10% of infected splenocytes.
We have previously shown that the majority (ca. 70–90%) of virally infected B cells, as indicated by YFP expression, exhibit a germinal center phenotype [23], [27]. Individual mice were assessed for YFP marking and, consistent with previous observations, we found a similar frequency of virus infected (YFP+) B cells with a germinal center phenotype for mice infected with either the MHV68.YFP or MHV68-M1st.YFP viruses, both showing an average of ca. 70% (Figure 3). These results further substantiate that a functional M1 gene is dispensable for establishment of MHV68 latency in B cells. In contrast, few infected germinal center B cells were marked by the M1pYFP virus (an average of ca. 20% of YFP+ cells) – indicating that the majority of M1 expressing cells do not have a germinal center B cell phenotype. Based on the ca. 10-fold lower frequency of splenocytes marked by the M1pYFP virus (see Figure 2), we estimate that M1 promoter activity is only detectable in ca. 5% of infected germinal center B cells. Based on these results it is clear that M1 is predominantly expressed in some other MHV68 infected cellular reservoir.
The other major cell population in the spleen that is infected by MHV68 are plasma cells (CD138hi, B220low) [23], [27]. During infection, virus infection (YFP marking) of splenic plasma cells reaches peak levels at day 14 post-infection (ca. 10–20% of virus infected splenocytes) and begins to wane by day 18 post-infection (ca. 5–10% of virus infected splenocytes) [23]. We observed marking of splenic plasma cells for both MHV68-YFP and MHV68-M1st.YFP infected mice at day 14 post-infection consistent with previous observations, with ca. 10% YFP+ cells exhibiting a plasma cell phenotype (no significant difference between these 2 groups) (Figure 4). Strikingly, when assessing YFP marking of the splenic plasma cell population by the M1pYFP virus, the vast majority of YFP+ cells exhibited a plasma cell phenotype (on average >75% of YFP+ cells) (Figure 4). Thus, this strongly argues that M1 gene expression is largely limited to the infected plasma cell population. Notably, MHV68 reactivation from latently infected splenocytes is tightly linked to plasma cell differentiation [27], which suggests that M1 expression is coupled to virus reactivation from B cells. Finally, when considering the frequency of M1pYFP marked cells with the frequency of MHV68-YFP and MHV68-M1st.YFP marked splenic plasma cells, it appears that the majority of virus infected plasma cells express M1.
Having identified the reservoir where M1 is expressed in vivo, we sought to characterize the structure of the M1 transcript and to identify the M1 promoter. Rapid amplification of cDNA ends (RACE) was done to identify the transcript initiation and termination sites in two cell lines: (i) infected NIH3T12 fibroblasts; and (ii) reactivated A20-HE2 cells. A20-HE2 cells are a stable lymphoblast B cell line which carry the MHV68 genome where viral reactivation can be induced by tetradecanoylphorbol acetate (TPA) [28]. RNA and protein were collected from both cell lines and lytic gene expression was confirmed prior to analysis (data not shown). Transcript analysis revealed four initiation sites and a single termination site from an unspliced transcript (Figure 5A, 5B). Though all transcript initiation sites were found in infected 3T12 cells, only transcripts starting at bp 2003 and bp 2013 were detected from reactivated A20-HE2 cells. The sizes of the predicted unspliced M1 transcripts were confirmed by northern analyses of RNA prepared from: (i) TPA stimulated A20-HE2 cells (a MHV68 latently infected B cells); and (ii) MHV68 infected NIH 3T12 fibroblasts (data not shown).
To identify the regulatory elements controlling M1 gene expression we next set out to characterize the M1 promoter. Serial truncations of the putative M1 promoter region were cloned into a luciferase reporter vector and tested for promoter activity in a variety of cell lines. Notably, minimal activity was detected in the murine B cell lines A20, WEHI, NSO, and BCL1-3B3 (data not shown) – perhaps consistent with the failure to observe significant M1 promoter-driven YFP activity in most splenic B cell populations with the MHV68-M1pYFP virus in mice. In addition, we failed to detect significant activity from these reporter constructs in the murine macrophage cell line RAW264.7 (data not shown). However, when these reporter constructs were transfected into the P3X68Ag8 murine plasmacytoma cell line significant basal promoter activity was observed (Figure 6). Similar levels of M1 promoter-driven luciferase activity were observed for the longer M1 promoter constructs (M1p/−1025 bp, M1p/−525 bp, and M1p/−245 bp), while truncation of sequences upstream of −100 bp significantly decreased activity (Figure 6). Activity was further decreased to near background levels when sequences upstream of −50 bp were deleted (Figure 6).
The region upstream of the M1 transcription initiation sites was screened for the presence of candidate transcription factor binding sites [University of Pennsylvania Transcription Element Search System (TESS)]. TESS and manual sequence analyses identified a number of candidate transcription factor binding sites for NFκB, GATA3, IRF8/IRF4, and RBPJκ. Because M1 promoter activity was detected in plasma cells in vivo, interferon regulatory factor 4 (IRF4), a transcription factor upregulated in plasma cells which plays a critical role in plasma cell differentiation as well as immunoglobulin class switch recombination ([29]–[31] and reviewed in [32]), was of particular interest.
To characterize IRF4 binding to the candidate IRF site in the M1 promoter, an electrophoretic mobility shift assay (EMSA) was carried out (Figure 7A). EMSA was performed using nuclear extracts from P3X63Ag8 cells grown under normal conditions, along with a [32P]-labeled oligonucleotide probe containing the candidate M1 promoter IRF4 binding site. As expected we observed shifted complexes, which could be competed away using unlabeled double stranded DNA probes containing the M1p IRF4 binding site, but not with a competitor containing an IRF binding site mutation which has previously been shown to disrupt IRF8 binding with DNA [33] (Figure 7A). Furthermore, binding of IRF4 was confirmed by supershift analysis using an antibody against IRF4 (Figure 7A). This analysis was extended by generating M1 promoter-driven luciferase reporter constructs in which mutations were introduced into the IRF binding site. Two mutations in the core interferon response sequence, which have previously been shown to ablate IRF8 DNA:protein interaction [33], were introduced into the M1 promoter. Notably, either mutation led to a significant loss in basal M1 promoter activity (ca. 8-fold decrease in promoter activity) (Figure 7B).
Several studies have established a link between gammaherpesvirus reactivation from latency and plasma cell differentiation [27], [34]–[39]. Given that our data shows: (i) M1 promoter expression is detected from plasma cells during in vivo infection; (ii) basal M1 promoter activity requires a functional IRF4 site; and (iii) viral reactivation is linked with plasma cell differentiation, we set out to evaluate whether the M1 promoter is responsive to the MHV68 viral lytic transactivator Rta. Expressing increasing amounts of Rta with an M1 promoter-driven reporter construct in the P3X63Ag8 plasmacytoma cell line resulted in a dosage dependent increase in M1 promoter activity (Figure 8A). Moreover, the ability of Rta to efficiently transactivate the M1 promoter in the P3X63Ag8 cell line was dependent on the presence of an intact IRF4 binding site (Figure 8B). To further assess whether Rta functionally synergizes with IRF4 to activate the M1 promoter, we chose a cell line (293T cells) which lacks expression of Rta and IRF4. In 293T cells we observed that either factor alone led to very modest increase in M1 promoter activity (ca. 5–10 fold) (Figure 8C). However, when the two factors were co-expressed there was a significant increase in promoter activity (ca. 250-fold) (Figure 8C). Importantly, disruption of the IRF4 binding site dramatically impaired the ability of IRF4 and Rta to synergistically activate the M1 promoter (Figure 8C).
Based on the synergy between Rta and IRF4 in activating the M1 promoter, we assessed whether these factors can physically interact with each other. A co-immunoprecipitation was performed with cell lysates from transfected 293T cells. Immunoprecipitation with anti-IRF4 antibody, followed by anti-Flag detection of Rta, resulted in detection of a 90 kD band corresponding to Rta that was present only when Rta and IRF4 were co-expressed in 293T cells (Figure 8D). Following detection of Rta the blot was stripped and probed for IRF4 to confirm expression of the 52 kD band corresponding to IRF4. IRF4 was detected in whole cell lysates and immunoprecipitated samples containing IRF4. The reciprocal blot using anti-flag for immunoprecipitation and anti-IRF4 for detection showed a 52 kD band corresponding to IRF4. Additionally, Rta was detected from whole cell lysates and immunoprecipitated samples containing Rta. These results are consistent with a physical interaction between Rta and IRF4 that likely facilitates that observed synergy of these factors in activating M1 gene expression.
Several investigators have identified Rta responsive elements in viral promoters for both KSHV and MHV68 [40]–[44]. To date, the known Rta responsive genes are either regulated through: (i) direct interaction of Rta and DNA through a core Rta binding sequence; or (ii) Rta DNA binding is facilitated through protein-protein interactions – in the case of KSHV Rta, through interaction with the cellular transcription factor RBPJκ (reviewed in [45]). In MHV68 gene 57 promoter, it appears that both types of Rta response elements may be utilized – although a role for RBPJκ in MHV68 Rta activation has not been formally demonstrated [40], [41]. Interestingly, neither of the binding sites identified in the MHV68 gene 57 promoter are present in the M1 promoter, suggesting a novel Rta interaction motif. To identify the Rta response element(s) in the M1 promoter, a series of promoter truncations were generated and tested in the P3X63Ag8 plasmacytoma cell line. A candidate Rta response element was identified by evaluating promoter constructs which lost the ability to be transactivated by Rta. Using this approach we identified a putative Rta response element between −82 and −72 bp in the M1 promoter (Figure 9A). This 12 bp sequence (5′-GGTCAGAAGGCT-3′) failed to show homology to any known Rta response element identified in the gammaherpesvirus family. However, a screen of the MHV68 genome identified a number of candidate sites upstream of other MHV68 replication-associated genes which share significant homology with the core 5′-TCAGAAG-3′ sequence in the putative M1 promoter Rta response element (Figure 9B). Mutations of the three most central residues of the predicted Rta response element (see M1pRREm in Figure 9B) resulted in an ca. 10-fold reduction in transactivation in the plasma cell line (Figure 9C), as well as an ca. 6-fold reduction in Rta and IRF4 synergistic transactivation of the M1 promoter in 293T cells (Figure 9D).
With the identification of a novel Rta response element, we next wanted to evaluate whether this element was functional in other viral promoters that appear to contain this RRE (see Fig. 9C). Reporter constructs for the putative promoter regions of the M2 gene (encoding an adaptor protein involved in B cell signaling), ORF8 (encoding glycoprotein B), ORF22 (encoding glycoprotein H), ORF63 (encoding a tegument protein), and ORF73 (encoding the MHV68 Latency Associated Nuclear Antigen (LANA) homolog) were generated. In addition, the gene 50 proximal, distal, and N4/N5 promoter constructs previously described in Wakeman et al. [46] were evaluated for response to Rta expression. We observed varying levels of promoter response, with the strongest responses from ORF50pp, ORF8p, ORF22p, ORF63p, intermediate responses from the M1p, ORF50dp and ORF50 N4/N5p, and weak responses from M2p and ORF73p (Figure 10A). To further investigate the role of the Rta response element in the observed transactivation, we engineered the same three nucleotide mutation used in the M1p (Figure 9B) into the proximal ORF50 promoter (Figure 10B). Notably, mutation of this sequence resulted in a 38-fold reduction in Rta transactivation (Figure 10C). Notably, with the exception of the M1 promoter, for all the other reporter construct we failed to observe any synergistic activation by the co-expression of Rta and IRF4 (data not shown).
Here we described the characterization of a recombinant MHV68 in which a gene encoding a fluorescent protein (YFP) has been introduced into the viral genome in place of a non-essential viral gene. This approach allows identification of the site and timing of viral gene expression in vivo for viral genes that are dispensable for replication and/or dissemination of virus. For viral genes that play an important role in either replication or dissemination, other approaches - such as the generation of fusion gene products - may be required. Information obtained from such studies can provide significant insights into viral gene function and their mode of action. In the case of M1, these analyses led to identification of the predominant cellular reservoir in which M1 is expressed, and subsequent identification of transcription factors involved in regulating M1 gene transcription.
Coppola et al. have previously demonstrated the ability of either B220+ cells, or T cell depleted splenocytes, isolated from MHV68 infected mouse spleen to stimulate Vβ4+ CD8+ T cell hybridomas [11]. However, they also found that B cell depleted splenocytes from MHV68 infected mice retained Vβ4+ CD8+ T cell stimulatory activity – which they interpreted as the presence of other non-B cells populations in the spleen that are infected by MHV68. However, based on our findings that M1 expression is largely restricted to plasma cells, it seems unlikely that either the isolation or depletion of B220+ cells would efficiently capture or eliminate, respectively, all MHV68 infected plasma cells. As such, one would anticipate Vβ4+ CD8+ T cell stimulatory activity in both the enriched and depleted fractions. This interpretation is consistent with the complete failure to observe any expansion of Vβ4+ CD8+ T cells in MHV68 infected B cell-deficient mice [9], [10] – even though we have previously shown robust MHV68 infection in the spleens of B cell-deficient mice under some experimental conditions (intraperitoneal inoculation of virus) in the absence of any detectable Vβ4+ CD8+ T cell expansion [10], [47].
As we have previously shown, MHV68 reactivation in the spleen is tightly linked to plasma cell differentiation [27]. The observation that M1 is predominantly expressed in plasma cells thus suggested that M1 expression is linked to virus reactivation/replication. This was substantiated by demonstration that Rta can strongly transactivate the M1 promoter in a plasma cell line (see Figure 8A). We propose that during infection, in response to viral reactivation and the transition from germinal center or memory B cell to plasma cell, M1 expression is activated by the synergistic effects of viral Rta and cellular IRF4. M1 protein is secreted from infected plasma cells and, by an undefined mechanism, stimulates Vβ4+ CD8+ T cell activation and expansion. It is likely that M1 activates Vβ4+ CD8+ T cells via a mechanism similar to classic viral super-antigens [2]. Activation does not require classical MHC class I molecules [11], [48], but does require an intact M1 protein - we have previously shown that proteolytic digestion, or denaturation of recombinant M1 renders it unable to activate Vβ4+ CD8+ T cell hybridomas [2].
Vβ4+ CD8+ T cells have been shown to traffic throughout the body, and can be detected in the blood, spleen, liver, lung, and peritoneal cavity [2], [3], [8]. These cells show cytolytic activity [8] and adopt an effector memory phenotype where upon re-stimulation with recombinant M1 protein ex vivo they degranulate and produce INFγ and TNF α ([2], unpublished observations). As IFNγ has been shown to regulate MHV68 reactivation from macrophages in the peritoneum, but not reactivation from splenic B cells [6], we would predict that the Vβ4+ CD8+ T cells traffic to sites in which infection is less tightly controlled, to suppress MHV68 reactivation through the secretion of IFNγ in a paracrine fashion. We find it noteworthy that MHV68 M1-null infected mice exhibit hyper-reactivation in the peritoneal cavity and persistent viral replication in the lung [1], [2], [49], further underscoring the importance of M1 expression in controlling viral infection.
Our findings demonstrate that the M1 promoter is regulated by MHV68 Rta, a viral transcription factor that is essential for induction of viral reactivation. Rta activation of the M1 promoter synergizes with IRF4, a transcription factor that plays a critical role in both plasma cell differentiation and immunoglobulin class switch recombination. Furthermore, we show that this interaction is likely mediated through both DNA-protein interactions with the M1 promoter sequence, as well as protein-protein interactions between Rta and IRF4. We propose that during MHV68 latency, the viral latency-associated gene product M2 is expressed in a sub-population of latently infected germinal center and memory B cells [21] leading to expression of high levels of IRF4 [50]. M2 appears to play an important role in virus reactivation from latency: (i) MHV68 M2 null mutants exhibit a profound reactivation defect from B cells, but not latently infected macrophages [51], [52]; (ii) exogenous expression of M2 in primary B cells results in acquisition of a pre-plasma memory phenotype [53]; (iii) M2 can drive B cell differentiation of a B lymphoma cell line in vitro [27]; (iv) M2 is required for efficient immunoglobulin class switch in infected B cells in vivo [27]; and (v) plasma cells are the primary source of viral reactivation from the spleen [27]. Taken together these data suggest that MHV68 is capable of driving plasma cell differentiation, and concurrent with this differentiation, viral reactivation. As a result of this transition, the increased expression of the transcription factors Rta and IRF4 lead to induction of M1 expression in plasma cells (Figure 11).
That M1 is responsive to viral Rta and cellular IRF4 highlights the importance of tightly regulated gene expression in response to host and viral cues. This promotes cell type specific expression coordinated with viral reactivation. Furthermore, the interaction with Rta and IRF4 suggests a conserved strategy for gene regulation in MHV68, allowing for better control of Rta responsive gene expression. In fact numerous lytic genes in MHV68 appear to share the Rta response element identified in the M1 promoter (Figure 9B).
Though our efforts to find other viral genes that are similarly responsive to the concerted effects of Rta and IRF4 were unsuccessful, we find it attractive to speculate that the partnership of Rta and IRF4 or other cellular transcription factors may mediate their gene expression in a cell type specific manner. However, many of the genes we evaluated showed response to the novel Rta response element. Our analysis was limited to the putative promoter regions of ORF50, ORF8, ORF22, ORF63, ORF73, and M2. However, many of these genes play critical role in the biology of the virus, either as structural genes- ORF8 and ORF 22 are both surface glycoproteins, or as genes involved initiating infection- ORF63 is a tegument protein; so it is not surprising to find a significant response to Rta but lack of synergy with IRF4. Furthermore, some of the candidate genes are known to be involved in viral latency, ORF73- or murine latency associated nuclear antigen (mLANA) a homolog of EBV and KSHV LANA, has many functions including: viral replication, episomal maintenance, transcriptional regulation, and dysregulation of cell cycle and cell division (reviewed in [54]). M2, a latency associate protein appears to play roles in both maintenance and establishment of latency, as well as in viral reactivation [50], [53], [55]. We therefore find it plausible that these genes would have less stringent requirements for cell specific expression, and that other unidentified genes, might be regulated by Rta and IRF4. Additionally, due to the differing functions of these genes in MHV68 biology, we were not surprised that in our studies we found differing levels of Rta responsiveness. Future studies using genome wide analyses will be necessary to identify genes which are temporally regulated by viral and host factors including Rta and IRF4.
Our identification of a partnered interaction between Rta and IRF4 suggests a conserved method for regulating MHV68 viral gene expression. Moreover, this mechanism appears throughout the gammaherpesviruses family as several studies have shown that Rta is capable of binding DNA through orchestration of complex protein-protein interactions. In KSHV, kRta has been shown to directly interact with cellular Oct1 and RBPJκ to regulate the KSHV bZip promoter [43]. This interaction with RBPJκ is maintained through a tetrameric protein complex of kRta flanking RBPJκ, and is mediated through a core “CANT” DNA repeat element found in the Mta promoter sequence [42]. Notably, kRTA has also been found to interact with viral IRF4 (vIRF4), one of several viral IRF homologs encoded by KSHV which in the case of vIRF4 is involved in counteracting innate antiviral defenses mediated by interferons to regulate vIRF1, vIRF4, PAN, and ORF57 gene expression [44].
In summary, the data reported here defines the timing and location of M1 expression during in vivo infection using a recombinant reporter virus – demonstrating that M1 is predominantly expressed from plasma cells. Furthermore, M1 gene transcription in plasma cells is driven by the viral immediate-early Rta in conjunction with cellular IRF4 – which potently synergize with each other to activate M1 gene transcription. Whether other viral (and perhaps cellular genes) are co-regulated by Rta and IRF4 remains to be determined, and will be the topic of future work. However, we find it interesting to speculate that this might be an effective strategy to target viral replication-associated gene expression in plasma cells.
This study was carried out in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Emory University Institutional Animal Care and Use Committee and in accordance with established guidelines and policies at Emory University School of Medicine (Protocol Number: YER-2002245-031416GN).
Six to eight week old female C57Bl/6 mice were obtained through Jackson Laboratory (Bar Harbor, ME) and housed at Emory University in accordance with university guidelines. Prior to infection mice were sedated with isofluorane and intranasally infected with 5×105 pfu in 20 ul of DMEM.
Cells were grown under normal conditions at 37°C with 5% CO2. A20-HE2 cell were grown in complete RPMI-1640 (supplemented with 10% FCS, 100 U/mL penicillin, 100 mg/mL streptomycin, 2 mM L-glutamine, and 50 mM β-mercaptoethanol); P3X63Ag8 (ATCC TIB-9) were grown in compete RPMI-1640 with the addition of 10 mM non-essential amino acids, 1 mM sodium pyruvate, and 10 mM HEPES; and 293T cells (a generous gift from Dr. Edward Mocarski) were grown in complete DMEM (supplemented with 10% FCS, 100 U/mL penicillin, 100 mg/mL streptomycin, and 2 mM L-glutamine).
To generate the M1 promoter driven YFP virus a 500 bp homology arm immediately upstream of M1 ORF was amplified with LFA_MluI_1521-1573 (5′-TCCCCAATGACGCCAAAGTCTAAGTCCCTGTACAGGCTTAACTTTTTTAGAAT-3′) and LFA_SpeI_2005-2022 (5′-GGTCGCCGCTGCTCAATG-3′) and cloned into pCR Blunt-eYFP vector (a kind gift from Dr. Chris Collins) using MluI and SpeI to generate pCR Blunt-eYFP M1 LFA Flank. Next a 495 bp homology arm immediately downstream of the M1 ORF was PCR amplified using RFA_NotI_3286-3307 (5′- GCCTGAATACATGTTTACTGGG-3′) and RFA_NsiI_3758-3780 (5′- AACCTACGCGGCCACTCAACAGA -3′) was cloned into pCR Blunt-eYFP using NotI and NsiI to create pCR Blunt-eYFP M1 LFA RFA flank. The eYFP flanked by left and right homology arms for the M1 locus was then PCR amplified to include BglII and NsiI restriction sites and was cloned into pGS284. The resulting plasmid pGS284-eYFP M1 LFA RFA flank was then electroporated into λPir electro-competent bacteria for allelic exchange with WT MHV68 BAC in GS500 RecA+ Escherichia coli. To generate the M1st-eYFP virus, GS500 containing M1st. BAC [2] were crossed with λPir containing pGS284-XL9CD-CMV-YFP-F for allelic exchange. Following allelic exchange virus preparation was performed as previously described [23].
Single cell suspensions of splenocytes were prepared and resuspended in PBS supplemented with 2% fetal bovine serum. Samples were stained using standard procedures. Following initial FC receptor block (CD16/32), samples were stained with a master mix containing: CD138-PE, CD3e-PerCP, CD95-PE, GL7-APC, B220-APC-Cy7, CD19-Pacific Blue. 1–2×106 events were recorded on BD LSRII flow cytometer and results were analyzed using FloJo software (Tree Star Inc).
A20-HE2 cells were stimulated with 20 ng/mL tetradecanoylphorbol acetate (TPA) for 48 hours prior to RNA isolation. NIH3T12 were infected with an MOI of 1.5 for 18 hours prior to RNA isolation. RNA was extracted using Trizol Reagent (Invitrogen Life Technologies) according to manufacturer's instructions and RNA concentration was determined. Prior to RACE analysis RT-PCR was performed to detect M1 and pol transcripts using primers described previously [56]. 5′ and 3′ RACE analysis was performed using GeneRacer Kit L1502-02 (Invitrogen Life Technologies) according to manufacturer's specifications. Gene specific primers were generated for detection of M1 transcript. For the first round of PCR M1ORF_Rd1_Fwd (5′-GGCCATTATGTGGACGTGAAGAGAATTGTAGGTAT-3′) was used to amplify the 3′ region and M1ORF_Rd1_Rvs (5′-CCTTGGTATCATCCTCAGGAAATGGGTAGGTTTCA-3′) was used to amplify the 5′ region. For the second round of PCR M1ORF_Rd2_Fwd (5′-GGAAAACTCTCCAGAGCTGCTGTCGTG GGGGATGAT-3′) was used to amplify the 3′ region and M1ORF_Rd2_Rvs (5′-GCCAGTGAGCTATGCTTTGGCCCAGTATGCAGGAA-3′) was used to amplify the 5′ region.
M1 promoter luciferase constructs were cloned into pGL4.10 (Promega) using BglII and KpnI restriction sites. With the exception of the M1pIRF4mut1, M1pIRF4mut2, and G50ppRREm binding site mutants, inserts were generated by PCR amplification of regions upstream of the M1 ORF, using WT BAC DNA as template, with primers listed in table 1. The M2 promoter construct was the generous gift from Shariya Terrell.
To generate M1pIRF4mut1 and 2 Overlapping PCR was used to introduce IRF4/IRF8 binding site mutations into the 197 bp M1 promoter region corresponding to nt. 1960–1961 and nt. 1961–1963 in the viral for mutants 1 and 2 respectively. Amplification of a 118 bp left flaking arm was done using 197 bp forward primer (table 1) and reverse primers: (5′-TCTTTCTTGGTGTGTTCACTTCTAAACATG-3′) and (5′-TCTTTCTTGGTGGGACCACTTCTAAACATG -3′) for mutants 1 and 2 respectively. Amplification of a 70 bp right flanking arm was done using 197 bp reverse primer (table 1) and forward primers: (5′-CATGTTTAGAAGTGAACACACCAAGAAAGA-3′) and (5′-CATGTTTAGAAGTGGTCCCACCAAGAAAGA-3′) for mutants 1 and 2 respectively. The left and right flanking arms were used as template and allowed to anneal for 6 rounds of the PCR cycle prior to the addition of the 197 bp forward and reverse primers. The resulting amplicon was then cloned into pGL4.10.
To generate the ORF50ppRREm MHV68 WT BAC DNA was used as a template for overlapping PCR. In the first PCR round left and right flanking arms were generated using ProxPromF (5′-GATCGCTAGCTCTTTATAGGTACCAGGGAA-3′) with ProxRREmR (5′-tcactctgttcaagaagttgcctgaggttcataaa-3′), and ProxPromR (5′-TAGCAGATCTGGTCACATCTGACAGAGAAA-3′) with ProxRREmF (5′-ttcattttcaggccatttatgaacctcaggcaact-3′) respectively. These products were used as a template for a second round PCR amplification with primers ProxPromF and ProxPromR, and amplicons were cloned into pGL4.10.
Expression constructs were cloned into pCDNA 3.1 (+) (Invitrogen) using NotI and XhoI restriction sites using primers listed in table 1. Both flag-tagged and non-tagged unspliced Rta were amplified from WT BAC DNA corresponding to viral genomic coordinates (66,761–69,374). Murine IRF4 was amplified from pMSCV-IRF4-IRES-GFP (a kind gift from Dr. Xiaozhen Liang).
All PCR amplification was carried out using high fidelity Phusion DNA polymerase (New England Biolabs), and sequence analysis confirmed completed plasmid constructs (Macrogen).
5×105 293T cells were seeded into 6 well plates, the following day cells were transfected with 2.5 ug firefly luciferase and protein expression plasmids and 10 ng of pRL-TK (Promega) using TransIT 293T (Mirus). 1×106 P3X63Ag8 cells were nucleofected with 5 ug firefly luciferase plasmids using Ingenio Electroporation Solution (Mirus) using setting X-01 on the Amaxa nucleofector. Reactions were done in triplicate for each condition, and 2–4 independent experiments were conducted. 48 hours later cells were lysed using passive lysis buffer (25 mM Tris-phosphate pH 7.8, 2 mM DTT, 2 mM DCTA, 10% glycerol, 1% Triton X-100). P3X63Ag8 cells were assessed for firefly luciferase activity using 10 µl lysate and 50 µl luciferase assay reagent (LAR) (75 mM HEPES pH 8, 4 mM MgSO4, 20 mM DTT, 100 µM EDTA, 53.0 µM ATP, 270 µM Coenzyme A, and 470 µM beetle Luciferin). A dual luciferase assay for firefly and renilla luciferase activity was performed on 293T cells using 10 µl cell lysate and 50 µl LAR, followed by the addition of 50 µl Stop & Glo reagent (Promega). Light units were read on a TD-20/20 luminometer.
Nuclear extracts of P3X63Ag8 cells grown under normal conditions were made as previously described [57]. Briefly, cells were washed with PBS, pelleted cells, resuspended in ice cold hypotonic lysis buffer and incubated on ice for 15 minutes. 10% Nonidet P-40 was added at 1/20 final volume and nuclei were spun down. Nuclei were then washed in hypotonic lysis buffer, resuspended in high salt buffer, and incubated with vigorous shaking for 2–3 hours at 4°C. Supernatants were collected following centrifugation and aliquoted on dry ice and stored at −80°C. Following isolation protein content in nuclear extract was quantified using DC Protein Assay (BioRad), and western blot was performed to confirm presence of IRF4.
Electrophoretic mobility shift assay was performed using nuclear extracts as previously described [58]. Briefly, a binding reaction containing 10 µg of nuclear extract, 0.2 ng 32P-labeled double stranded oligonucleotide probe containing IRF4 consensus binding sequence (underlined) (sense-5′-TTGGTGGTTTCACTTCTAAACA-3′), and 2 ug poly (di-dC) was made up in binding buffer (10 mM Tris-HCl (pH 7.5), 10 mM HEPES, 50 mM KCl, 1.1 mM EDTA, and 15% glycerol, with 1.25 mM DTT) and incubated on ice for 30 minutes. Supershift assays included 1 ug of IRF4 antibody (clone M17, Santa Cruz Biotech.) or isotype control pSTAT1 antibody (clone Tyr 701, Santa Cruz Biotech.) incubated with nuclear extracts slow shaking at 4°C for 1 hour. Competition experiments were performed with 2X and 20X unlabeled oligonucleotides containing WT or mutated (underlined) IRF4 consensus binding sequence (sense 5′-TTGGTGGGACCACTTCTAAACA-3′). Nucleoprotein complexes were run on 5% native polyacrylamide gel in 0.5X Tris Buffered EDTA at 180 V for 1 hour. Gel was dried under vacuum and analyzed with PhosphorImager analysis (Typhoon 9410; Amerisham Bioscience).
10 cm dishes were seeded with 4×106 293T and were transfected the next day using TransIT-293T (Mirus). 48 hours later cells were washed 2 times with ice cold PBS, and lysed while rocking at 4°C, in 1 mL Triton X Lysis Buffer (50 mM Tris HCl pH 7.4–7.5, 150 mM NaCl, 1 mM EDTA, 0.1% Triton; supplemented with 1 mM NaF, 1 mM activated Na3V04, and Roche EDTA free protease inhibitor cocktail tablet for 50 mL volume). Following lysis, membranes were pelleted and lysates were transferred to pre-chilled tubes. Protein concentration was determined using DC Protein Assay (BioRad). 1 mg of cell lysate was precleared with prepared protein G beads (Pierce), then 8 ug of IRF4 antibody (clone M17, Santa Cruz Biotech) or flag antibody (clone M2, Sigma) was added and lysates were incubated overnight at 4°C. Lysates were transferred to freshly prepared protein G beads for binding and were incubated at 4°C for 2 hours. Following wash, protein was eluted and samples were electrophoresed on 10% polyacrylamide gels, and transferred onto nitrocellulose membranes for western blot. The following detection antibodies were used: IRF4 (clone H140, Santa Cruz Biotech.) and Flag (clone M2, Sigma).
|
10.1371/journal.pntd.0001970 | Chronic Helminth Infection Does Not Exacerbate Mycobacterium tuberculosis Infection | Chronic helminth infections induce a Th2 immune shift and establish an immunoregulatory milieu. As both of these responses can suppress Th1 immunity, which is necessary for control of Mycobacterium tuberculosis (MTB) infection, we hypothesized that chronic helminth infections may exacerbate the course of MTB.
Co-infection studies were conducted in cotton rats as they are the natural host for the filarial nematode Litomosoides sigmodontis and are an excellent model for human MTB. Immunogical responses, histological studies, and quantitative mycobacterial cultures were assessed two months after MTB challenge in cotton rats with and without chronic L. sigmodontis infection. Spleen cell proliferation and interferon gamma production in response to purified protein derivative were similar between co-infected and MTB-only infected animals. In contrast to our hypothesis, MTB loads and occurrence and size of lung granulomas were not increased in co-infected animals.
These findings suggest that chronic filaria infections do not exacerbate MTB infection in the cotton rat model. While these results suggest that filaria eradication programs may not facilitate MTB control, they indicate that it may be possible to develop worm-derived therapies for autoimmune diseases that do not substantially increase the risk for infections.
| Tuberculosis prevalence is high in areas that are endemic for helminths, suggesting that many people are chronically infected with both pathogens. As parasitic helminths can suppress the host immune system to facilitate their own survival, they frequently impact the host immune response to bystander antigens. Thus, while helminth infections ameliorate allergies and autoimmune diseases, they also decrease immune responses elicited by vaccines. Several studies have shown that helminth exposure impairs Mycobacterium tuberculosis-specific immune responses, raising the possibility that helminth infections may decrease the host's ability to control M. tuberculosis infection. To test this, we analyzed whether chronic infection of cotton rats with the filarial worm Litomosoides sigmodontis exacerbates the course of M. tuberculosis infection. Cotton rats are an excellent model organism to study human M. tuberculosis as they develop, in contrast to mice, distinct granuloma formation during infection. In addition, cotton rats are the natural host for L. sigmodontis, a nematode that establishes long-lived infections (>2 years) with circulating microfilariae in these animals. The results of this study demonstrate that chronic filarial infection does not exacerbate M. tuberculosis-associated pathology or mycobacterial burdens in cotton rats and suggest that filaria-induced immunoregulation can be overcome to respond effectively to newly acquired infections.
| Tuberculosis and helminth infections affect approximately one third of the world's population. The geographic distributions of both diseases overlap substantially, making co-infections with both pathogens common.
In contrast to infections with most bacterial, viral, protozoan, and fungal pathogens, chronic helminth infections are associated with Th2 immune responses characterized by eosinophilia, elevated IgE levels, and production of type 2 cytokines such as IL-4, IL-5, and IL-13 [1]. Over time, however, chronic helminth infections induce immunoregulatory networks through regulatory T cells, alternatively activated macrophages, and the inhibitory cytokines IL-10 and TGFβ [1]. The effects of these immune responses on the host are complex. While helminth-induced immunoregulation enhances parasite survival in the host, it also impacts the immune response to bystander antigens. As a benefit to the host, helminth-induced immunoregulation appears to play a role in protection against allergies and autoimmune diseases [2], [3], . Negatively, though, helminth infections hamper the development of adequate immune responses to vaccines like BCG, tetanus toxin, and cholera vaccine [8], [9], [10].
As infections with Mycobacterium tuberculosis (MTB) require a protective IFNγ-driven Th1 immune response [11], and as both Th2 and immune regulatory responses induced by helminths can suppress Th1 immunity, it has been hypothesized that helminth infections may impair development of a protective immune response against MTB [12], [13]. To date, however, the clinical impact chronic helminth infections have on co-infections with organisms such as MTB, Plasmodium, or HIV is controversial and not sufficiently understood [13].
The primary limitation of experimental helminth and mycobacteria co-infection studies reported to date is the utilization of mouse models of mycobacterial infection. In humans, mycobacterial infections routinely result in the development of distinct granulomas with central caseating necrosis. The formation of these granulomas in humans is believed to be necessary for immunologic control of the bacteria. Murine mycobacterial infections, however, develop diffuse infection patterns without well-formed granulomas. Recently, this limitation in rodent models of mycobacteria has been overcome with the development of the cotton rat model of MTB infection. In this system, granulomas consist of macrophages that surround the bacteria and exhibit central caseous necrosis similar to human granulomas [14]. In addition to being a useful model for mycobacterial disease, cotton rats are the natural host for the long-lived filarial nematode Litomosoides sigmodontis [15]. The adults of this parasite live in the pleural space and, after 7–8 weeks, release their offspring, the microfilariae, which circulate in the blood.
For our experiments, cotton rats chronically infected with L. sigmodontis and uninfected controls were challenged intranasally with MTB and nine weeks later euthanized to evaluate PPD-specific splenocyte proliferation and IFNγ production, lung histology, and bacterial load in the lung and spleen by quantitative culture.
Experiments were performed with 6–8-week-old female cotton rats (Sigmodon hispidus) that were obtained from Virion Systems, Inc. and maintained at the Uniformed Services University of the Health Sciences (USU) animal facility. The cotton rats are considered inbred since they have been brother sister mated in excess of 20 generations. Animals were housed individually and obtained water and food ad libitum.
Cotton rats were infected by subcutaneous injection with 100 infectious L. sigmodontis L3 larvae in media (RPMI-1640, Mediatech) as previously described [16]. After the development of a chronic L. sigmodontis infection at 11 weeks post infection, a subset of helminth-infected animals and uninfected controls were challenged with 5×104 M. tuberculosis bacteria (H37Rv) by intranasal inoculation.
The M. tuberculosis strain H37Rv was originally obtained from the Institute Pasteur, Paris, France (a kind gift of Prof. G. Marchal) and is now maintained at Sequella, Inc.. Mycobacteria stock was prepared by suspending Mycobacteria in 7H9 broth supplemented with bovine serum albumin (BSA), dextrose, and catalase. The mycobacterial suspension was cultured two successive times in roller bottles at 37°C for 7 days. The final culture was washed in PBS with 0.05% Tween 80, resuspended in PBS with 0.01% BSA and 0.05% Tween 80, aliquoted, and frozen at −80°C. CFU of the frozen aliquots were determined after thawing by plating serial 10-fold dilutions on 7H10 agar.
Animal experiments were performed under a protocol approved by the USU Institutional Animal Care and Use Committee.
L. sigmodontis worms reside in the pleural cavity where they molt into adult worms around 30 days post infection [15]. Around 8 weeks post infection, microfilariae, the offspring of adult worms, are released and enter the peripheral blood. Peripheral blood microfilaria counts were performed as described previously [16] at eleven weeks (immediately before MTB challenge) and at the end of the study, twenty weeks post L. sigmodontis infection. In brief, 10 µl of peripheral blood was obtained and added to 1 ml of ACK lysis buffer (Quality Biological, Inc.). After centrifugation the supernatant was removed and the remaining pellet was completely analyzed for microfilariae numbers by microscopy. Those numbers were divided by 10 to obtain microfilariae per µl of peripheral blood. As the microfilarial burden was too high to count at 20 weeks post L. sigmodontis infection, we resuspended the microfilaria-containing pellet in 100 µl PBS and evaluated microfilaria levels in 10 µl of this suspension to obtain total numbers of microfilariae per µl of peripheral blood. Adult worms were enumerated 20 weeks after helminth infection by careful removal from the pleural cavity using a dissection probe.
Peripheral blood was obtained by orbital bleeding or puncturing the inferior vena cava following laparotomy under a lethal dose of sodium pentobarbital from 5-week and 11-week L. sigmodontis infected cotton rats and uninfected controls. Automated differential cell blood counts were performed using a Bayer Advia 120 differential leukocyte counter.
At different time points after L. sigmodontis infection (5, 11, 20 weeks post infection) cotton rats were euthanized and spleen cells were isolated. Single cell suspensions were obtained (0.22 µm filter, BD Bioscience) and red blood cells lysed (ACK lysis buffer). Spleen cell proliferation and IFNγ production was determined from 2×106/ml spleen cells cultured with 20 µg/ml M. tuberculosis Tuberculin PPD (Statens Serum Institut), 20 µg/ml crude L. sigmodontis adult worm antigen (LsAg, prepared as previously described [17]), 10 µg/ml Staphylococcus enterotoxin B (SEB, Toxin Technology, Inc.), or cell culture media alone (Iscove's modified Dulbecco's media (Mediatech), 10% FCS (Valley Biomedical), 1% L-glutamine (Mediatech), 1% insulin-transferrin-selenium (Invitrogen Inc.), 1% penicillin-streptomycin (Mediatech)). After 48 h BrdU was added for 16 h and cellular proliferation subsequently determined according to the manufacturer's recommendations (Roche Diagnostics GmbH). In parallel cultures, IFNγ production was determined in cell culture supernatants after 72 h using a cotton rat specific ELISA according to the manufacturer's recommendations (R&D Systems, Inc.).
For microscopic evaluation of histopathology, the left lung was inflated through the trachea to its normal volume with 10% buffered formalin and sections were stained with hematoxylin and eosin (H&E). Modified acid-fast tissue stain was used to confirm the presence of acid-fast bacilli. The area of the lung containing granulomas was estimated in a blinded fashion by a single investigator (VGH).
Colony-forming units (CFUs) were assessed from equal amounts of homogenized tissue of spleen and the right lung and plated in 10-fold serial dilutions on 7H10 agar.
Statistical analysis was performed using GraphPad Prism software (GraphPad Software). Differences between multiple groups were tested for significance using the Kruskal-Wallis test followed by Dunn's post-hoc multiple comparisons. Differences between two groups were tested for significance with the Mann-Whitney-U-test. Correlations were tested using the spearman test. Impact of chronic L. sigmodontis infection on MTB was investigated in two independent experiments. Results are shown as representative examples from one experiment.
To confirm that L. sigmodontis infection induces a Type 2 immune response and a hyporesponsive milieu in cotton rats, we infected cotton rats with 100 L3 larvae and analyzed eosinophil counts, spleen cell proliferation, and IFNγ production 5 and 11 weeks after infection. Those time points were chosen as they reflect acute infection, with adult worms present in the pleural cavity prior to the release of microfilariae, and chronic infection after onset of microfilaria release into the circulation. As seen in figure 1A, L. sigmodontis infection of cotton rats induces a substantial increase in numbers of peripheral eosinophils at both 5 and 11 weeks post infection, though the differences between these timepoints and uninfected cotton rats did not reach statistical significance.
While L. sigmodontis antigen induced non-specific proliferation of splenocytes from uninfected cotton rats, a trend towards increased proliferation in response to parasite antigen was observed in splenocytes from infected animals (Fig. 1B). Splenocytes from 11 week, but not 5 week, L. sigmodontis infected cotton rats exhibited significantly reduced proliferation in response to stimulation with SEB, a superantigen which induces polyclonal activation of T-cells, compared to uninfected controls (Fig. 1C).
Parasite-specific IFNγ production from splenocytes was reduced in animals that were chronically infected for 11 weeks with L. sigmodontis compared to uninfected and 5 week-infected animals (Fig. 1D). While splenocytes from 5 week-infected cotton rats produced more IFNγ in response to SEB than uninfected animals, by 11 weeks of infection IFNγ production from splenocytes had decreased compared to 5 weeks (Fig. 1E). These findings, in addition to the decreased proliferation induced by SEB at 11 weeks, are consistent with the development of an immune regulated state during chronic filariasis.
We tested whether chronic helminth infection exacerbates MTB infection by infecting cotton rats with 100 L. sigmodontis L3 larvae. Subsets of these animals and uninfected controls were challenged 11 weeks later by intranasal inoculation of 5×104 M. tuberculosis bacteria and euthanized 9 weeks later (Fig. 2). This experiment was conducted twice.
During the first co-infection experiment, four out of 24 cotton rats that were infected with L. sigmodontis died before the MTB challenge. Two additional cotton rats that received MTB-only challenge died 3 and 5 days post MTB inoculation. No animals died at later timepoints and none of the co-infected animals died during the experiment. In the second experiment two out of eight L. sigmodontis-only infected cotton rats (18 weeks post L. sigmodontis infection), but none of the MTB only or co-infected cotton rats, died post L. sigmodontis infection. We assume that the observed death of cotton rats was due to the duration of the experiments rather than as a consequence of excessive filarial or MTB burden. These deceased cotton rats were not included in the analysis. As no co-infected animals died, inclusions of these animals in the final analysis would have only strengthened our ultimate conclusion that chronic helminth infection does not hinder control of MTB.
At study endpoint, histopathology clearly demonstrated successful infection of cotton rats with MTB and L. sigmodontis in all animals. 9 weeks after the challenge with MTB, lungs from cotton rats showed macroscopic granuloma formation (Fig. 3A) with central necrosis (Fig. 3B) and presence of acid-fast stained bacteria (Fig. 3C). Infection with L. sigmodontis was confirmed by the occurrence of microfilariae in the peripheral blood (Fig. 3D) and the presence of adult worms in the pleural space adjacant to the lungs (Fig. 3E).
Cotton rats are the natural host for L. sigmodontis and develop chronic infections. Infectious L. sigmodontis larvae migrate after the infection to the pleural cavity and molt into adult worms around 30 days post infection [15]. Microfilariae, the offspring of adult worms, start to be released 8 weeks post infection and circulate in the blood. Peripheral microfilaria counts obtained immediately before the challenge with MTB, 11 weeks post L. sigmodontis infection, revealed similar microfilaria levels between both groups (co-infection group: range 0–380 microfilariae/µl, median 180; L. sigmodontis-only group: range 82–394, median 152 p = 0.54, data not shown).
The first time we conducted the experiment co-infected cotton rats had significantly fewer microfilariae and adult worms at study endpoint than those in the helminth-only group (median co-infected: 15 adult worms (range 3–27), 540 microfilariae/µl (range 0–4492) vs. median single infected: 32 adult worms (range 15–41), 1163 microfilariae/µl (range 580–6450), Fig. 4A,B). The two cotton rats that did not produce detectable microfilaraemia at study endpoint each had three living adult worms at the end of the experiment and thus were included in the analysis.
However, the reduced adult worm and microfilaria burden in co-infected cotton rats did not occur in the repeat experiment (median co-infected: 71.5 adult worms (range 36–90), 1288 microfilariae/µl (763–3090) vs. median single infected: 54.5 adult worms (39–82), 1416 microfilariae/µl (1287–2146), Fig. 4C, microfilaria counts not shown). These results suggest that MTB co-infection does not have a consistent impact on burden of L. sigmodontis infection in cotton rats.
As IFNγ-driven Th1 immune responses are considered necessary for protection against MTB infection, we tested whether spleen cells from helminth co-infected cotton rats produce less IFNγ in response to PPD compared to cells from MTB-infected controls. In vitro stimulation of splenocytes demonstrated that both MTB-challenged groups produced significantly more IFNγ in response to PPD than uninfected and L. sigmodontis-only infected animals (IFNγ in pg/ml: uninfected median 1610 (range 265–9230), L. sigmodontis-only 520 (0–7284), MTB-only 13902 (6106–19540), co-infected 10745 (2356–28580), Fig. 5A). Importantly, IFNγ production in co-infected animals was not different than that of cotton rats infected with MTB-only and did not correlate with adult worm (r = −0.182) or microfilaria burden at study endpoint (r = 0.027). Similarly, in the repeat experiment co-infected cotton rats exhibited no reduced PPD-specific IFNγ production from splenocytes compared to MTB-only infected animals, though for all groups levels of PPD-specific IFNγ were lower in the 2nd experiment (data not shown). The total capacity to produce IFNγ from spleen cells was not changed 20 weeks post L. sigmodontis infection or 9 weeks after MTB infection, as all groups studied showed similar levels of IFNγ in response to SEB (IFNγ in pg/ml: uninfected median 13060 (range 5740–20990), L. sigmodontis-only 19000 (13120–33280), MTB-only 19600 (8640–34280), co-infected 17720 (8560–27800), Fig. 5B). Adult worm numbers tended to be negatively correlated with SEB-induced IFNγ levels (r = −0.5) whereas microfilariae levels had no clear impact (r = −0.22). SEB-induced IFNγ release from splenocytes of uninfected, L. sigmodontis-only, and co-infected animals were also similar in the repeat experiment, though MTB-only challenged cotton rats had significantly reduced IFNγ levels compared to uninfected controls (IFNγ in pg/ml: uninfected median 7920 (range 1430–10451), L. sigmodontis-only 6274 (2720–7930), MTB-only 1479 (0–5278), co-infected 4383 (1643–7587), data not shown).
Although chronic helminth infections induce a suppressive, hyporesponsive milieu in their hosts that reduces antigen-specific cell proliferation and can affect the immune response to bystander antigens, helminth co-infection did not reduce PPD-specific spleen cell proliferation during active infection with MTB. Spleen cells from co-infected cotton rats proliferated at least as well as splenocytes from MTB-only challenged animals in response to PPD, and both groups showed significantly increased proliferation rates compared to uninfected or L. sigmodontis-only infected animals (proliferation as stimulation index (OD of stimulated cells/baseline): uninfected median 1.65 (range 1.27–2.20), L. sigmodontis-only 1.19 (0.85–1.79), MTB-only 4.25 (1.82–9.92), co-infected 6.77 (2.04–15.34), Fig. 6A). PPD-induced spleen cell proliferation indices in the repeat experiment were low and not significantly increased in MTB-only challenged or co-infected cotton rats compared to uninfected or L. sigmodontis-only infected animals (uninfected median 1.23 (range 0.64–2.29), L. sigmodontis-only 1.00 (0.64–1.50), MTB-only 1.30 (0.81–2.86), co-infected 1.04 (0.80–2.69), data not shown).
Spontaneous and SEB-induced spleen cell proliferation were not affected by L. sigmodontis or MTB infection in the first experiment and showed similar results among the different treatment groups (proliferation as stimulation index: uninfected median 12.98 (range 10.28–19.07), L. sigmodontis-only 14.99 (9.16–22.74), MTB-only 15.46 (3.86–28.73), co-infected 15.01 (8.03–20.98), Fig. 6B). In the 2nd experiment, while SEB-induced spleen cell proliferation was not significantly different between the various groups, SEB-induced spleen cell proliferation was lowest in L. sigmodontis-only infected animals (uninfected median 6.05 (range 1.03–10.51), L. sigmodontis-only 2.50 (1.35–9.89), MTB-only 3.63 (0.88–10.66), co-infected 5.42 (3.18–7.81), data not shown).
Total spleen cell numbers were increased in the L. sigmodontis-only (median 89.5×106, range 56–110×106), MTB-only (median 105×106, range 42–130×106), and co-infected group (median 83.5×106, range 66–110×106) compared to uninfected controls (median 45×106, range 13–85×106), although this difference was only statistically significant for MTB-only infected animals (p<0.01, data not shown).
To assess whether helminth infection increases susceptibility to primary MTB infection, lungs from co-infected and MTB-only infected cotton rats were analyzed for granuloma formation and quantitative MTB cultures were conducted on lung and spleen tissues. Helminth co-infection was not associated with greater granulomatous inflammation in the lung compared to MTB-only infected animals (median granuloma area as a percentage of total lung tissue in co-infected animals = 2%, range 0–40% vs. 10% for MTB-only, 1–30%, Fig. 7A). One co-infected animal did not develop lung granulomas, whereas all MTB-only infected animals had lung granulomas. Comparable results were obtained during the repeat experiment, though in general the granuloma-covered area was higher in the repeat experiment (co-infected: median 40%, range 15–75%; MTB-only: median 40%, range 30–75%, data not shown).
Similar to granuloma formation, L. sigmodontis co-infection did not increase MTB bacterial burdens in lungs. CFUs from L. sigmodontis co-infected and MTB-only infected cotton rats were not significantly different in lung (Fig. 7B), but tended to be lower in the co-infected animals (median 1.2×106 in co-infected vs. 7.8×107 in MTB-only). There was no correlation between adult worm burdens (r = 0.103, 20 weeks post L. sigmodontis infection) or microfilaria levels (11 weeks post L. sigmodontis infection: r = 0.057; 20 weeks post L. sigmodontis r = 0.299) and MTB CFUs in the lungs. Similarly, lung CFUs from co-infected animals tended to be reduced in the repeat experiment compared to MTB-only infected cotton rats (median 3.25×107, range 1.96×106–6.33×108 vs. 2.06×108, range 1.53×107–6.91×108, data not shown). Combined results from both experiments resulted in significantly reduced CFUs in lungs of co-infected animals compared to MTB-only infected cotton rats (p = 0.027). Whereas all animals challenged with MTB-only had positive lung cultures, three co-infected animals in the first experiment had no detectable CFUs in the lung, though two of them had lung granulomas and positive spleen cultures. The co-infected animal without lung granulomas and negative spleen and lung cultures showed the strongest PPD-specific IFNγ production and cell proliferation, suggesting there had been an initial MTB infection which had been successfully cleared. CFUs in the spleen were quantified in the first experiment and were 3–4 logs lower than the ones observed in the lung. Spleen CFUs were similar between helminth co-infected and MTB-only infected cotton rats (median 1300 vs. 2900, Fig. 7C). Similar to lung CFUs, there was no correlation between adult worm burdens (r = 0.048, 20 weeks post L. sigmodontis infection) or microfilaria levels (11 weeks post L. sigmodontis infection: r = 0.062; 20 weeks post L. sigmodontis r = 0.002) and CFUs in spleen. Quantitative MTB cultures from spleens of 2/11 co-infected and 1/11 MTB-only infected animals were negative.
In contrast to the hypothesis that chronic helminth infections worsen the course of MTB, the results of this study demonstrate that chronic filarial infection does not alter control of MTB in the cotton rat. Histological examinations and quantitative MTB cultures from two independent experiments clearly demonstrated equivalent or reduced mycobacterial burden in co-infected animals compared to those infected with only MTB. These findings are supported by immunological studies revealing that PPD-specific cellular proliferation and IFNγ production were not suppressed in co-infected animals.
These results are unexpected since it is documented that chronic helminth infection can alter the immune response to bystander antigens. Indeed, recent studies have shown that chronic filarial infection is associated with decreased PPD-specific IFNγ and IL-17 responses in individuals latently infected with tuberculosis [18]. Similarly, active filaria infection in patients latently infected with MTB correlates with a reduction in TLR2 and TLR9 activation in response to MTB antigens that normalizes after anti-filarial treatment [19].
The results of our study, however, suggest that systemic filaria-induced immunomodulation can be overcome in the setting of an active MTB infection. Immunomodulatory effects in cotton rats chronically infected with L. sigmodontis were confirmed in our model. A time course study showed that cotton rats infected with L. sigmodontis developed eosinophilia, which correlates with the induction of a Type 2 immune response. Additionally, splenocytes of cotton rats infected for 11 weeks exhibited reductions in parasite-specific cytokine production and splenocyte proliferation in response to polyclonal activation.
Among previous in-vivo co-infection studies utilizing helminths and mycobacteria, two showed no effect of helminths on mycobacterial infection and three observed worsened control [20], [21], [22], [23], [24]. The first two studies which demonstrated negative impact used an intravenous Mycobacterium bovis infection challenge into mice either chronically infected with Schistosoma mansoni [23] or acutely infected with Strongyloides venezuelensis [20]. A possible explanation for the observed difference between those studies and ours may be that distinct helminths have different effects on mycobacteria co-infection. Filariae, strongylids, and schistosomes all reside in different tissue spaces, have markedly different lifecycles, and release different excretory/secretory factors. For example, it has been shown that in vitro exposure of human dendritic cells to microfilariae of the human filaria Brugia malayi results in decreased expression of DC-sign, a receptor for MTB [25], providing a potential mechanism by which filariae may actually have some host-protective effects against tuberculosis.
In addition to the different effects parasites may induce on host cells, the anatomical niche used by helminths inside the host may be important for impacting the immune response to mycobacteria. The helminth utilized in our study lives in close proximity to MTB. L. sigmodontis adult worms live in the pleural space abutting the lungs, and microfilariae enter the peripheral blood via the lung capillaries and regularly transit through the spleen. Thus, there is potential for L. sigmodontis to exert local effects on MTB co-infection. For example, the influx of cells into the pleural cavity induced by adult L. sigmodontis worms could potentially facilitate clearance of MTB bacteria.
Alternatively, it is possible that the differences observed between the M. bovis models [20], [23] and ours were due to the different mycobacterial models used. One of the strengths of our study was the use of the cotton rat model of MTB infection. Unlike murine mycobacterial models, MTB infection of cotton rats results in discrete granulomas containing macrophages, mycobacteria, and central necrosis similar to that observed in human tuberculosis. MTB and schistosome co-infection in the cotton rat may reveal whether individual helminths have different effects on the course of mycobacterial infection.
The third study that has shown a negative impact of helminth infection on control of mycobacteria utilized Nippostrongylus brasiliensis infection in mice [21]. In this study, acute infection with 500 tissue-invasive N. brasiliensis larvae transiently worsened control of M. tuberculosis infection in an acute setting [21]. The contrasting outcomes of this study and ours are likely due to differences in the helminth models as well as the timing of the MTB challenge. Whereas N. brasiliensis L3 infections induce a short-lived infection in mice, chronic L. sigmodontis infection persists in cotton rats for years. Thus, it can be assumed that L. sigmodontis worms are better adapted to the immune system of their natural host, the cotton rat. As such, chronic L. sigmodontis infection in cotton rats is likely a good immunologic model for long-term persistent human filarial infections. Another key difference between the N. brasiliensis model and ours is the timing of MTB infection. Whereas MTB challenge was given only days after N. brasiliensis infection, when type 2 immune responses are increasing, we challenged rats with MTB 11 weeks after helminth infection, a timepoint at which chronic helminth infection and immunoregulatory responses have become established. Whether acute L. sigmodontis infection imparts a transient reduction in control of MTB infection similar to N. brasiliensis is not known and may be the topic of future studies.
In addition to the in vivo animal studies that showed a negative impact of helminths on mycobacterial infection, Elias et al. showed that acute MTB infected patients had an increased frequency of helminth infection compared to MTB negative household contacts [26]. This discrepancy with our results may also be due to differences in the helminth species present in the hosts. While in our experiments a filarial nematode was used, Elias et al. observed an increased frequency of helminth infection with Schistosomes and intestinal nematodes (hookworms, Ascaris, Trichuris, Strongyloides). In contrast, a different epidemiological study done in South India found no impact of either intestinal or filarial infection on frequencies of PPD positivity [27].
It is important to note that our study did not evaluate the effects helminth infections have on latent MTB. As the immune response required for control of latency may be different than that required for control of active disease, it may be worthwhile exploring whether chronic helminth infection alters the risk of reactivation in a latent MTB model.
Interestingly, in the first co-infection experiment we conducted adult L. sigmodontis worm numbers and microfilaria counts were significantly decreased in MTB co-infected cotton rats. While Th2 immune responses are generally considered protective against helminth infections, we speculate that the decreased worm burden was due to the pro-inflammatory environment created by the MTB co-infection, as it is known that IFNγ can contribute to resistance against L. sigmodontis [28]. In accordance with this speculation, IFNγ production from splenocytes of all groups were lower in the second experiment, correlated with a higher worm burden 20 weeks post L. sigmodontis infection, and was associated with no difference in worm burdens of co-infected and single infected groups.
In conclusion, our data demonstrates that chronic filaria infection does not exacerbate the course of acute MTB in the cotton rat model. While results of prior studies investigating the effects helminth infections have on MTB co-infection have been conflicting, we believe that the use of an animal in which the host develops granulomas to MTB in combination with a chronic helminth infection in its natural host makes this study the most likely to approximate chronic helminth infection and MTB co-infection in humans.
While our results indicate that filaria eradication programs may not have a substantial impact on MTB control, they also suggest that it may be possible to develop worm-derived therapies for autoimmune diseases which do not substantially increase the risk for severe infections. Future studies evaluating effects of different helminths utilizing the same MTB model and assessing the impact of helminths in MTB latency models will provide important insights for further understanding the effects helminth co-infections have on MTB.
|
10.1371/journal.pgen.1005390 | A Role for Macro-ER-Phagy in ER Quality Control | The endoplasmic-reticulum quality-control (ERQC) system shuttles misfolded proteins for degradation by the proteasome through the well-defined ER-associated degradation (ERAD) pathway. In contrast, very little is known about the role of autophagy in ERQC. Macro-autophagy, a collection of pathways that deliver proteins through autophagosomes (APs) for degradation in the lysosome (vacuole in yeast), is mediated by autophagy-specific proteins, Atgs, and regulated by Ypt/Rab GTPases. Until recently, the term ER-phagy was used to describe degradation of ER membrane and proteins in the lysosome under stress: either ER stress induced by drugs or whole-cell stress induced by starvation. These two types of stresses induce micro-ER-phagy, which does not use autophagic organelles and machinery, and non-selective autophagy. Here, we characterize the macro-ER-phagy pathway and uncover its role in ERQC. This pathway delivers 20–50% of certain ER-resident membrane proteins to the vacuole and is further induced to >90% by overexpression of a single integral-membrane protein. Even though such overexpression in cells defective in macro-ER-phagy induces the unfolded-protein response (UPR), UPR is not needed for macro-ER-phagy. We show that macro-ER-phagy is dependent on Atgs and Ypt GTPases and its cargo passes through APs. Moreover, for the first time the role of Atg9, the only integral-membrane core Atg, is uncoupled from that of other core Atgs. Finally, three sequential steps of this pathway are delineated: Atg9-dependent exit from the ER en route to autophagy, Ypt1- and core Atgs-mediated pre-autophagsomal-structure organization, and Ypt51-mediated delivery of APs to the vacuole.
| ER-quality control (ERQC) ensures delivery of “native” proteins through the secretory pathway. Currently, ER-associated degradation (ERAD), which delivers misfolded proteins for degradation by the proteasome, is considered a major ERQC pathway, with autophagy as its backup. Until now, the role of autophagy, which shuttles cellular components for degradation in the lysosome through autophagosomes (APs), in ERQC was ill defined. Recently, the process of ER degradation induced by ER stress was defined as micro-ER-phagy, which does not require autophagic machinery and does not pass through APs. Here, we characterize the macro-ER-phagy pathway, which delivers excess membrane proteins for degradation in the lysosome, as a novel ERQC pathway. This pathway functions in the absence of cellular or ER stress and can be further induced by overexpression of a single integral-membrane protein. Unlike the micro-ER-phagy pathway, the marco-ER-phagy pathway requires core autophagy-specific proteins, Atgs, and Ypt/Rab GTPases. In addition, for the first time, the function of the only membrane core Atg, Atg9, was uncoupled from that of the other core Atgs. Whereas Atg9 plays a role in the assembly of ER-to-autophagy membranes (ERAM), other core Atgs and Ypt1 assemble the Atg-protein complex on ERAM to form the pre-autophagosomal structure.
| One third of all newly synthesized proteins enter the endoplasmic reticulum (ER). However, only a small fraction is transported to their final destination. A large fraction (30–75%) fails to fold and mature properly, does not pass the ER quality control (ERQC) and gets degraded [1]. Two different cellular pathways shuttle proteins from the ER for degradation: ER associated degradation (ERAD) and autophagy. Whereas the importance of ERAD in ERQC has been studied extensively and is well established, not much is known about the role of autophagy in ERQC [2].
ERAD delivers proteins from the ER for degradation by the cytoplasmic proteasome. ERAD substrates include soluble and integral-membrane proteins that fail to fold properly or assemble into complexes. Substrate recognition happens in the lumen or the membrane of the ER by chaperones (e.g., BiP). These substrates are translocated back to the cytoplasm where they are ubiquitinated and degraded by the proteasome [3,4]. Under conditions that stimulate accumulation of misfolded proteins (e.g., DTT and tunicamycin, inhibitors of disulfide-bond formation and glycosylation, respectively), ER stress and the conserved unfolded-protein response (UPR) are induced. In yeast, UPR induction requires two proteins, the endonuclease Ire1 and the transcription factor Hac1, which binds to UPR elements and stimulates the transcription of ERAD machinery components [5]. Multiple human disorders have been associated with ERAD [2].
In autophagy, cargo is delivered for degradation in the lysosome (vacuole in yeast), a major recycling cellular compartment. There are three major types of autophagy: macro, micro and chaperone mediated (CMA) [6]. Macro-autophagy, the best studied type, is a collection of cellular degradation pathways in which cargo is engulfed by a double-membrane organelle termed the autophagosome (AP) that fuses with the lysosome. All macro-autophagy pathways start with the formation of the pre-autophagosomal structure (PAS), which is mediated by the core autophagy-related proteins (Atgs). PAS includes subunits of the Atg protein complex and membranes; the latter are thought to be supplied by the only integral-membrane core Atg, Atg9 [7,8]. Macro-autophagy can be nonselective, when induced by stress, or selective, e.g., cytoplasm-to-vacuole transport (CVT), mitophagy (autophagy of mitochondria), pexophagy (autophagy of peroxisomes) [8], and ER-phagy (autophagy of the ER), which is discussed here.
Micro-autophagy and CMA, about which less is known, do not require Atgs and their cargos are not delivered through APs [6]. In micro-autophagy, cargo enters the lysosome through invagination of its membrane. For example, under certain growth conditions, peroxisome clusters can enter the lysosome via micro-pexophagy [9]. Likewise, dispensable portions of the nucleus can be delivered into the lysosome via the piecemeal micro-autophagy of the nucleus [10]. CMA, which has been described so far only in mammalian cells, is highly specific and involves translocation of unfolded proteins through the lysosomal membrane [6].
Until recently, the terms ER-phagy and reticulophagy, autophagy of the ER, have been loosely used to describe any process that delivers ER for degradation in the lysosome. For example, ER was identified as one of multiple membrane sources for AP biogenesis [11]. In addition, during prolonged ER stress, nonselective macro-autophagy can be induced [12], and was suggested to serve as a backup for ERAD [4,13]. Most notably, ER-phagy induced by ER stress [14], was recently characterized as micro-ER-phagy, which does not require the macro-autophagy machinery components [15]. Thus, almost nothing is currently known about selective macro-autophagy of the ER [16]. We have recently used the term ER-phagy to describe a Ypt1 GTPase-dependent pathway through which an overexpressed membrane protein is delivered to the vacuole for degradation [17]. The major point of that paper was to clarify that accumulation of GFP-Snc1, a marker traditionally used for endosome-to-Golgi transport, in mutant cells defective in Ypt1 function is not due to a defect in endosome-to-Golgi transport, but, rather to its role in autophagy emanating from the ER. However, under what conditions this pathway is induced, the nature of its cargos and the role of core Atgs in it, are not known. This pathway is defined here as macro-ER-phagy.
The eleven yeast Ypt GTPases and their seventy Rab homologues regulate and coordinate the multiple intra-cellular trafficking pathways [18]. These GTPases are activated by their guanine-nucleotide exchange factors (GEFs) to recruit their various effectors, which in turn mediate vesicular transport steps [19,20]. At least three Ypt GTPases were implicated in regulation of autophagy. We have shown that Ypt1, a Rab1 homolog, coordinates the first steps of two different pathways that emanate from the ER: secretion and autophagy [21,22]. Ypt1 does this in the context of two distinct GTPase modules that contain different GEFs and effectors [23]. Vps21, a Rab5 homolog, regulates two different pathways, endocytosis and autophagy, in the context of the same module [24,25]. Finally, Ypt7, a Rab7 homolog, regulates fusion of APs with the vacuole [26].
We have previously shown that components of the autophagy-specific Ypt1 module, including the Trs85-containing TRAPP III GEF, Ypt1-GTPase, and the Atg11 effector, regulate delivery of an overexpressed integral plasma membrane (PM) protein from the ER to the vacuole for degradation. The membrane protein we used was GFP-tagged Snc1-PEM; a variant of the vSNARE Snc1 that cannot be internalized by endocytosis [27]. We used the term ER-phagy for this pathway [22]. However, especially in light of recent characterization of drug-induced ER-phagy as micro-ER-phagy, it is crucial to determine if and how ER-phagy induced by overexpression of an integral-membrane protein is different from other ER-phagy processes described recently, e.g., micro-ER-phagy and backup for ERAD. Here, we characterize this pathway as macro-ER-phagy, and determine that it plays a role in vacuolar recycling of some ER-resident proteins even under normal growth conditions and can be further induced by overexpression of a single integral-membrane protein. We also show that this pathway requires core Atgs, and identify its cargos. Moreover, we define three sequential steps in this pathway, which are dependent on Atg9, Ypt1 and core Atgs, and Vps21.
Budding yeast was instrumental for current conceptual understanding of intracellular trafficking [28], autophagy [8,29], and the role of Ypt/Rab GTPases in these processes [30] in human cells. Moreover, machinery components that mediate these processes are conserved from yeast to humans and are relevant to health and disease. For example, the Ypt1 human homolog hRab1A was defined recently as an oncogene [31]. Therefore, there is no doubt that the principles of macro-ER-phagy characterized here would pertain to human cells. The potential relevance of clearance of excess membrane proteins by macro-ER-phagy to human disease is discussed.
We have previously shown that mutations in components of the autophagy-specific Ypt1 GTPase module, ypt1-1, trs85∆, and, atg11∆, are defective in delivery of overexpressed GFP-Snc1-PEM, a PM integral membrane protein, from the ER to the vacuole for degradation. We termed this pathway ER-phagy [17]. Because here we identify this pathway as macro-ER-phagy, and to distinguish it from the recently characterized micro-ER-phagy [15], we will hereinafter use the term macro-ER-phagy.
To identify other autophagy machinery components involved in macro-ER-phagy, we determined whether deletions of several Atgs result in a defect in delivering overexpressed GFP-Snc1-PEM to the vacuole through this pathway, using the following three criteria: increase in the protein level of GFP-Snc1-PEM, its accumulation in aberrant ER structures, and induction of UPR. These aberrant structures were previously identified as a cluster of ER-derived membrane-bound vesicles using immune-electron microscopy and anti-Hmg1 antibodies [17]. Their molecular composition is further characterized below.
Like Atg11, deletion of the two core autophagy components, Atg1 and Atg8, result in an increase in the GFP-Snc1-PEM protein level (Fig 1A), accumulation of aberrant Snc1-PEM intracellular structures that co-localize with the ER marker Sec61 (Fig 1B), and induction of UPR (Fig 1C). While >70% of all three mutant cells accumulate aberrant GFP-Snc1-PEM intracellular structures, the levels of protein accumulation and UPR induction vary, and are lower than those observed in ypt1-1 mutant cells (see S1A Fig). Deletion of another core autophagy component, Atg17, by itself had no effect on macro-ER-phagy. However, the double mutant atg11∆ atg17∆ displayed more severe defects in all three assays than those of the atg11∆ alone, similar to that of the ypt1-1 mutation (S1A–S1C Fig). More severe defects in other autophagy types were previously observed for the double atg11∆ atg17∆ mutant than for the single deletions [32]. These results indicate that the role of Ypt1 in macro-ER-phagy is mediated by both Atg11-dependent and-independent GTPase modules.
Atg11 plays a role in all selective autophagy processes [33]. To determine whether other selective-autophagy Atgs play a role in macro-ER-phagy, we analyzed the effect of deletions of Atg19, Atg32, and Atg36, which are required for CVT, mitophagy and pexophagy, respectively, on delivery of overexpressed GFP-Snc1-PEM to the vacuole [33]. In all three assays, atg19∆, atg32∆ and atg36∆ mutant cells behaved like wild-type cells (Fig 1D–1F). However, when combined with atg11∆, all three double mutant cells behaved like atg11∆ single mutant cells (S1D–S1F Fig).
Together, these data indicate that macro-ER-phagy requires the core autophagy machinery, but not any of the known selective autophagy components other than Atg11.
Atg9, the only integral-membrane core Atg protein, is required for PAS formation in all autophagy processes [34]. Analysis of the atg9∆ effect on delivery of overexpressed GFP-Snc1-PEM to the vacuole revealed a phenotype different from that of mutations in other core Atgs and Ypt1 (shown in Fig 1). The level of GFP-Snc1-PEM protein was >3-fold higher than its level in WT cells, comparable to the increase in atg11∆ mutant cells (Fig 2A and 2D). A 2.5-fold increase was also observed in the cytoplasmic fluorescence of GFP-Snc1-PEM (S2A Fig). However, whereas ~75% of atg11∆ mutant cells accumulate GFP-Snc1-PEM in large intracellular structures, only 20% of the atg9∆ mutant cells contain small intracellular GFP-Snc1-PEM structures (Fig 2B and 2E), which co-localize with the ER marker Sec61-mCherry (S2A Fig). Moreover, UPR was not induced in atg9∆ mutant cells (Fig 2C and 2F). In addition, while overexpression of GFP-Snc1-PEM had no effect on the growth WT or atg11∆ mutant cells, it caused a growth defect in atg9∆ mutant cells (S2B Fig).
Two other Atgs, Atg2 and Atg18, affect the function of Atg9 by mediating its recycling from APs to peripheral sites [35]. The effects of the atg2∆ and atg18∆ mutations on macro-ER-phagy were determined. Whereas atg2∆ showed phenotypes similar to those of atg9∆, atg18∆ showed no effect (Fig 2A–2C). Different effects of Atg2 and Atg18 on autophagy were previously observed in Drosophila [36].
The different macro-ER-phagy phenotype of atg9∆ mutant cells allowed us to perform epistasis analyses with atg11∆, atg1∆, and ypt1-1. The level of GFP-Snc1-PEM in the double mutants atg9∆ atg11∆ and atg9∆ atg1∆ is similar to its level in the single mutants (Figs 2D and S2C, respectively). The fact that the level does not increase in the double mutant indicates that Atg9 functions in the same pathway as Atg11 and Atg1. Importantly, both the microscopy and the UPR assays show that the atg9∆ mutation “masks” the phenotypes of atg11∆ and atg1∆. Specifically, whereas >75% of atg11∆ and atg1∆ single-mutant cells accumulate large intracellular GFP-Snc1-PEM structures, small structures were observed only in <25% the double mutants, atg9∆ atg11∆ and atg9∆ atg1∆, similar to the phenotype of the single atg9∆ mutation (Figs 2E and S2D). Likewise, whereas UPR is induced in atg11∆ and atg1∆ single-mutant cells, it is not induced in atg9∆ atg11∆ and atg9∆ atg1∆ double mutants, similar to the phenotype of the single atg9∆ mutation (Figs 2F and S2E). Both atg9∆ and atg9∆ atg11∆ mutant cells are not defective in UPR induction, as UPR can be induced in these cells by tunicamycin (Fig 2G).
The macro-ER-phagy phenotype of the single ypt1-1 mutation is more severe than those of a single deletion of any Atg. Importantly, atg9∆ suppresses the ER-phagy phenotype of ypt1-1. First, the level of GFP-Snc1-PEM in ypt1-1 mutant cells is ~20-fold higher than its level in WT cells. In ypt1-1 atg9∆ double mutant cells, the level is reduced to ~5.5 fold (Fig 3A, left). Second, whereas 85% of ypt1-1 mutant cells contain large intracellular GFP-Snc1-PEM structures, only 28% of the ypt1-1 atg9∆ double mutant cells contain small structures, similar to atg9∆ mutant cells (Fig 3B, left). Third, the UPR is also lower in the ypt1-1 atg9∆ double mutant then in the single ypt1-1 mutant cells (Fig 3C, left).
Together, these results show that Atg9 functions upstream of Ypt1 and core Atgs in macro-ER-phagy. The fact that GFP-Snc1-PEM does not accumulate in large aberrant structures in atg9∆ mutant cells as it does in the other mutant cells, suggests that Atg9 plays a role in exit of macro-ER-phagy cargo from the ER. Interestingly, UPR induction seems to be dependent on the assembly of these structures.
One prediction from these results is that Atg9 would accumulate on aberrant ER structures in ypt1-1 mutant cells. To determine whether this is the case, WT and ypt1-1 cells expressing endogenously tagged Atg9-mCherry and overexpressing GFP-Snc1-PEM were analyzed by live-cell microscopy. Whereas in WT cells the Atg9 puncta do not co-localize with the PM-localized GFP-Snc1-PEM, in ypt1-1 mutant cells Atg9 co-localizes with the aberrant GFP-Snc1-PEM structures (Fig 3D). Thus, Atg9 is required for the formation of the aberrant ER structures in ypt1-1 mutant cells and is present on their membrane.
To better characterize macro-ER-phagy, we determined whether it is affected by mutations in Sec12 and Vps4, which mediate the exocytic pathway and autophagy-independent delivery to the lysosome, respectively. Sec12 mediates ER-to-Golgi transport [37], which is also regulated by Ypt1 [38]. Because in addition to their ER-to-Golgi block, sec12ts mutant cells are also defective in ERAD at their restrictive temperature [39], it was expected that they would accumulate some GFP-Snc1-PEM in their ER even at permissive temperature. Indeed, the level of GFP-Snc1-PEM is increased ~3.5 fold when compared to WT cells, 40–50% of the cells accumulate some of it in their ER, and UPR is induced (S3 Fig). However, there are three main differences between the accumulation of GFP-Snc1-PEM in sec12-ts and ypt1-1 mutant cells. First, whereas overexpression of GFP-Snc1-PEM in ypt1-1 mutant cells results in a two-fold increase of the UPR, UPR is higher in sec12ts mutant cells that do not overexpress GFP-Snc1-PEM (S3C Fig). This result suggests that UPR induction in sec12ts mutant cells is due to a defect in ER exit of multiple proteins, and triggering the macro-ER-phagy pathway by excess GFP-Snc1-PEM might partially relieve the ER stress in these mutant cells. Second, while deletion of ATG9 masks the phenotypes of ypt1-1 (e.g., GFP-Snc1-PEM structures accumulation and UPR induction), it exacerbates those of sec12ts mutation (Fig 3A–3C, right). This result indicates that Atg9 and Sec12 do not function in the same pathway. Most importantly, we have previously shown that ypt1-1 mutant cells, regardless if they are defective in vacuolar proteolysis or not, accumulate GFP-Snc1-PEM outside their vacuole [22]. In contrast, whereas there is no GFP-Snc1-PEM in vacuoles of sec12ts mutant cells, it does accumulate in the proteolysis-defective vacuoles of sec12ts pep4∆ double-mutant cells (Fig 3E). Because GFP-Snc1-PEM can get to the vacuole only through autophagy and not through endocytosis [27], sec12ts mutant cells are not defective in autophagy. Together, these results show that unlike Ypt1, Sec12 is not required for macro-ER-phagy.
Vps4 is required for transport to the lysosome from the PM or the Golgi through late endosomes [40]. To confirm that GFP-Snc1-PEM reaches the vacuole from the ER and not from the Golgi, macro-ER-phagy of overexpressed GFP-Snc1-PEM was determined in vps4∆ mutant cells. In all three aforementioned assays, vps4∆ mutant cells behave like WT cells (S4A–S4C Fig).
Recently, a micro-ER-phagy process that does not require known autophagy machinery was described [15]. This process can be induced by adding DTT or tunicamycin to cells and observed by the presence of ER membrane “whorls” in vacuole of proteolysis-defective mutants. To confirm that the Ypt1 is not required for micro-ER-phagy, DTT-dependent formation of ER-membrane whorls was compared in YPT1 and ypt1-1 mutant cells defective in vacuolar proteolysis (pep4∆ prb1∆). Accumulation of ER whorls was similar in both strains (S4D Fig), indicating that Ypt1 does not play a role in micro-ER-phagy. Therefore, GFP-Snc1-PEM accumulation in ypt1-1 mutant cells is caused by a defect in a process distinct from micro-ER-phagy.
Our finding that the core autophagy machinery is required for delivery of GFP-Snc1-PEM to the vacuole supports the idea that the process described here is macro-ER-phagy. To confirm this idea, we determined whether yDsRed-Snc1-PEM passes through APs en route to the vacuole. Recently, we showed that the Rab5 homolog Vps21 plays a role in autophagy as vps21∆ mutant cells accumulate AP structures marked by Atg8 under starvation [24]. This indicates that autophagy is blocked in vps21∆ mutant cells after the formation of APs. Under normal growth conditions, both WT and vps21∆ mutant cells have one Atg8 dot per cell, representing the phagophore or AP [41]. Co-localization of overexpressed yDsRed-Snc1-PEM with yEGFP-Atg8 was observed in ~70% of vps21∆ mutant cells, as compared to ~4% of WT cells (Fig 4A, top). This vps21∆ phenotype is also different from that of the ypt1-1 single mutation and the ypt1-1 vps21∆ double mutation. As we have previously shown, ypt1-1 mutant cells are defective in PAS formation and contain multiple Atg8 dots per cell [22,42]. Even though yDsRed-Snc1-PEM accumulates in ypt1-1 single- and ypt1-1 vps21∆ double-mutant cells, it does not co-localize with the multiple Atg8 dots (Fig 4A, bottom). The finding that the ypt1-1 mutation overrides the vps21∆ phenotype indicates that Ypt1 and Vps21 function sequentially in the same pathway. Moreover, the co-localization of yDsRed-Snc1-PEM with Atg8 in vps21∆ mutant cells indicates that it passes through APs en route to the vacuole and confirms that this pathway is macro-ER-phagy.
UPR plays a role in ERAD [5] and is induced when macro-ER-phagy is blocked [17]. To determine whether UPR is required for macro-ER-phagy, accumulation of overexpressed GFP-Snc1-PEM was tested in ire1∆ and hac1∆ mutant cells, which are defective in UPR induction. In WT (YPT1) cells, deletion of either IRE1 or HAC1 abolishes UPR, but does not result in accumulation of GFP-Snc1-PEM (Fig 4B–4D, left). In ypt1-1 mutant cells, deletion of either IRE1 or HAC1 abolishes UPR induction, but does not affect accumulation of GFP-Snc1-PEM (Fig 4B–4D, right). These results show that Ire1/Hac1-dependent UPR induction is not required for macro-ER-phagy and does not affect the formation of aberrant ER structures in ypt1-1 mutant cells.
We have shown that overexpressed GFP-Snc1-PEM, a chimeric integral PM protein, can serve as a cargo for macro-ER-phagy [17]. We wished to identify a native protein that when overexpressed would be a cargo for ER-phagy. For this purpose we chose the PM multi-drug transporter Snq2 that contains multiple trans-membrane domains. Snq2 tagged at its C-terminus with yEGFP was overexpressed from a 2μ plasmid in WT and ypt1-1 mutant cells. In ypt1-1 mutant cells, like GFP-Snc1-PEM, the level of Snq2-yEGFP was increased (by ~10-fold), it accumulated in aberrant ER structures in 80% of the cells, and UPR was induced (Fig 5A–5C). Thus, overexpressed Snq2-yEGFP, like GFP-Snc1-PEM, is transported through macro-ER-phagy for degradation in the vacuole.
To study the effect of overexpression of multiple membrane proteins on macro-ER-phagy, Snq2-yEGFP and DsRed-Snc1-PEM were co-overexpressed in WT and atg11∆ mutant cells. In WT cells, while overexpression of a single membrane protein did not have any effect, overexpression of two proteins resulted in some increase in the protein levels, and they co-localized in the vacuole of ~12% of the cells (Figs 5D and 5E and S5A). In atg11∆ mutant cells, while overexpression of a single membrane protein had a moderate phenotype, overexpression of the two resulted in a synergistic effect. The protein levels were increased >25 fold over that of WT cells (single protein levels were increased 2.5–6 fold) (Fig 5D). In 87% of the cells the two proteins co-localized in aberrant intra-cellular structures (compared to 50–65% for single protein) (Figs 5E and S5A). Moreover, the fluorescence of these intra-cellular structures was 5-fold brighter than that of each single protein (Fig 5F). Finally, whereas overexpression of a single membrane protein did not affect the growth of atg11∆ mutant cells, overexpression of two caused a growth defect (S5B Fig). These results suggest that when macro-ER-phagy is partially defective, as in atg11∆ mutant cells, the extra burden of overexpression of multiple membrane proteins can be detrimental.
Together, these results indicate that excess of integral membrane proteins, such as Snc1-PEM and Snq2, are transported to the vacuole through macro-ER-phagy.
We wished to determine which ER-resident proteins are transported together with GFP-Snc1-PEM to the vacuole via the macro-ER-phagy pathway. Five ER-resident proteins were used for this analysis: three integral membrane proteins, Sec61 (translocon subunit), Hmg1 (sterol biogenesis), and Sec12 (ER-exit sites), a cytoplasmic coat protein, Sec13 (ER exit sites), and an ER-lumen chaperon, Kar2 (BiP) (Fig 6A). Protein accumulation was determined by immunoblot and microscopy analyses in WT (YPT1) and ypt1-1 mutant cells in which vacuolar proteolysis was normal (PEP4 PRB1) or defective (pep4∆ prb1∆). The analysis was done without and with over-expression of GFP-Snc1-PEM.
We have previously shown that some resident-ER proteins accumulate in ypt1-1 mutant cells when compared to WT cells even without overexpression of GFP-Snc1-PEM. This phenotype was observed using live-cell microscopy and tagged resident ER proteins, e.g., Hmg1 and Sec61 [17]. Here, the microscopy analysis is supported by an immunoblot analysis. The levels of Sec61 and Hmg1 were determined in cells that do not overexpress GFP-Snc1-PEM. In YPT1 (WT) cells defective in vacuolar proteolysis (pep4∆ prb1∆), the levels of Sec61 and Hmg1 were increased by 1.2–1.6 fold. In ypt1-1, regardless if the vacuolar proteolysis is normal or defective, the levels of Sec61 and Hmg1 were increased by ~2–2.6 fold (Figs 6B and 6D and S6A). In contrast, the levels of Sec12, Sec13 and Kar2 were not increased in these mutant cells (Figs 6C and 6D and S6B and S6C). Therefore, ~20–40% of some ER resident proteins, but not all, are shuttled to the vacuole for degradation in WT and ypt1-1 mutant cells even in cells that do not over-express GFP-Snc1-PEM.
The previously observed Hmg1 and Sec61 accumulation in ypt1-1 mutant cells was exacerbated when GFP-Snc1-PEM was overexpressed. In addition, we have previously shown that in YPT1 (WT) cells GFP-Snc1-PEM is transported to the vacuole for degradation [17]. We wished to determine whether ER-resident proteins are shuttled to the vacuole for degradation with over-expressed GFP-Snc1-PEM.
Hmg1 and Sec61 behaved like Snc1-PEM in wild type cells that overexpress GFP-Snc1-PEM. Immuno-blot analysis shows that their levels increased by 10 fold in cells defective in vacuolar proteolysis, indicating that they were delivered to the vacuole (Figs 6F and 6H and S6F). The degradation rate of Sec61 was compared in WT cells, which either over-express GFP-Snc1-PEM or not, using cycloheximide to inhibit protein translation. The half-life of Sec61 is shortened by 7 fold in cells over-expressing GFP-Snc1-PEM (from 29 to 4 hours; S6D Fig). The finding that the stability of the ER resident protein Sec61 is reduced in WT cells over-expressing GFP-Snc1-PEM further supports the idea that it is shuttled to the vacuole for degradation under these conditions. However, in spite of the fact that ~95% of Sec61 and Hmg1 were degraded in cells overexpressing GFP-Snc1-PEM, their steady-state level did not change (Figs 6E and S6E). In contrast to Sec61 and Hmg1, the levels of Sec12, Sec13 and Kar2 in vacuolar proteolysis defective cells overexpressing GFP-Snc1-PEM were very slightly changed (1.5–2 fold) (Figs 6G and 6H and S6G–S6H, left). Thus, in WT cells, >90% of some, but not all, ER resident proteins are shuttled for degradation in the vacuole with overexpressed GFP-Snc1-PEM.
In ypt1-1 mutant cells that overexpress GFP-Snc1-PEM, the levels of Hmg1 and Sec61, like that of Snc1-PEM, were increased by ~15 fold regardless whether vacuolar proteolysis was normal or defective, indicating that they accumulate before reaching the vacuole. In contrast, the levels Sec12, Sec13 and Kar2 increased by 4, 3, and 2 fold, respectively (Figs 6F–6H and S6F–S6H, right).
In the microscopy analysis, Hmg1 and Sec61 also behaved like GFP-Snc1-PEM. In YPT1 (WT) cells defective in vacuolar proteolysis, 60–70% of the intra-cellular GFP-Snc1-PEM co-localized with mCherry-tagged Hmg1 and Sec61 (Figs 7A and 7D and S7A). The proteins probably co-localize in the vacuole because co-localization is observed only in cells defective in vacuolar proteolysis and, using a vacuolar membrane stain, we have previously shown that GFP-Snc1-PEM accumulates inside the vacuole of pep4∆ cells [17]. This point was confirmed for Hmg1, which co-localizes with Snc1-PEM in the vacuole, using three-color fluorescence microscopy (Fig 7B). In contrast, very little co-localization was observed for Sec13, Sec12 and Kar2 (Figs 7A and 7D and S7A). In ypt1-1 mutant cells, >70% of GFP-Snc1-PEM co-localized with Hmg1 and Sec61, regardless of the vacuolar proteolysis state. For Sec12, Sec13 and Kar2, the levels were lower: 38, 14 and 12%, respectively (Figs 7C and 7D and S7B).
These results show that ~20–40% of some, but not all, ER-resident proteins are transported to the vacuole through macro-ER-phagy in WT cells. In cells over-expressing excess integral membrane proteins such as Snc1-PEM, >90% of these ER proteins are corralled to this pathway. Interestingly, even though the ER resident protein Sec61 is shuttled to the vacuole for the degradation when GFP-Snc1-PEM is overexpressed, its steady state level does not change.
Results presented here define a novel ER quality-control pathway, macro-ER-phagy. This pathway delivers excess of integral-membrane proteins from the ER to the lysosome for degradation and requires core autophagy machinery and at least two Ypt GTPases, Ypt1 and Vps21 (Ypt51). It was defined as macro-ER-phagy based on the requirement of core Atgs and the accumulation of the cargo in Atg8-marked autophagic structures (phagophore and APs) in vps21∆ mutant cells. Macro-ER-phagy is different from nonselective autophagy because it does not require starvation or prolonged ER stress, and from other specific autophagy processes, because it does not require known specific Atgs (like Atg19, Atg32, and Atg36). It is also different from micro-ER-phagy, which does not require Atgs [15] or Ypt1 GTPase (shown here). Finally, macro-ER-phagy is different from the other ERQC pathway ERAD in which proteins are extracted from the ER lumen through the membrane and delivered to the proteasome for degradation (Fig 8A). When macro-ER-phagy is impaired, UPR is induced, indicating that accumulation of its cargo causes ER stress. However, UPR induction is not required for macro-ER-phagy.
The cargo shuttled through the macro-ER-phagy pathways is different from cargo delivered by other known ERQC pathways. Whereas the cargo of the ERAD and micro-ER-phagy pathway are misfolded proteins (recognized in the ER lumen or in the ER membrane) and extra membrane, respectively [3,15], the cargo of the macro-ER-phagy is excess of integral-membrane proteins. Based on the fact that one such cargo, GFP-Snc1-PEM, does not have a lumenal domain, we propose that the macro-ER-phagy cargo is recognized in the cytoplasm.
What fraction of integral-membrane proteins is transported through basal macro-ER-phagy? Without overexpression of a membrane protein, about 20–50% of the ER resident proteins Sec61 and Hmg1 are transported to the vacuole for degradation. This estimate is based on a 1.2–1.6 fold increase in cells defective in vacuolar proteolysis, and on the 2–2.5 fold accumulation in ypt1-1 mutant cells regardless of whether vacuolar proteolysis is normal or not. This estimate is within the 30–75% range previously reported for ER proteins that do not pass the ERQC [1].
Importantly, cells that overexpress the integral-membrane protein GFP-Snc1-PEM, shuttle ~95% of this protein to the vacuole for degradation via macro-ER-phagy. This fraction is estimated based on the following two results: ~20-fold increase in GFP-Snc1-PEM level in cells defective in macro-ER-phagy (e.g., ypt1-1), and >10-fold increase in cells defective in vacuolar proteolysis (pep4∆ prb1∆). Moreover, >90% of some ER-resident membrane proteins (e.g., Sec61 and Hmg1) are also transported through macro-ER-phagy for degradation in cells overexpressing GFP-Snc1-PEM. In spite of this increase in the degradation of ER resident proteins upon over-expression of a single integral membrane protein, the steady state level of Sec61 and Hmg1 remained unchanged.
We propose that macro-ER-phagy is a “housekeeping” pathway that delivers excess resident-ER membrane proteins to the lysosome for degradation. When an integral membrane protein is overexpressed, ~95% of this protein is shuttled through this pathway together with >90% of the ER that contain certain resident-ER membrane proteins. Therefore, macro-ER-phagy of some resident ER proteins is induced from 20–50% to >90% by over-expression of a single membrane protein. The observation that the steady state level did not plunge under these conditions suggests that cells regulate the level of these ER resident proteins.
Based on data presented here, we delineate three sequential steps of the macro-ER-phagy pathway: In the first Atg9-dependent step, membranes containing macro-ER-phagy cargo form ER-to-autophagy membranes (ERAM). If this step is blocked, cargo accumulates in the ER, but UPR is not induced. In the second step, which is dependent on Ypt1, ERAM and the core Atgs form the PAS. If this step is blocked, ERAM accumulate and UPR is induced, even though UPR induction is not required for macro-ER-phagy. In the third Atg8- and Vps21-mediated step, APs are formed and fuse with the lysosome (see model in Fig 8B).
Atg9, the only integral-membrane core Atg, was implicated in delivering membrane to PAS [7,8]. However, until now its function could not be separated from that of the other PAS organizers. Because the macro-ER-phagy phenotype of atg9∆ is different from, and can mask, the phenotypes of ypt1-1, atg1∆ and atg11∆, the two steps could be separated and their order could be defined. This is the first time that the function of Atg9 is uncoupled from that of the other core Atgs in any autophagy process. In macro-ER-phagy, Atg9 is required for ERAM formation and is present on these membranes. Interestingly, in the absence of over-expressed cargo, a large portion of peripheral Atg9 structures co-localize with ER [43].
Ypt1 regulates ER-to-Golgi transport and all types of macro-autophagy in the context of different GEF-GTPase-effector modules [23]. Even though we established the ypt1-1 mutation affects autophagy and not ER-to-Golgi transport [21,22], it was important to ensure that mutations in other ER-to-Golgi and Golgi-to-vacuole regulators do not exhibit macro-ER-phagy phenotypes. Sec12 mediates exit from the ER towards the Golgi. In sec12-ts mutant cells, some membrane proteins are trapped in the ER, and ER stress is high. However, these mutant cells are not defective in macro-ER-phagy because excess membrane proteins can reach the vacuole, and atg9∆ enhances their phenotype rather than masking it. In addition, inhibition of transport from the Golgi or endosomes in vps4∆ mutant cells did not affect macro-ER-phagy. Moreover, even though the macro-ER-phagy phenotype of ypt1-1 is more severe than deletion of any single Atg, the observation that the phenotype of the atg11∆ atg17 double mutant is similar to that of ypt1-1 suggests that the ypt1-1 phenotype is caused by a defect in autophagy. All these results support the idea that the function of Ypt1 in macro-ER-phagy is unrelated to its function in ER-to-Golgi transport.
Vps21 was used here to show that macro-ER-phagy cargo utilizes APs like other macro-autophagy pathways. We have recently shown that Vps21 is required for specific and nonspecific autophagy, and vps21∆ mutant cells accumulate autophagic structures marked by Atg8 [24]. Here we show that Vps21 also plays a role in macro-ER-phagy because a macro-ER-phagy cargo accumulates in Atg8-marked autophagic structures of vps21∆ mutant cells. The observation that this phenotype can be masked by ypt1-1 supports the idea that Ypt1 functions before Vps21 in this pathway.
While UPR is induced when the second step of macro-ER-phagy is blocked, it is not required for this process. This idea is based on the observation that deletion of IRE1 or HAC1, which completely abolish the UPR, does not affect macro-ER-phagy in YPT1 (WT) cells, and can uncouple UPR from the formation of ERAM in ypt1-1 mutant cells. While it is not clear why UPR is induced in mutant cells defective in the second step of macro-ER-phagy (e.g., ypt1-1, atg1∆, atg8∆ and atg11∆), the finding that Kar2 (BiP) is excluded from ERAM in ypt1-1 mutant cells supports the model that sequestration of Kar2 from Ire1 (which presumably is retained) is one of the requirements for UPR induction [44].
While overexpression of integral PM proteins was used here to detect clear defects in the pathway, partial defects in delivery of some ER-resident proteins to the vacuole were observed without it in ypt1-1 mutant cells [17]. Moreover, we show here that some resident ER proteins are delivered to the vacuole through basal macro-ER-phagy also in wild type cells, and overexpression of a single integral PM protein further induces this pathway. This suggests that macro-ER-phagy is a constitutive ERQC process that clears excess membrane proteins from the ER. Our results also imply the existence of a complementary process that ensures that the steady state level of certain resident ER proteins is maintained when macro-ER-phagy is induced by over-expression of membrane proteins. While it is not clear whether this steady state level is regulated at the level of mRNA or protein, it seems that it does not involve UPR, which regulates the expression of multiple genes [45], because UPR is not induced in wild type cells overexpressing integral-membrane proteins (Figs 5C and S3C). Specific mechanisms that underlie basal and induced macro-ER-phagy and the regulation of ER-resident proteins level when macro-ER-phagy is induced, need to be deciphered.
Other open questions to be addressed in the future are how macro-ER-phagy cargo is recognized and sorted to ERAM and the relationship between this pathway and ERAD. We propose that macro-ER-phagy cargo is recognized either on the cytoplasmic side of the ER or in the ER membrane, because at least one cargo, GFP-Snc1-PEM, does not have a lumenal domain. Under ER stress, the ERAD-M and ERAD-C pathways recognize misfolded intra-membrane and cytosolic domains of membrane proteins, respectively [46]. While it is possible that excess integral-membrane proteins, especially when tagged with a fluorescence moiety, are partially misfolded, it seems that ERAD does not play a major role in their clearance because deletion of Ire1 and Hac1, which are required for ERAD induction, does not affect this clearance. To determine whether ERAD plays a role in the degradation of excess ER membrane proteins, the effect of ERAD mutations, alone and in combination with autophagy mutations, on this process should be determined in future experiments. It is intriguing if and how proteins are sorted between macro-ER-phagy and ERAD for degradation, whether they are sequestered to different domains, and whether yet unknown ER-phagy-specific Atgs play a role in this process. In addition, cooperation between ER-phagy and other ERQC processes has been proposed [4]. The idea is that each pathway can serve as a backup for the other. The synergistic effect of overexpression of multiple membrane proteins in atg11∆ mutant cells supports this idea. We propose that these mutant cells cope with excess of a single protein using another ERQC pathway, possibly ERAD, while overexpression of two proteins saturates this outlet.
Another question is whether ER-exit sites towards the Golgi (ERES) and ER-phagy overlap. Recently, ERES were proposed to be the site of AP biogenesis [47]. Our result that Sec12, an ERES integral-membrane protein, is not delivered to the vacuole through macro-ER-phagy, does not support a role for ERES in this process.
What is the relevance of degradation of excess membrane proteins by macro-ER-phagy? First, specific impairment of macro-ER-phagy should be considered in studies that require overexpression of membrane proteins. For example, ERQC was identified as the bottleneck of overexpression of heterologous proteins in yeast. Co-overexpression of chaperons, like BiP, did not affect this block [48], supporting the idea that macro-ER-phagy, and not ERAD, is the likely cause of this bottleneck. Most importantly, overexpression of membrane proteins has been associated with multiple human diseases. For example, amplification of HER2, a human epidermal growth factor receptor, occurs in ~20% of breast cancers and is associated with a more aggressive disease [49,50]. Likewise, overexpression of the M oncostatin receptor is associated with increased aggressiveness of cervical cancer and is considered a therapeutic target [51]. Moreover, overexpression of the P-glycogen efflux pump is associated with bone inflammation and is considered a potential therapeutic target [52]. Finally, chronic ER stress is associated with neurodegenerative diseases and therapeutic benefits of chemical autophagy activators were reported [53]. Therefore, if the mechanism we unraveled here is conserved from yeast to human like all other basic cellular processes, macro-ER-phagy of excess membrane protein would be relevant to human disease.
While this paper was under revision, Mochida et al [54] reported the identification of two new receptors, Atg39 and Atg40, required for autophagy of the ER during nitrogen starvation, which induces non-selective autophagy. Interestingly, Atg40 is similar to the FM134B reticulon, an autophagy receptor of ER in mammals, which was implicated in neuropathy in humans [55]. Future studies should clarify whether these new receptors play a role in macro-ER-phagy of ER proteins under normal growth conditions described here.
Details about strains, plasmids and reagents used in this manuscript are given in Supplementary Experimental Procedures (S1 File), and S1 and S2 Tables. Construction of strains and plasmids is described in Supplementary Experimental Procedures (S1 File).
All yeast strains were grown in rich (YPD) or minimal (SD) media containing the necessary amino acid supplements. In tunicamycin experiments, cells were incubated with 5 μg/ml tunicamycin for 1.5 hours before collecting them. Experiments involving DTT treatment were performed as previously described [15].
The protein level of GFP-Snc1-PEM, HA-tagged Sec61, Sec13 and Sec12 was determined as previously described [17]. The protein level of mCherry-tagged Hmg1 and Kar2 was determined according to the previously described methods [56,57]. Quantification was done using ImageJ and adjusted to the loading control.
For live-cell microscopy, cells expressing fluorescently tagged proteins were grown to mid-log phase in appropriate media. Fluorescence microscopy was performed using deconvolution Axioscope microscope (Carl Zeiss, Thornwood, NY) with FITC (GFP, yEGFP) and TexasRed (mCherry) sets of filters. Labeling of vacuole membranes with FM4-64 was done using a 5 min pulse followed by 60 min chase as previously described [58]. Labeling the vacuolar lumen with CMAC-Arg was done according to the manufacturer’s instructions. Immunofluorescence microscopy of yeast cells was performed as previously described [59].
The β-gal assay was done with cells as previously described [17]. Briefly, cells were transformed with a plasmid (pJC104) expressing the LacZ gene behind four UPR elements [60]. The level of β-galactosidase in cell extracts was determined as β-gal (Miller) units [61]. The values are 0.6–1.0 units for WT cells. Graphs represent percent of UPR in experimental cells (e.g., mutant cells) as fold of β-Gal units in WT cells; averages and error bars (represent STDEV) were calculated from two independent reactions; each result represents at least two different experiments from independent transformants. Statistical significance (p value) was calculated from two different experiments using independent transformants, each with two independent reactions (total of four reactions). In our experiments, tunicamycin treatment (5 μg/ml for 90 min) results in a 12-22-fold induction over untreated WT cells.
|
10.1371/journal.pgen.1007220 | Potential and limits for rapid genetic adaptation to warming in a Great Barrier Reef coral | Can genetic adaptation in reef-building corals keep pace with the current rate of sea surface warming? Here we combine population genomics, biophysical modeling, and evolutionary simulations to predict future adaptation of the common coral Acropora millepora on the Great Barrier Reef (GBR). Genomics-derived migration rates were high (0.1–1% of immigrants per generation across half the latitudinal range of the GBR) and closely matched the biophysical model of larval dispersal. Both genetic and biophysical models indicated the prevalence of southward migration along the GBR that would facilitate the spread of heat-tolerant alleles to higher latitudes as the climate warms. We developed an individual-based metapopulation model of polygenic adaptation and parameterized it with population sizes and migration rates derived from the genomic analysis. We find that high migration rates do not disrupt local thermal adaptation, and that the resulting standing genetic variation should be sufficient to fuel rapid region-wide adaptation of A. millepora populations to gradual warming over the next 20–50 coral generations (100–250 years). Further adaptation based on novel mutations might also be possible, but this depends on the currently unknown genetic parameters underlying coral thermal tolerance and the rate of warming realized. Despite this capacity for adaptation, our model predicts that coral populations would become increasingly sensitive to random thermal fluctuations such as ENSO cycles or heat waves, which corresponds well with the recent increase in frequency of catastrophic coral bleaching events.
| Coral reefs worldwide are suffering high mortality from severe thermal stress episodes induced by acute ocean warming events. Under the current rate of warming, will corals be gone before the end of this century? Here we combine population genomics with biophysical and evolutionary modeling to investigate adaptive potential of a common reef-building coral from the Great Barrier Reef. To approach this task, we have developed a predictive model of polygenic adaptation in a system of multiple inter-connected populations that exist in a heterogeneous and changing environment. Applying this model to our coral species, we find that populations successfully adapt to diverse local temperatures along the range of the Great Barrier Reef despite high migrant exchange and should collectively harbor enough adaptive genetic variants to fuel region-wide thermal adaptation for another century and perhaps longer. In the same time, the model predicts that random thermal fluctuations will induce increasingly severe coral mortality episodes, which aligns well with observations over the last few decades.
| Mass coral bleaching, caused by global warming, is devastating coral reefs around the world [1] but there is room for hope if corals can adapt to increasing temperatures from generation to generation [2]. Many coral species have wide distributions that span environments that differ dramatically in their thermal regimes, demonstrating that efficient thermal adaptation has occurred in the past [3]. But can coral adaptation keep up with the unprecedentedly rapid current rate of global warming [4]? One way for corals to achieve rapid thermal adaptation is through genetic rescue, involving the spread of existing heat tolerance alleles from warm-adapted populations to now-warming regions via larval migration [5,6]. We have previously demonstrated the presence of genetic variants conferring high thermal tolerance in a naturally warm low-latitude population of A. millepora on the Great Barrier Reef (GBR, [5]). It can be assumed that the effectiveness of GBR-wide adaptation based on these pre-existing variants would depend on the prevailing migrant exchange pathways and on total amount of genetic variation in populations of this coral. While considerable genetic connectivity along the latitudinal range of the GBR has been documented in corals [7–9], previous approaches have not been able to resolve the directionality of the migrant exchange to confirm that redistribution of heat-tolerance alleles from warm- to cooler-adapted populations is indeed taking place. In addition, recent declines in coral cover [10] could have already taken a toll on the total amount of standing genetic variation. Here, we used population genomics coupled with model-based allele frequency spectrum (AFS) analysis to establish directionality of migrant exchange and estimate contemporary and historical effective population sizes (measure of genetic diversity) in Acropora millepora, a common but heat sensitive reef-building coral representing the most ecologically prominent and diverse coral genus in the Indo-Pacific. The genomics-based migration rates were cross-validated with biophysical model of larval dispersal. The resulting demographic estimates were used to parameterize a newly developed metapopulation adaptation model to predict future persistence of A. millepora on the GBR.
We used samples collected in 2002–2009 from five populations of A. millepora along the latitudinal range of the GBR (Fig 1A). Environmental parameters (obtained from http://eatlas.org.au/) varied widely among these locations (Fig 1C). Importantly, maximum summer temperature (the major cause of bleaching-related mortality) followed the latitudinal gradient with one notable exception: one of the near-shore populations from the central GBR (Magnetic Island) experienced summers as hot as the lowest-latitude population examined here (Wilkie Island, Fig 1A and 1B).
We genotyped 18–28 individuals per population using 2bRAD [11] at >98% accuracy and with a >95% genotyping rate. Analysis of population structure based on ~11,500 biallelic SNPs agreed with previous results [8,12] and revealed very low levels of genetic divergence, with only the Keppel Islands population being potentially different from the others (Fig 1C and 1D). Pairwise FST did not exceed 0.014 even between the southernmost and northernmost populations (Keppel and Wilkie). We did not verify whether these FST measures were significant because our main statistical analysis was based AFS modeling.
Here we applied Diffusion Approximation for Demographic Inference (∂a∂i, [13]), a methodology of demographic analysis based on allele frequency spectrum (AFS). AFS is essentially a histogram of number of genetic variants binned by frequency (S1 Fig). AFS-based analysis is enabled by the multitude of molecular markers provided by next-generation sequencing and offers a number of advantages compared to classical population genetics approaches applied previously to GBR corals [7–9,14], Most importantly, AFS analysis does not rely on assumptions of genetic equilibrium (stability of population sizes and migration rates for thousands of generations) or equality of population sizes. It can fit any user-defined demographic model, for example involving asymmetric migration or population growth and declines, to maximize the likelihood of generating the observed AFS. This approach also allows likelihood-based model comparisons (likelihood ratio tests or tests based on Akaike Information Criterion, AIC) to prove the importance of parameters included in the model.
We used bootstrap-AIC approach to confirm that our populations are separate demographic units. For each pair of populations we generated 120 bootstrapped datasets by resampling genomic contigs and performed delta-AIC comparison of two demographic models, a split-with-migration model and a no-split model (S2 Fig). The split-with-migration model assumed two populations that split some time T in the past with potentially different sizes N1 and N2, and exchange migrants at different rates (m12 and m21) depending on direction. The no-split model allowed for ancestral population size to change but not for a population split, so the experimental data were modeled as two random samples from the same population of size N. The majority of bootstrap replicates (64–100%) showed AIC advantage of the split-with-migration model for all pairs of populations except Sudbury-Magnetic (41% support, S2 Fig). This indicates that most A. millepora populations on the GBR are in fact demographically distinct, despite often non-significant FST reported by previous studies based on allozymes [7,15] and microsatellite markers [8], or our own PCA and ADMIXTURE results (Fig 1C and 1D). This underscores the higher sensitivity of AIC-base AFS analysis compared to classical equilibrium population genetics methods.
AFS-based analysis allows statistically rigorous estimation of unidirectional migration rates between populations. The classical FST−based approach only allows estimating bi-directional migration rate [15] and even this calculation has been criticized because its underlying assumptions are rarely realistic [16]. We determined unidirectional migration rates from the split-with-migration model and estimated their confidence limits from bootstrap replicates. In theory, migration rate can be confounded with population divergence time, since in the AFS higher migration often looks similar to more recent divergence [17]. To confirm that the model with ancient population divergence and migration is preferable to the model with very recent divergence and no migration, we performed the delta-AIC bootstrap comparison between these models and obtained overwhelming support for the model with ancient divergence and migration (S3 Fig). Notably, for all pairwise analyses migration in southward direction exceeded northward migration, and this difference was significant in seven out of ten pairwise comparisons (Fig 2A and S2A Fig). Linear mixed model analysis of direction dependent median migration rates with a random effect of destination (to account for variation in total immigration rate) confirmed the overall significance of this southward trend (PMCMC <1e-4). Full listing of parameter estimates and their bootstrap-derived 95% confidence limits is given in S1 Table.
To investigate whether the southward migration bias was due to higher survival of warm-adapted migrants (due to ongoing sea surface temperature increase) rather than currents, we developed a biophysical model of coral larval dispersal on the Great Barrier Reef. This model quantified the per-generation migration potential among coral reef habitat patches in the GBR based on ocean currents and parameters of larval biology [18,19], the latter including pre-competency period, competency period and mortality rate [20]. The biophysical model predicted very similar migration rates as our genetic model (Mantel r = 0.58, p = 0.05), recapitulating the southward bias (Fig 2A–2C). Importantly, the same southward bias was predicted for population pairs in which southward migration corresponded to movement to the same-temperature or even to warmer location, such as migrations to the Magnetic Island. This indicates that southward migration bias is predominantly driven by ocean currents and not by preferential survival of warm-adapted coral genotypes migrating to cooler locations. More generally, this result indicates that currents remain the major factor affecting larval dispersal in our study species, with environmental selection pressures playing comparatively minor, if any, role.
The GBR has warmed considerably since the end of last century [21], which may have already reduced genetic diversity in A. millepora populations. We used ∂a∂i to infer effective population sizes, which is a measure of genetic diversity and one of the key parameters determining the population’s adaptive potential [22]. The results of the split-with-migration model (Fig 2A) were consistent for all population pairs and indicated that Keppel population was about one-fifth the size of others (Fig 2D and 2E). This result was not surprising since the Keppel population has frequently suffered high mortality due to environmental disturbances compared to the other populations studied [12]. To investigate whether there was a detectable recent drop in genetic diversity associated with GBR-wide coral decline [10] we have used stairwayPlot, a model-free method of past effective population size reconstruction based on AFS [23]. Reconstruction of the most recent changes draws information from the rarest alleles and therefore requires large sample sizes, which is why we pooled three highly similar populations from the Central GBR (Sudbury, Orpheus, and Magnetic) to increase the sample size to 84 individuals. We also analyzed Keppel population separately since it was distinct from the rest (Fig 1D and 1E) and was less genetically diverse (Fig 2D). This analysis confirmed the long-term decline of the Keppel population, possibly since the time of its separation from the main GBR stock, but did not find significant recent decline in the central GBR (Fig 2E). To see if a very recent decline was at all detectable with a sample size of 84, we used SLiM [24] to simulate evolution of 20,000 2bRAD loci in ten populations exponentially declining from 30,000 individuals to 1/10th or 1/3rd of that size over the last 20 generations (in our coral this would correspond to approximately 100 years). As a control, we simulated populations maintaining their size. With the sample size of 84 we could detect the ten-fold population crash in nine cases out of ten (albeit with low confidence) but we could not detect the three-fold decline (S4 Fig). Thus, although we did not detect recent drop in genetic diversity in our species, substantial population decline might still be occurring.
To evaluate whether local thermal adaptation could facilitate rapid adaptation of the whole A. millepora metapopulation to the simulated gradual warming, we developed an individual-based polygenic model of metapopulation adaptation in the SLiM software environment [24]. The model’s code is highly flexible and can simulate any number of populations with any configuration of population sizes, migration rates, and environmental trends. The number and effect sizes of QTLs, mutation rate, heritability, and breadth of tolerance can also be varied in this model. Here, we used population sizes and migration rates inferred from the genetic analysis (Fig 2A and 2D) and incorporated differences in mid-summer monthly mean temperature among populations (Fig 1A). Initially, populations were allowed to adapt to local thermal conditions while exchanged migrants and were at the state of genetic equilibrium at the start of the warming periods (S5 Fig). Warming consists in the temperature increase at a rate of 0.05°C per generation in all populations, corresponding to the projected 0.1°C warming per decade [25]. Both during pre-adaptation and warming periods the temperature was allowed to fluctuate randomly between generations to approximate El Nino Southern Oscillation (ENSO) or similar but random acute temperature events. During the warming period, a population declining in fitness would shrink in size and stop contributing migrants to other populations.
The fist important result of the model was that high migration rates (on the order of 0.1–1% of immigrants every generation, Fig 2C and S1 Table) did not lead to “migrational meltdown” [26] of local adaptation: all populations successfully adapted to local thermal conditions, although at the settings for lower efficiency of selection (low heritability and/or broad tolerance) this adaptation became increasingly imperfect (Fig 3B and 3D and S7 Fig). Moreover, under all parameter settings the pre-adapted metapopulation as a whole was able to persist through the gradual warming for at least 20–50 generations (100–250 years) although the initially warm-adapted populations were going extinct relatively rapidly (Figs 3, S6 and S7). Migration substantially contributed to this persistence (Fig 3E and 3F), underscoring the importance of divergent local adaptation and genetic rescue [5,6] in promoting and redistributing the standing genetic variation.
A notable tendency observed with all parameter settings was that during warming the fitness (and hence the size) of adapting populations began to fluctuate following random thermal anomalies, and the amplitude of these fitness fluctuations increased as the warming progressed even though the amplitude of thermal anomalies did not change (Fig 3G). These fluctuations correspond to severe mortality events induced by thermal extremes that can occur as a result of ENSO and heat waves and affected warm-adapted populations most, which very much resembles the situation currently observed throughout the world [1].
Efficiency of selection depends on how strongly the phenotype is determined by genotype (heritability), and also on how steeply the fitness declines if the phenotype does not match the environment (breadth of thermal tolerance). In our model, heritability becomes lower with more random variation (determined by the Esd parameter, Figs 3, S6 and S7) added to the breeding value. Lower heritability notably diminished the efficiency of local adaptation–during pre-adaptation period, mean fitness of each population was lower (S6 Fig) and the population’s mean phenotype failed to achieve full match to the environment (S7 Fig), ostensibly due to less efficient selection against maladapted immigrants. Yet, lower heritability did not result in reduced persistence of the metapopulation (Figs 3, S5 and S6). This is good news for reef-building corals since heritability of thermal tolerance in them is expected to be low: much of natural variation in this trait is due to the type of algal symbionts (Symbiodinium spp. [27]). Photo-symbionts are not transmitted from parent to offspring in the majority of coral species [28], and although host genetics can have some effect on the choice of Symbiodinium in the next generation [29] environment has stronger effect on this association [27,30].
Broader thermal tolerance (determined by parameter σ, Figs 3, S6 and S7) also reduces the efficiency of selection but increases the population’s mean fitness, counteracting the fitness-diminishing effect of lower heritability (S6 Fig). During warming, it prevents extinction despite increasingly poorer match between the population’s mean phenotype and the environment, and thus facilitates longer persistence (Figs 3, S6 and S7). It is also notable that both low heritability and broader tolerance decrease the sensitivity of populations to random thermal fluctuations (Fig 3A and 3C and S6 Fig).
There are many uncertainties in our model associated with coral biology. Below we argue that, while more research is certainly needed to resolve them, our parameter settings were for the most part set to under-estimate adaptive potential.
Mutation rates are generally difficult to estimate [31] and therefore in our model we had to rely on order-of-magnitude guesses. Encouragingly, even under relatively low per-locus mutation rate of 1e-6 per gamete per generation [32] we have observed occasional “evolutionary rescue” events: brief periods of accelerated phenotypic evolution due spread of novel beneficial mutations [33]. These events substantially contributed to the metapopulation persistence (Fig 3B and S7 Fig). Furthermore, even under ten-fold lower mutation rate (1e-7) the initial adaptive response during the first ~100 generations was still observed (Fig 4A), although there was no subsequent adaptation. While ten-fold higher mutation rate (1e-5) allowed for indefinite adaptation (Fig 4C), such a high mutation rate is unlikely to be realistic [32,34].
Changes in two other parameters could result in considerably longer persistence: larger population sizes and more fine-grained genetic architecture (more QTLs with less effect). Both of these strongly facilitate adaptation beyond the initial “genetic rescue” period (Fig 5). Unlike the high mutation rate, these settings are relatively realistic. Coral population size is likely larger than assumed in our model, which used effective population sizes suggested by genetic analysis as census sizes. However, in highly fecund marine organisms effective population sizes tend to be much smaller than census sizes, sometimes by orders of magnitude [35]. It is also notable that stairwayPlot predicted substantially larger effective population sizes than the ∂a∂i-derived estimates used for the model (Fig 2D and 2E). As for genetic architecture, our assumption of only ten thermal QTLs was conservative; the actual number of thermal QTLs in acroporid corals is likely much larger [36]. However, there is currently no data on the distribution of effect sizes of these QTLs, which would be an important subject for future research to improve the model.
As for breadth of thermal tolerance, in simulations shown on Fig 3 σ = 0.5 and σ = 2 corresponded to 86% and 13% decline in fitness if the individual’s phenotype mismatched the environment by 1°C. The existing data on the breadth of coral thermal tolerance are somewhat conflicting. One study shows that acroporid corals can successfully acclimatize to environments differing in maximum temperatures by as much as 2°C [37]; however, another study found that coral grew 52–80% more slowly when transplanted among locations differing by 1.5°C average temperature, [38]. Although it is not possible to directly interpret these results in terms of breadth of thermal tolerance function, the former study likely supports the broader tolerance setting while the latter study suggests narrower tolerance. It must also be noted that both these studies involved in situ transplantations and hence the effect of temperature remains confounded with colony history and other local fitness-affecting environmental parameters. Also, adult corals likely have narrower tolerance than larvae and recruits, which are expected to exhibit non-reversible developmental plasticity associated with metamorphosis and establishment within a novel environment [39]. One particularly important event during this developmental transition is establishment of association with local algal symbionts. Since symbionts also adapt to local thermal conditions [30] this would elevate the fitness of the coral host despite possible mismatch between its own genetically determined thermal optimum and local temperature, which we can model as broadening of the thermal tolerance. Future experiments that expose multiple genetically distinct coral individuals to a range of temperatures under controlled laboratory settings are required to rigorously quantify variation in thermal tolerance curves in natural populations.
It could be argued that our samples are genetically out of date, not capturing the effects of recent disturbances such as mass bleaching, large cyclones and a Crown-of-Thorns outbreak that have happened on the GBR since the time of their collection (2002–2009). However, very recent demographic events (in this case, 2–3 generations ago) are undetectable at the level of neutral genetic variation (S4 Fig) unless the study’s sample size is comparable to the number of disturbance-surviving individuals (i.e., either when the disturbance was truly catastrophic or the sample size is very large). Thus, our samples can still be considered representative of major patterns of genetic diversity of our study species.
Finally, our model assumed that recovery from high mortality events would happen without impediment, through reseeding by survivors and migrant influx from other coral populations. However, severe mortality across large spatial scales or ecological feedbacks such as shifts to an alternative ecological stable state [40] might substantially decrease the rate of reseeding and recovery of affected reefs. In that case, the increase in severity of bleaching-related mortality might lead to much faster coral extinction than predicted by our model.
We found that genetic diversity of Acropora millepora was not yet strongly affected by climate change and that the migration patterns were well positioned to facilitate persistence of the GBR metapopulation for a century or more. Our results underscore the pivotal role of standing genetic variation and migrant exchange in the future metapopulation persistence, suggesting management interventions such as assisted gene flow [41] by moving adult reproductively active colonies or by outplanting lab-reared offspring produced by crossing corals from different populations. With the estimated natural migration rates on the order of 0.1–1% (10–100) migrants per generation, human-assisted genotype exchange could appreciably contribute to the genetic rescue without risking disruption of the natural local adaptation patterns [42]. What might get in the way of assisted gene flow is adaptation of transplanted corals to other environmental parameters at home, for example, light levels or concentration of inorganic nutrients. The extent to which such adaptation can limit survival of transplanted corals and naturally dispersing larvae (the effect called “isolation by environment” [43] or “phenotype-environment mismatch” [44]) requires further study. Importantly, despite good prospects for short-term adaptation, coral populations are predicted to become increasingly more sensitive to random thermal anomalies, especially in the originally warm-adapted populations. The 10–85% mortality in the Northern GBR as a result of 2016 bleaching event [45] could be a particularly sobering recent manifestation of this trend. Finally, to validate the model’s predictions and further fine-tune its parameters, long-term monitoring of genetic variation in natural coral populations must be initiated to track ongoing evolutionary changes.
This study relied predominantly on samples described by van Oppen et al [8] with addition of several samples from Orpheus and Keppel islands that were used in the reciprocal transplantation experiment described by Dixon et al [46]. These samples were collected under Great Barrier Reef Marine Park Authority permits number G99/441 and G09/29894.1.
The samples were genotyped using 2bRAD [11] modified for Illumina sequencing platform; the latest laboratory and bioinformatics protocols are available at https://github.com/z0on/2bRAD_denovo. BcgI restriction enzyme was used and the samples retained for this analysis had 2.3–20.2 (median: 7.45) million reads after trimming and quality filtering (no duplicate removal was yet implemented in this 2bRAD version). The reads were mapped to the genome of the outgroup species, Acropora digitifera [47,48], to polarize the allelic states into ancestral (as in A. digitifera) and derived, e.g., [49,50]. Genotypes were called using GATK pipeline [51].
Preliminary analysis of sample relatedness using vcftools [52] revealed that our samples included several clones: four repeats of the same genotype from the Keppel Island (van Oppen et al [8] samples K210, K212, K213 and K216), another duplicated genotype from Keppel (samples K211 and K219), and one duplicated genotype from Magnetic Island (samples M16 and M17). All other samples were unrelated. We took advantage of these clonal replicates to extract SNPs that were genotyped with 100% reproducibility across replicates and, in addition, appeared as heterozygotes in at least two replicate pairs (script replicatesMatch.pl with hetPairs = 2 option). These 7,904 SNPs were used as “true” SNP dataset to train the error model to recalibrate variant quality scores at the last stage of the GATK pipeline. During recalibration, we used the transition-transversion (Ts/Tv) ratio of 1.438 determined from the “true” SNPs to assess the number of false positives at each filtering threshold (as it is expected that an increase of false positive calls would decrease the Ts/Tv ratio towards unity). We chose the 95% tranche, with novel Ts/Tv = 1.451. After quality filtering that restricted the calls to only bi-allelic polymorphic sites, retained only loci genotyped in 95% or more of all individuals, and removed loci with the fraction of heterozygotes exceeding 0.6 (possible lumped paralogs), we ended up with 25,359 SNPs. In total, 2bRAD tags interrogated 0.18% of the genome. The genotyping accuracy was assessed based on the match between genotyped replicates using script repMatchStats.pl. Overall agreement between replicates was 98.7% or better with the heterozygote discovery rate (the fraction of matching heterozygote calls among replicates) exceeding 96%. All but one representative of each clonal group were excluded from all subsequent analysis.
To begin to characterize genome-wide divergence between populations we used pairwise genome-wide Weir and Cockerham’s FST calculated by vcftools [52], principal component analysis (PCA) using R package adegenet [53], and ADMIXTURE [54]. For PCA and ADMIXTURE, the data were thinned to keep SNPs separated by 5kb on average and by at least 2.5 kb, choosing SNPs with highest minor allele frequency (script thinner.pl with options ‘interval = 5000 criterion = maxAF’), resulting in 11,426 unlinked SNPs. The optimal K in ADMIXTURE analysis was determined based on the cross-validation procedure incorporated within ADMIXTURE software; the lowest standard error in cross-validation was observed at K = 1.
Prior to demographic analysis, Bayescan [55] was used to identify sites potentially under selection among populations, and 73 sites with q-value <0.5 were removed. This aggressive removal of potential non-neutral sites resulted in better agreement between bootstrap replicates compared to an earlier analysis where only 13 sites with q-value < 0.05 were removed. Demographic models were fitted to 120 bootstrapped datasets, which were generated in two stages. First, three alternatively thinned datasets were generated for which SNPs were randomly drawn to be on average 5 kb apart and not closer than 2.5 kb (10,042–10,074 SNPs in each). This time the SNPs were drawn at random to avoid distorting the allele frequency spectrum (unlike thinning for PCA and ADMIXTURE where the highest minor allele frequency SNPs were selected). Then, 40 bootstrapped replicates were generated for each thinned dataset by resampling contigs of the reference genome with replacement (script dadiBoot.pl). The fitted model parameters were summarized after excluding bootstrap replicates that fell into the lowest 15% likelihood quantile and the ones where model fitting failed to converge, leading to some parameters being undetermined or at infinity (less than 10% of total number of runs). Delta-AIC values were calculated for each bootstrap replicate that passed these criteria for both compared models, and summarized to obtain bootstrap support value, the percentage of replicates favoring the alternative model. While fitting ∂a∂i models, the data for each population were projected to sample sizes maximizing the number of segregating sites in the analysis, resulting in 6143–7193 segregating sites per population. Initially, our models included a parameter designed to account for ancestral state misidentification rate when constructing the polarized AFS (e.g., [56]), but since this parameter was consistently estimated to be on the order of 0.001 and had negligible effect on the models’ likelihood, we removed it from the final set of models.
To convert ∂a∂i -reported parameter values (θ, T and M) into time in years (t), effective population sizes in number of individuals (Ne) and migration rates (the fraction of the total population that are new immigrants in each generation, m), we estimated the mutation rate (μ) from the time-resolved phylogeny of Acropora genus based on the paxC intron [57], at 4e-9 per base per year. Although A. millepora can reproduce after 3 years [58] we assumed a generation time of 5 years reasoning that it would better reflect the attainment of full reproductive potential as the colony grows. Assuming a genome size of 5e+8 bases [47] the number of new mutations per genome per generation is 10. Since the fraction of the genome that is sequenced using 2bRAD was 1.8e-3 (calculated by dividing the total length of genotyped RAD loci by the size of the reference genome), the mutation rate per 2bRAD-sequenced genome fraction per generation is μ = 0.018. This value was used to obtain:
A spatially-explicit biophysical modeling framework [18,59] was used to quantify migration between coral reef habitats of the broader region surrounding the Great Barrier Reef, thereby revealing the location, strength, and structure of a species' potential population connectivity. The model’s spatial resolution of ca. 8 km coincides with hydrodynamic data for the broader region (1/12.5 degree; HYCOM+NCODA Reanalysis and Analysis product; hycom.org). Our biophysical dispersal model relies on geographic data describing the seascape environment and biological parameters capturing coral-specific life-histories. Coral reef habitat data are available from the UNEP World Conservation Monitoring Centre (UNEP-WCMC; http://data.unep-wcmc.org/datasets/1) representing a globally-consistent and up-to-date representation of coral reef habitat. To capture specific inter-annual variability, two decades of hydrodynamic data were used from 1992 to 2013 [60].
Coral-specific biological parameters for A. millepora included relative adult density (dependent on the habitat), reproductive output, larval spawning time and periodicity (e.g., the majority of colonies at Magnetic Island spawn a month earlier than the majority of colonies on other GBR sites [61]), maximum dispersal duration, pre-competency and competency periods, and larval mortality [20]. The spatially explicit dispersal simulations model the dispersal kernel (2-D surface) as a ‘cloud’ of larvae, allowing it to be concentrated and/or dispersed as defined by the biophysical parameters. An advection transport algorithm is used for moving larvae within the flow fields [62].
Simulations were carried out by releasing a cloud of larvae into the model seascape at all individual coral reef habitat patches and allowing the larvae to be transported by the currents. Ocean current velocities, turbulent diffusion, and larval behavior move the larvae through the seascape at each time-step. Larval competency, behavior, density, and mortality determine when and what proportion of larvae settle in habitat cells at each time step. When larvae encounter habitat, the concentration of larvae settling with the habitat is recorded at that time-step. From the dispersal data, we derived the coral migration matrix representing the proportion of settlers to each destination patch that came from a source patch, which is analogous to the source distribution matrix [63] and is equivalent to migration matrices derived from population genetic analysis. It is important to note that migration matrices extracted for the field sites represent the potential migration through all possible stepping-stones.
The model was implemented in SLiM [24], the forward evolutionary simulator, by modifying the provided recipe “Quantitative genetics and phenotypically-based fitness”. The model simulated Wright-Fisher populations with discreet generations. At the start of the simulation, populations were established with specified pairwise migration rates. Mutations (at the rate of 1e-6 per locus per gamete) had the effect size drawn from a normal distribution with mean zero and specified standard deviation. To rapidly generate and equilibrate the standing genetic variation, we used the fact that the allele frequency spectrum is the function of the product of population size and mutation rate. Since smaller populations equilibrate proportionally faster and are much faster to simulate, the first 5,000 generations were performed at 100-fold smaller population size than the final target value (12,500 for all populations except K, which was five times smaller) but 100-fold higher mutation rate, followed by 5,000 generations with 10-fold smaller population size and 10-fold higher mutation rate, followed by 10,000 generations with the target population size and mutation rate. This step-wise strategy resulted in rapid generation and equilibration of genetic diversity both within individual populations and in the whole metapopulation (S5 Fig). The phenotype of each individual was calculated as breeding value (sum of all QTL effects) plus Gaussian-distributed noise (of the magnitude set by the Esd parameter) to simulate a non-heritable phenotypic component. Then, fitness of each individual was calculated based on the difference between the individual’s phenotype (thermal optimum), temperature of the environment, and the setting for the breadth of thermal tolerance curve (σ parameter, the standard deviation of the Gaussian slope of fitness decline away from the phenotypic optimum). Each generation parents were chosen to produce the next generation according to their fitness; parents for immigrant individuals were chosen from among individuals in the source population. New mutations at QTLs happened at the specified rate at the transition to the next generation and the effect of a new mutation was added to the previous QTL effect. To model fitness-dependent population demography, we implemented linear scaling of the population size and migration rates with the population’s mean fitness. In the model described here this scaling was applied during 500 generation preceding warming and during the warming period, so that populations declining in fitness relative to their historical level shrunk in size and stopped contributing migrants to other populations. Genetic variation shown on S5 Fig was calculated as standard deviation of breeding values, representing the average difference between genetically-determined thermal tolerance of an individual and the population’s mean in °C (genetic variance is this value squared).
Adjustable model parameters and their settings in this study were:
The model’s code is designed for general modeling of polygenic adaptation in metapopulations. It can read user-supplied files of environmental conditions, population sizes and migration matrices for an arbitrary number of populations.
Here, we modeled our five populations with effective population sizes and pairwise migration rates inferred by ∂a∂i. We modeled identical thermal trends across populations with population-specific offsets. During the pre-adaptation period lasting 20,000 generations, the temperature was constant on average but experienced random fluctuations across generations drawn from a normal distribution with a standard deviation of 0.25°C (to approximate ENSO events). The temperature was offset by +1.6°C in Wilkie and Magnetic populations and by -1.8°C in the Keppel population, to model differences in midsummer monthly mean temperature among populations (Fig 1). After 20,000 generations, a linear increase at 0.05°C per generation was added to simulate warming.
All combinations of parameter settings were run ten times to ensure consistency. We found that with population sizes in thousands, such as in our case, the results were very consistent among independent runs. We therefore did not aggregate results over many replicated runs but instead show one randomly chosen run for each tested parameter combination.
|
10.1371/journal.ppat.1002119 | Clathrin Facilitates the Morphogenesis of Retrovirus Particles | The morphogenesis of retroviral particles is driven by Gag and GagPol proteins that provide the major structural component and enzymatic activities required for particle assembly and maturation. In addition, a number of cellular proteins are found in retrovirus particles; some of these are important for viral replication, but many lack a known functional role. One such protein is clathrin, which is assumed to be passively incorporated into virions due to its abundance at the plasma membrane. We found that clathrin is not only exceptionally abundant in highly purified HIV-1 particles but is recruited with high specificity. In particular, the HIV-1 Pol protein was absolutely required for clathrin incorporation and point mutations in reverse transcriptase or integrase domains of Pol could abolish incorporation. Clathrin was also specifically incorporated into other retrovirus particles, including members of the lentivirus (simian immunodeficiency virus, SIVmac), gammaretrovirus (murine leukemia virus, MLV) and betaretrovirus (Mason-Pfizer monkey virus, M-PMV) genera. However, unlike HIV-1, these other retroviruses recruited clathrin primarily using peptide motifs in their respective Gag proteins that mimicked motifs found in cellular clathrin adaptors. Perturbation of clathrin incorporation into these retroviruses, via mutagenesis of viral proteins, siRNA based clathrin depletion or adaptor protein (AP180) induced clathrin sequestration, had a range of effects on the accuracy of particle morphogenesis. These effects varied according to which retrovirus was examined, and included Gag and/or Pol protein destabilization, inhibition of particle assembly and reduction in virion infectivity. For each retrovirus examined, clathrin incorporation appeared to be important for optimal replication. These data indicate that a number of retroviruses employ clathrin to facilitate the accurate morphogenesis of infectious particles. We propose a model in which clathrin contributes to the spatial organization of Gag and Pol proteins, and thereby regulates proteolytic processing of virion components during particle assembly.
| The assembly and maturation of infectious retroviruses is driven by two viral proteins, Gag and Pol. Additionally, a number of cellular proteins are found in retrovirus particles, many of which lack a known functional role. One such protein is clathrin, which normally mediates several physiological processes in cells and was previously thought to be only passively incorporated into virions. In this study we show that clathrin is actively, specifically and abundantly incorporated into retrovirus particles. In several cases, retroviral proteins encode peptide motifs that mimic those found in cellular adaptor proteins that are responsible for clathrin recruitment. The range of retroviruses into which clathrin is packaged includes human and simian immunodeficiency viruses as well as other murine and simian retroviruses. Manipulations that prevented clathrin incorporation into virions also caused a variety of defects in the genesis of infectious retroviruses, including viral protein destabilization, inhibition of particle assembly and release, and reduction in virion infectiousness. The precise nature of the defect varied according to which particular retrovirus was examined. Overall these studies suggest that clathrin is frequently employed by retroviruses to facilitate the accurate assembly of infectious virions.
| To establish a productive infection in host cells, retroviruses have evolved strategies that employ numerous host factors to facilitate their replication. Recently, several groups have applied genome-wide RNAi screens to identify hundreds of candidate host factors that may facilitate human immunodeficiency virus-1 (HIV-1) and murine leukemia virus (MLV) infection [1], [2], [3]. Other strategies to identify host factors that facilitate virus replication include the identification of proteins that bind to viral proteins [4] and analysis of the proteomes that are incorporated into virions [5], [6], [7]. Indeed, host proteins involved in HIV-1 budding, such as Tsg101 [8] and ALIX [9] can be found in virions. Additionally, proteins that modulate virion infectivity such as cyclophilin A (CypA) [10], and Hsc70 [11] can also be demonstrated to be virion components. However, while proteomic analyses of purified HIV-1 or MLV particles have revealed dozens of virion-associated host proteins, no biological significance has been attached to the virion association of many of them. One such protein is clathrin, which previous reports suggest is only passively incorporated into particles [6].
Clathrin has been intensively studied in the context of cell biology (reviewed in [12]). It is a cytosolic protein that functions in vesicle genesis and transport and, specifically, mediates endocytosis from the plasma membrane and cargo trafficking from the trans-Golgi network (TGN). Clathrin is comprised of a trimer of 180 kDa heavy chains (HC) that are arranged with their N-terminal adaptor binding domains at the extremities of each leg of a triskelion, while clathrin light chains (LC) bind to heavy chains close to their C-termini. Clathrin adaptors (such as the AP family) govern the sorting of specific cargoes into clathrin-coated vesicles and recruitment of clathrin to membranes (reviewed in [13]). Many of these adaptors contain motifs such as LΦXΦ[DE] (Φ indicates a bulky hydrophobic residue), or in the case of AP180, repeated motifs with the sequence DLL, which bind to the clathrin N-terminal β–propeller domain, and facilitate the recruitment of clathrin to the plasma membrane [14], [15].
Here, we demonstrate that clathrin is abundantly and specifically incorporated into a range of diverse retrovirus particles, including HIV-1, SIVmac, MLV and M-PMV through interactions with Gag or Pol proteins. Indeed, several retroviral Gag proteins were found to encode peptide motifs that drive clathrin incorporation and mimic those found in cellular clathrin adaptor proteins. We also show that mutations in the motifs that mediate clathrin recruitment have a range of effects on the accuracy of particle morphogenesis, or on virion infectiousness, depending on the particular retrovirus that was examined. In several cases, we demonstrate that these effects can be recapitulated by reducing the available levels of clathrin in cells.
Initially, to discover potential host factors involved in HIV-1 replication, we set out to identify cellular proteins that are incorporated into HIV-1 particles, using a slightly different strategy compared to previous studies. To minimize the “noise” in such experiments, such as contaminating cell debris, or passively incorporated proteins, and prevent degradation of incorporated cellular proteins by the viral protease, plasmids expressing codon-optimized HIV-1 Gag alone, or protease-inactive (D25A, PR-) GagPol proteins were transfected into 293T cells. This generated a very high yield of virus-like particles (VLPs) without apparent cytotoxicity. Silver and Coomassie blue staining of SDS-PAGE gels (Figure 1A), loaded with iodixinal gradient-purified GagPol VLPs revealed that five cellular proteins were abundantly incorporated into GagPol VLPs and were resistant to digestion by externally applied subtilisin. Mass spectroscopic analysis of the excised bands identified these proteins as CypA, Hsp70, Hsp90, AIP1/ALIX and clathrin HC (Figure 1A), all of which have previously been identified as components of HIV-1 particles [5], [6], [9], [10], [11]. Strikingly, and in contrast to the other proteins that were found in HIV-1 GagPol VLPs, both silver staining and Western blot analysis revealed that clathrin HC was undetectable in particles simultaneously generated using the HIV-1 Gag protein alone (Figure 1A). Moreover, visual inspection of silver or Coomassie stained gels suggested that clathrin HC was packaged into GagPol VLPs at exceptionally high level, i.e. the level of clathrin HC in virions approached that of GagPol (Figure 1A, right panel). To verify this finding with authentic virion particles, the T-cell lines CEMX174 and MT2 were infected with VSV-G pseudotyped, Env-defective HIV-1 at a multiplicity of infection of ∼1, washed extensively and progeny virions were harvested 40 h later (Figure 1B). Western blot analysis revealed that clathrin HC was abundantly incorporated into these virions. Indeed, in experiments where virions were simultaneously generated via infection of T-cell lines or transfection of 293T cells and the amount of clathrin incorporated into virions directly compared by quantitative fluorescence-based Western blotting (LI-COR), we found that the T-cell derived virions incorporate as much or more clathrin than the 293T-derived particles (Figure S1). Moreover, we determined that 6 to 8% of the total clathrin HC in the HIV-1 infected T-cell cultures appeared to be present in extracellular virions rather than cells (Figure S1).
Consistent with previous findings that clathrin LC binds to the C-terminal portion of clathrin HC and forms a stable complex with it [12], clathrin LC was also incorporated into HIV-1 virions. Indeed, when green fluorescent HIV-1 VLPs were generated by coexpression of HIV-1 GagGFP and GagPol in cells stably expressing fluorescently tagged clathrin LC (DsRed-clathrin LC), about 40% of the VLPs were labeled with sufficient DsRed-clathrin LC to be visualized by deconvolution microscopy (Figure 1C, Table S1). In contrast, only a few VLPs generated by coexpression of GagGFP and Gag were red-fluorescent. In similar experiments, authentic virions were generated by transfection with a proviral plasmid encoding YFP embedded in the stalk region of the MA domain of Gag into cells stably expressing DsRed-clathrin LC. In this case, 70–80% of YFP+ virions contained sufficient DsRed-clathrin LC to be visualized as colocalizing red fluorescent puncta (Figure 1D, Table S1).
Given that clathrin incorporation into HIV-1 particles appeared Pol-dependent, we next attempted to map the sequences responsible for its incorporation, using particles generated by expressing protease-defective GagPol proteins and detection of virion associated clathrin using Western blot or microscopy assays. Precise deletion of the reverse transcriptase (RT) or integrase (IN) domain in this context did not markedly affect VLP release but, surprisingly, both manipulations abolished clathrin incorporation (Figure 1C and Table S1). Similarly, several Pol truncations or point mutations, constructed in the context of protease-defective proviral plasmid, also inhibited or abolished clathrin incorporation (Figure 1E and Table S1). Specifically, mutations in reverse transcriptase (RT), including L234A, that inhibit RT dimerization [16] prevented clathrin incorporation as did deletion of the IN C-terminal domain (CTD) as well as two so-called ‘class II’ IN mutations, N184L and F185K [17]. Western blot analysis of virions indicated that, while the levels of GagPol protein in cells and virions were not apparently affected by these mutations, clathrin incorporation was inhibited (Figure 1E). We even found that the presence of the non-nucleoside RT inhibitor efavirenz, which stimulates RT dimerization [16], during the production of VLPs, inhibited clathrin incorporation into virions (Table S1). In contrast, no effect on clathrin incorporation was observed for the active site IN point mutant E152K (Figure S2). Taken together, these findings demonstrate that multiple Pol domains are strictly required for clathrin incorporation into HIV-1 VLPs. This suggests that the overall conformation of Pol is critical for clathrin packaging, and effectively made it impractical to map small motifs in Pol that might be responsible for clathrin recruitment.
Clathrin adaptors GGA and AP-1, 2 and 3 are known to bind to the N-terminal 7-bladed β–propeller domain of clathrin HC [13]. This clathrin domain, encoded by residues 1–494aa is known to fold autonomously when expressed in the absence of other clathrin domains [18]. Co-expression of N-terminal 1–494aa or 1–363aa fragments of clathrin along with HIV-1 GagPol resulted in specific incorporation of these clathin fragments into wild-type VLPs, but not into IN δCTD VLPs (Figure 1F), mimicking the property of endogenous full-length clathrin. Thus, the N-terminal adaptor binding domain of clathrin HC was sufficient to drive its incorporation into HIV-1 GagPol VLPs.
To determine whether clathrin incorporation into virions is a general feature of primate lentiviruses, we next determined whether it also occurred in SIVmac. In contrast to results obtained with HIV-1 Gag, analysis of VLPs generated using only SIVmac Gag showed that clathrin was abundantly incorporated into VLPs in the absence of Pol (Figure 2A). Mapping experiments in which chimeric SIVmac/HIV-1 Gag proteins (Figure 2B) were used to generate VLPs revealed that Gag proteins encoding the SIVmac p6 domain, specifically SIV(HIV MA) and HIV(SIV p6), yielded VLPs containing clathrin (Figure 2B and 2C lane 4 and lane 6). Conversely, the reciprocal chimeric Gag proteins, HIV(SIV MA) and SIV(HIV p6), generated VLPs that did not incorporate clathrin (Figure 2B and 2C lane 3 and lane 5). Thus, the differential abilities of HIV-1 and SIVmac Gag VLPs to incorporate clathrin were clearly governed by the p6 domain.
Inspection of the SIVmac p6 protein sequence revealed two ‘DLL’ motifs (positioned at p6 residues 21–23 and 51–53) that are absent in HIV-1 p6 (Figure 2D). Because the clathrin adaptor AP180 employs multiple copies of a DLL motif in its C-terminal domain to directly bind to clathrin HC [15], we mutated either or both DLL motifs in SIVmac p6 and tested the ability of the mutant Gag proteins to drive clathrin incorporation into VLPs. Western blot analysis indicated that mutations in both motifs (D21A, D51A) or the second motif only (D51A or L52S) reduced clathrin incorporation to almost undetectable levels, while mutation in the first DLL motif alone had little effect (Figure 2E).
To determine the clathrin domain that mediates packaging into SIVmac Gag VLPs, expression plasmids encoding the 1–494aa or 1–363aa clathrin HC N-terminal domains were coexpressed with SIVmac Gag. As shown in Figure 2F, SIVmac VLPs efficiently incorporated the clathrin HC N-terminal 1–494aa fragment, but unlike HIV-1, the 1–363aa fragment was poorly incorporated (Figure 2F, bottom panel). The significance of the difference in clathrin sequence requirements for incorporation into HIV-1 versus SIVmac virions is unclear at present, but in both cases the N-terminal adaptor binding domain appeared to be responsible for driving incorporation.
Next, we tested whether SIVmac Pol, like HIV-1 Pol, could also drive clathrin incorporation. To facilitate clathrin detection, an HA tagged clathrin N-terminal domain (1–494aa) was co-transfected with plasmids expressing either wild-type SIVmac GagPol, or mutants in which either (i) the DLL motifs in p6 were mutated (DLL- GagPol), (ii) IN was mutated (N184L) in a way analogous to that which blocks clathrin incorporation into HIV-1 GagPol VLPs, or (iii) both p6 and IN were mutated. SIVmac DLL- GagPol exhibited greatly diminished clathrin packaging into VLPs (Figure 2G). However, some clathrin incorporation was observed in SIVmac DLL- GagPol VLPs, and this incorporation was completely abolished by additional mutation at IN residue N184 (Figure 2G). Thus, in SIVmac, both Gag and Pol contribute to clathrin incorporation into virions, but Gag appears to play the dominant role and drives significantly more clathrin incorporation than Pol.
The finding that clathrin was specifically packaged into HIV-1 and SIVmac VLPs prompted us to ask whether clathrin incorporation is a general property of retroviruses. Inspection of a variety of retroviral Gag protein sequences revealed that some, but not all, encoded putative clathrin binding peptides, including DLL and LLTLD motifs in their Gag proteins. In particular, a prototype gammaretrovirus, MLV and a prototype betaretrovirus, M-PMV were selected for further investigation. Putative clathrin binding motifs, DLL and DLISLD respectively, were found in their Gag proteins proximal to their late domains (a DLL motif at 156–158aa in MLV Gag and a DLISLD motif at 129–133aa of M-PMV Gag respectively, Figure 3A). Each of these viruses was found to package either endogenously expressed clathrin HC, or coexpressed N-terminal adaptor-binding domain fragments of clathrin HC (1–494aa or 1–363aa, Figure 3B,C,D,E) into virions. This incorporation was specific because mutations in respective putative clathrin binding sites in Gag (DLL to ALL in MLV or DLISLD to DAASLD in M-PMV) dramatically reduced this incorporation. Notably, the yield of virion particles was unaffected by mutations in these clathrin recruiting motifs (Figure 3B,C,D,E). Mutations in other candidate clathrin HC binding sites in Gag (533LLTLD537 at the C-terminus of MLV Gag and 30DLL32 in the matrix domain of M-PMV, respectively) caused no reduction in clathrin incorporation, indicating that these other candidate motifs do not play a critical role in clathrin packaging (unpublished observations).
Overall, these findings demonstrated that clathrin can be specifically incorporated into virions from widely divergent retroviruses, and that this incorporation is driven by the adaptor-binding domain at the N-terminus of clathrin HC. Moreover, clathrin incorporation sometimes occurs via the action of peptide motifs in viral structural proteins that mimic those found in cellular clathrin adaptors.
To probe the role of clathrin in retrovirus replication, we adopted a variety of approaches, including the analysis of viral mutants that were defective for clathrin incorporation, as well as depletion of clathrin using siRNA based approaches. Attempts to deplete clathrin using siRNA were complicated by the fact that it is highly abundant, has a relatively long half-life (20 h–50 h) [19], [20] and is essential for various cellular functions and for cell viability. Therefore, while we were able to reproducibly deplete about 70–80% of endogenous clathrin HC (Figure S3A), it proved nearly impossible to completely deplete the intracellular clathrin to a sufficient extent such that a clathrin-deficient viral phenotype could be analyzed. Moreover, since clathrin plays a critical role in a number of cellular pathways, including trafficking of proteins through the secretory pathway, distinguishing the direct effects of depletion from indirect effects is challenging. Therefore, in order to investigate the role that clathrin plays in retrovirus life cycles, we employed multiple approaches including characterizing viruses with mutations that prevent clathrin incorporation, reducing clathrin expression by siRNA depletion (Figure S3A), and overexpressing the C-terminal domain of the clathrin adaptor AP180 to induce clathrin sequestration (Figure S3B) [21], [22].
Several mutations in HIV-1 Pol were found to block clathrin incorporation, including mutations in RT at the dimer interface (L234A), as well as class II and CTD-truncating mutations in IN (N184L and δCTD) (Figure 1E). Unfortunately, these mutations are pleiotropic, and may therefore have multiple effects on viral replication by perturbing the tertiary structure and function of the Pol protein. Although these mutations did not have discernable effects on Gag and GagPol precursor protein levels in cells or VLPs when the viral protease was inactivated by mutation (Figure 1E), these mutations induced aberrant proteolytic cleavage of Pol proteins and reduced levels of GagPol precursor in cells and virions when protease was active [23], [24] (Figure 4A, Figure S4). Specifically, when HIV-1 proviruses were expressed in 293T cells, the WT and Pol-mutant viruses generated similar levels of Gag protein and its processed derivatives, but the levels of GagPol precursor, partly processed intermediates and mature IN proteins, detected using an anti-IN antibody, were diminished in the mutants that failed to package clathrin into virions (Figure 4A, Figure S4). Similarly, in a series of constructs that were made to express GagPol protein with an HA-epitope fused at C-terminus of Pol, reduced GagPol levels were detected using an anti-HA antibody when mutations that blocked clathrin were introduced, and aberrant processed derivatives were detected (Figure 4B).
While the RT and IN mutations both blocked clathrin incorporation and induced PR-dependent Pol depletion, it was not clear whether or how these two effects were causally related to each other, given the pleiotropic nature of these mutations. However, these results did suggest the possibility that clathrin recruitment by Pol might inhibit its viral protease-dependent depletion. To investigate this possibility, we examined Pol processing in cells where the clathrin HC N-terminal domain (1–494aa) or the full-length clathrin HC was overexpressed. Notably, for two Pol mutants, namely RT(L234A) and IN(F185K), clathrin HC (1–494aa) or full length clathrin HC overexpresssion led to an increase in the levels of GagPol derived proteins, including mature IN (Figure 4B and 4C). Conversely clathrin overexpression had no, or only a slight effect, on IN (N184L or δCTD) mutants (Figure 4B and 4C). Importantly, the effects of clathrin HC overexpression were specific to Pol; there was no significant effect on Gag precursor or processed derivative levels. To further explore the idea that clathrin recruitment might stabilize Pol in the presence of an active viral protease, we fused the clathrin recruiting domain from SIVmac Gag, namely p6 (mutated in such a way so as not bind to Tsg101 and ALIX), to the C-terminus of HIV-1 GagPol (IN δCTD). This chimera expressed higher levels of Pol protein expression than an equivalent construct containing the SIVmac p6 domain in which the DLL motifs were mutated (Figure 4D). Together, these results are consistent with the notion that clathrin recruitment results in the stabilization of Pol proteins in the presence of an active viral protease. In support of this idea, Western blot analyses revealed that clathrin HC depletion using siRNA reduced the levels of GagPol precursor, IN, and partly processed intermediate Pol proteins in cells transfected with a WT HIV-1 proviral construct (Figure 4E). Importantly, the levels of Gag protein (which are translated from the same viral mRNA species) were not reduced by this manipulation. Furthermore, AP180C overexpression also reduced the levels of intracellular GagPol, IN and intermediate proteins without affecting Gag levels (Figure 4F), again suggesting that clathrin specifically stabilizes Pol proteins in HIV-1 infected cells.
When introduced into GagPol expression vectors that were then used to transfect 293T cells, the class II HIV-1 IN mutations that blocked clathrin incorporation (N184L, F185K or δCTD) had only minor effects on overall particle yield (Figure 5A). However, as expected, viruses encoding these mutations exhibited extremely low infectivity (Figure 5B). More importantly, depletion of clathrin HC using siRNA did not affect overall particle yield from 293T cells (Figure 5C), but caused significant decrease in infectiousness (5 to 20-fold) of HIV-1 particles generated by cotransfection with an HIV-1 proviral plasmid (Figure 5D). Alternatively, overexpression of AP180 C-terminal domain (AP180C), which binds clathrin and induces its mislocalization [21]–[22], reduced the incorporation of clathrin into HIV-1 particles (Figure S5) and also reduced the infectiousness of HIV-1 particles generated from a proviral plasmid by >100 fold (Figure 5E) without affecting physical particle yield (Figure 5C). However, clathrin HC depletion using siRNAs, or perturbation by AP180C overexpression, had comparatively modest, but nevertheless significant, effects (∼3-fold) on the infectivity of VSV G-pseudotyped HIV-1 particles (Figure 5F and 5G). Thus, while clathrin depletion and sequestration clearly impacted Pol protein levels and, consequently, virion infectivity, there was a significant discrepancy in the magnitude of the infectivity impairment induced by clathrin perturbation when VSV-G pseudotyped HIV-1 particles were examined as compared to those generated using the natural HIV-1 envelope. Given clathrin's role in the secretory pathway, it was possible that clathrin perturbation might have effects on the HIV-1 envelope protein that impact virion infectivity, independent of its effects on Pol. In fact, generation of HIV-1 particles in the presence of AP180C resulted in virions that contained primarily unprocessed gp160 envelope protein, and very little gp120 (Figure S6). Therefore, it appears likely that clathrin affects the trafficking of the HIV-1 Env protein, or cellular factors required for Env maturation. While these results indicate caution in the interpretation of the effects of clathrin perturbation on virion infectivity, results with M-PMV (see below) suggest the clathrin perturbation using siRNA or AP180C expression does not cause a non-specific effect on the infectivity of retroviruses, and in particular the function of the VSV-G envelope protein. Thus, the effect of clathrin siRNA and AP180C on VSV-G pseudotyped HIV-1 infectivity (Figure 5F and 5G) should be independent of effects on the envelope protein.
In SIVmac, MLV and M-PMV Gag proteins, we noticed that motifs responsible for clathrin incorporation into virions were situated proximal to motifs responsible for recruitment of factors (ESCRT pathway-associated proteins or ubiquitin ligases) involved in viral budding (Figure 2D and 3A). In the case of SIVmac, one of the two DLL motifs is positioned overlapping the putative ALIX binding site at the C-terminus of p6 (Figure 6A). Therefore, to facilitate an examination of the role for clathrin in SIVmac replication, we first identified residues that were critical for ALIX recruitment, as well as residues that were critical for clathrin recruitment that could be mutated without disrupting ALIX recruitment. An SIVmac Gag expression plasmid was subjected to scanning mutagenesis throughout residues 41–60 of p6 (Figure 6A), with mutations selected so as not to alter the underlying p6-Pol (p6*) protein sequence. Mutant SIVmac Gag proteins were coexpressed with HA-tagged ALIX, and ALIX incorporation into VLPs was assessed by Western blot analyses (Figure 6B). Mutations P44L Y45S, L52S and D51A/L52S dramatically affected ALIX incorporation into VLPs, while the mutations D21A, D51A, L58P and F59S, did not (Figure 6B) These results are consistent with a recent study that also mapped the ALIX binding site in SIVmac p6 [25]. Thus, the SIVmac p6 domain was capable of recruiting clathrin and ALIX into VLPs through overlapping peptide sequences (D21LL D51LL for clathrin and P44Y45 L52 for ALIX), but mutants were readily identified that separately inhibited these activities. Therefore, in the ensuing studies, an SIVmac Gag protein encoding the D21A and D51A mutations in p6 was termed (DLL-) and used as a clathrin–recruitment defective mutant while the P44L/Y45S mutant was termed (PY-) (Figure 6A) and used as a ALIX-recruitment defective mutant.
To investigate the potential role of clathrin recruitment by Gag in SIVmac replication, we first compared the infectivity of particles generated by wild type and (DLL-) SIVmac proviral plasmids in a single round infectivity assay. A modest reduction (2–3 fold) in infectiousness was observed as a consequence of DLL motif mutations, with no obvious effect on physical particle yield (Figure 6C and 6D). Similar results were obtained when VSV-G pseudotyped SIVmac particles were used (unpublished observations). However, in spreading replication assays conducted in C8166 cells, this modest difference was apparently amplified, and the SIVmac (DLL-) virus was highly attenuated compared to the wild-type counterpart (Figure 6E).
Given the proximity of the DLL motifs to the L-domains in SIVmac Gag (Figure 6A), we next tested the effects of mutations predicted to abolish the recruitment of Tsg101 (PTAP-), clathrin (DLL-) and ALIX (PY-), either alone or in combination, on particle release. Initially this was done in the context of SIVmac Gag, in the absence of the viral protease (Figure 6F). Consistent with a previous report [26], mutation of the PTAP motif alone had a surprisingly modest effect on SIVmac particle yield. Moreover, mutation of the ALIX binding sites had no discernable impact on particle yield, when introduced either alone or in combination with the PTAP mutation (Figure 6F and 6G). Strikingly, while the DLL- mutation had no effect on VLP yield when introduced alone, the combination of this mutation with the PTAP- mutation dramatically diminished VLP yield (Figure 6F). This finding held true when the SIVmac Gag protein was expressed using a human codon optimized construct (Figure 6G). Overexpression of ALIX has previously been reported to rescue PTAP-mutant HIV-1 particle release [27], [28] and we found that overexpression of ALIX restored the defect in VLP release associated with the PTAP- DLL- double mutant (Figure 6G). This activity required an intact ALIX binding site, because ALIX overexpression failed to rescue the budding defect in the PTAP-DLL-PY- triple mutant (unpublished observations). This finding indicates that the PTAP-DLL- double mutant is fully competent to assemble into virions, but requires ALIX overexpression to complete budding.
Importantly we also found that overexpression of AP180C recapitulated the effect of the DLL- mutation in SIVmac Gag, and inhibited the release of VLPs assembled using PTAP- Gag (Figure 6H). AP180C overexpression did not inhibit particle release when wild-type Gag was used and this finding suggested that clathrin facilitates the completion of assembly and/or budding of SIVmac particles, particularly when budding is impaired or retarded by inhibition of Tsg101 recruitment.
To further characterize the defects in VLP release imposed by the PTAP and DLL- mutations, we analyzed 293T cells expressing wild-type and mutant SIVmac Gag proteins via scanning electronic microscopy. To eliminate variations from transfection levels, Gag expression constructs were utilized which contained an internal ribosomal entry site (IRES) linked to a GFP coding sequence on the same mRNA as Gag, and cells expressing equivalent amounts of GFP were chosen for analysis. Virions assembled using Gag proteins bearing individual PTAP- or DLL- mutations exhibited no gross morphological abnormalities, and numerous apparently spherical particles were observed on the plasma membrane of cells (Figure 6I and Figure S7). However, PTAP- DLL- double mutant Gag proteins exhibited a clear morphogenesis defect (Figure 6I and Figure S7). Specifically, cells exhibited hemispherical protrusions from their surfaces, but complete spherical particles were almost never observed. Crucially, this morphological defect could be induced using the PTAP- single mutant SIVmac Gag protein (but not the wild type Gag protein) upon overexpression of AP180C (Figure 6J and Figure S8). Together, these data strongly suggest that clathrin plays a facilitating role in the morphogenesis of SIVmac virions that is modest when measured in the context of a single cycle of SIVmac assembly, but is sufficient to strongly enhance replication, and becomes particularly evident when budding is inhibited or slowed by mutation of the PTAP L-domain.
When SIVmac Gag proteins were expressed in the context of an active viral protease, either using GagProtease (Figure 7A) or full-length GagPol (Figure 7B) expression plasmids, the DLL- single mutation had little or no effect on levels of cell associated Gag protein (Figure 7A and 7B). However, when the DLL- mutation was present in combination with the PTAP- mutation, the steady state cell-associated levels of the Gag precursor and processed derivatives were significantly diminished (Figure 7A and 7B). Remarkably, overexpression of ALIX rescued this defect and restored the level of PTAP- DLL- mutant SIVmac Gag proteins to those expressed by the wild type Gag-protease and GagPol expression plasmid (Figure 7A and 7B). Notably, the ability of ALIX to restore PTAP- DLL- mutant SIVmac Gag protein levels required an ALIX binding site, as this effect was not observed when a PTAP- DLL- PY- mutant SIVmac GagPol expression plasmid was used (Figure 7B). The effect of the DLL- mutation on SIVmac Gag expression levels was partially recapitulated by clathrin sequestration using AP180C. Specifically, AP180C overexpression caused a reduction in the levels of Gag protein expressed by a PTAP- PY- mutant, but not wild-type, GagPol expression plasmid (Figure 7C). Taken together, these findings lead to the conclusion that clathrin interaction with SIVmac Gag facilitates virion morphogenesis, with consequent effects on Gag protein stability in the presence of the viral protease, especially when budding is impaired.
The effects of the DLL- mutation on SIVmac Gag protein levels were specific to Gag, and did not affect Pol protein levels (Figure 7B). Similarly, a mutation in SIVmac Pol (analogous to the HIV-1 IN N184L) that blocked Pol dependent clathrin incorporation (Figure 2G) caused a reduction in the level of SIVmac Pol protein to an almost undetectable level, while Gag levels were unaffected (Figure 7B). In contrast to the effect of ALIX on SIVmac Gag expression, this Pol expression defect could not be rescued by overexpression of ALIX. Thus, the effects of mutations that reduce protease-dependent Gag stability and clathrin incorporation were independent of those that affected Pol stability and clathrin incorporation.
Like SIVmac, MLV has a DLL motif proximal to its L-domain that is responsible for clathrin incorporation (Figure 3A). Transfection of an MLV GagPol expression plasmid in which the DLL motif was mutated to either ALL or DAA, along with plasmids encoding VSV-G and a GFP expressing MLV vector resulted in no defect in the yield of physical particles (Figure 8A), but mutant particles were >100-fold less infectious than WT particles (Figure 8B). Similar results were obtained using full-length MLV proviral plasmids (unpublished observations). Similarly, siRNA mediated clathrin depletion did not affect particle yield (Figure 8C). However, clathrin depletion resulted in only a modest reduction of MLV infectivity (Figure 8D). This may be attributable to incomplete clathrin knockdown (approximately 20% of normal clathrin levels remained in siRNA transfected cells, Figure S3A). Consistent with this notion, coexpression of WT and DLL- mutant MLV GagPol at various ratios revealed that particles were nearly fully infectious, even when a small fraction of the Gag proteins contained therein harbored an intact DLL motif (Figure 8E). Therefore, we hypothesize that the residual clathrin recruited by WT virus assembled in siRNA treated cells may be nearly sufficient to fulfill its functional role.
Western blot analysis showed that substitutions in the MLV DLL motif had no effect on cell- or extracellular virion- associated levels of viral precursor Gag (Pr65) or processed p30 CA proteins when expressed in the context of MLV GagPol or proviral plasmids (Figure 8A and unpublished observations). This held true when the DLL mutation was introduced alone, or in combination with mutations in PPPY and/or PSAP motifs (unpublished observations). However, Western blot analysis with an anti-p12 monoclonal antibody revealed a dramatic decrease in the levels of the mature p12 protein associated with virions (Figure 8F). Conversely the partly processed MA-p12 intermediate was present at equivalent abundance in wild type and mutant virions. This finding excluded the possibility that the p12 antibody failed to recognize the DAA mutant p12 sequence and instead suggested the possibilities that either the MA-p12 junction was not efficiently processed in the DAA mutant, or that the DAA mutation created a p12 that was aberrantly cleaved and as a consequence could not be recognized by the p12 antibody. To investigate these possibilities, a small amount of wild-type or DLL-mutant MLV Gag-Pol expression plasmid was co-transfected with increasing amounts of wild-type or mutant Gag expression plasmids (in a Gag-Pol ∶ Gag ratio of 1∶0.5 to 1∶16, (Figure S9)). This analysis suggested that the DAA mutant and WT Gag proteins could be processed in trans by coexpressed GagPol bearing the WT p12 sequence, yielding mature p12 protein in extracellular virions with comparable efficiency. Conversely, the DAA mutant GagPol protein failed to generate the fully processed p12 in virions, even when a WT Gag protein was provided in trans (Figure S9). In a similar experiment, we used a different MLV-related GagPol expression plasmid (from XMRV) whose p12 sequence is poorly recognized by the anti MLV p12 antibody (Figure S10). When XMRV GagPol was coexpressed with WT or DLL mutant MLV Gag proteins, the WT and DLL mutant MLV p12 proteins were detected at approximately equivalent levels in the resulting virions (Figure S10). Thus, both the WT and mutant MLV MA-p12 junction can be efficiently processed by MLV or XMRV protease in trans resulting in the appearance of the mature p12 protein in virions. In other words, it appeared that the DLL motif regulates the activity of the MLV protease in cis, but not in trans, and the absence of the DLL motif in cis to the protease caused aberrant Gag processing and absence of p12 in virions.
While the absence of the p12 protein in DLL-mutant MLV virions appeared symptomatic of the defect induced by the clathrin binding site mutation, it was not clear whether this lesion was the direct cause of the infectivity defect. To test this, we generated MLV particles using a construct, similar to one previously described [29], in which the MA-p12 cleavage site was mutated (S1KK) so that it could not be cleaved, resulting in the absence of the mature p12 protein in virions (Figure 8G). In this context, the DLL- mutation retained its deleterious effect on MLV infectivity (Figure 8H). Thus, these findings suggested the existence of a generalized morphological defect that included, but was not limited to, p12 deficiency in MLV particles consequent to mutation of the clathrin binding site.
Electron microscopic analysis of WT and DLL mutant MLV expressing cells did not reveal any gross morphological alterations in assembling or assembled particles (Figure S11A). Therefore, to test for more subtle morphogenesis defects, we examined whether DLL- mutant particles retained the ability to saturate the TRIM5α restriction factor. This assay should test for the presence of a stable, morphologically accurate capsid lattice, which is expected to be required for efficient binding to, and saturation of TRIM5α. Thus, increasing amounts of N-tropic MLV particles generated using WT or DLL- mutant GagPol expression plasmids (Figure S11B) were applied to human TE671 cells, which express a TRIM5α protein that can recognize and restrict N-tropic MLV capsids [30], [31]. Simultaneously, a fixed and equal amount of GFP-expressing WT B-tropic or N-tropic indicator virus was applied to monitor TRIM5α saturation (Figure 8I). Wild-type N-tropic MLV particles efficiently saturated TRIM5α and thereby facilitated N-tropic MLV infection. However, particles containing a mutation in the DLL motif in p12 were approximately 10-fold less active in this TRIM5α saturation assay (Figure 8I and S11B). The finding that the mutant virions were less ‘visible’ to human TRIM5α suggests that their cores were not properly formed, or unstable and, therefore, that the clathrin-binding motif in p12 is important for accurate MLV particle morphogenesis.
Like SIVmac and MLV, the betaretrovirus M-PMV harbours a motif (DLISLD) proximal to its L-domains that is responsible for recruiting clathrin into virions. Notably, however, the morphogenesis pathway for M-PMV is quite different to that of SIVmac and MLV, in that complete immature capsids are assembled in the cytoplasm and move thereafter to the plasma membrane for envelopment.
Mutation of the clathrin-recruiting motif in M-PMV Gag reduced the infectiousness of M-PMV virions by 5-fold, without affecting the yield of physical particles (Figure 9A and 9B) or causing gross morphological abnormalities that could be visualized by electron microscopic examination (unpublished observations). Notably, clathrin depletion using siRNA or perturbation by overexpression of AP180C had a negative effect on M-PMV infectivity of similar magnitude to that of the clathrin binding site mutation (Figure 9B and 9C). Importantly, however, the effect of depletion or sequestration of clathrin on M-PMV infectivity was specific to the WT virus; these manipulations had no effect on the infectivity of the mutant M-MPV that could not recruit clathrin (Figure 9B and 9C), suggesting that the clathrin binding site mutation and the clathrin perturbation affected the same process.
It was surprising that the clathrin knockdown had a greater effect on M-MPV infectivity compared to MLV infectivity (compare Figure 8D and 9B), while M-PMV was less affected by clathrin binding site mutations than was MLV (compare Figure 8B and 9B). However, experiments in which M-MPV virions were generated using mixtures of WT and DLISLD mutant proviral plasmids showed that reduction in infectivity was approximately proportional to the fraction of the Gag protein that was mutant versus WT (Figure 9D). Thus, M-MPV infectivity appeared more sensitive to partial depletion of clathrin, or partial removal of clathrin binding sites from virions, than did MLV. Nonetheless, optimal infectiousness and replication of both viruses was clearly dependent on clathrin recruitment by their respective Gag proteins.
In this study, we identified clathrin as a component in a variety of retrovirus particles, including members of the lentivirus (HIV-1 and SIVmac), gammaretrovirus (MLV) and betaretrovirus (M-PMV) genera. Specifically, the Gag proteins of SIVmac, MLV and M-PMV recruit clathrin HC using DLL or DLISLD motifs, which mimic those commonly found in classical clathrin adaptors. Additionally, HIV-1 Pol and, to some extent, SIVmac Pol were capable of recruiting clathrin into virions, although the motifs responsible could not be mapped because clathrin incorporation appeared to be dependent on the conformational integrity of multiple domains of the HIV-1 Pol protein.
Phenotypic characterization of mutant viruses and a combination of other techniques to perturb clathrin in virus-producing cells (siRNA-based clathrin HC knockdown or AP180C overexpression), revealed that clathrin can have a range of apparently distinct effects on virion morphogenesis, depending on the particular retrovirus examined. These effects are summarized in Table S2. In the case of HIV-1, mutations or drugs (efavirenz) that affect RT dimerization as well as class II IN mutations are known or thought to affect protease activation [17], [23], [24], [32], [33], [34], [35]. We found that at least some of these mutations blocked clathrin incorporation, and also found that clathrin overexpression could ameliorate deficits in the levels of Pol protein harboring some of the aforementioned mutations. Moreover, depletion or sequestration of clathrin could reduce the levels of Pol proteins in cells. This suggests that clathrin may be directly involved in regulating protease activity, or perhaps in stabilizing or retaining the products of Pol proteolytic cleavage during HIV-1 morphogenesis.
Similarly, mutations in the clathrin-recruiting motif in the p6 domain of SIVmac Gag, as well as sequestration of endogenous clathrin, decreased the level of the viral Gag protein in the presence of the viral protease. However, these effects were only truly evident when the viral PTAP motif was also mutated. Again this finding is consistent with the notion that clathrin regulates proteolysis of the viral protein, or stabilizes the products of proteolysis, and the magnitude of the effect is exaggerated to easily detectable levels if budding is blocked or retarded. Notably, the apparent protease-dependent instability of the PTAP- DLL- mutant Gag protein could be largely reversed by overexpression of ALIX, strongly suggesting that the PTAP- DLL- mutant Gag protein is not generically unstable or acutely sensitive to proteolysis simply as a direct consequence of the introduced mutations. Rather, this observation reinforces the notion that protease-dependent SIVmac Gag depletion, consequent to a failure to recruit clathrin, is exaggerated by retarding the rate of particle budding.
In the case of MLV, there was no apparent effect of the DLL- mutation on Gag precursor stability, even when the proximal PPXY L-domain motif was mutated. However, there was a very clear effect on the products of Gag proteolysis, in that p12 was absent from virions. Moreover, the DLL mutation had major effects on the infectivity of MLV particles, as well as the accuracy of their morphogenesis, or their stability, as evidenced by the relative inefficiency with which MLV cores were apparently recognized by the TRIM5α restriction factor. However the precise nature of the biochemical lesion responsible for this infectivity defect remains to be completely defined. The absence of p12 from virions could contribute to the infectivity defect, as p12 has been shown to be a component of MLV preintegration complexes [36]. However, mutations that prevent the cleavage of p12 from MA did not abolish the effect of the DLL-mutation on MLV infectivity, suggesting additional effects of clathrin on particle morphogenesis.
In addition to effects on viral protein proteolysis, clathrin appears to affect the morphogenesis and release of SIVmac particles in the absence of the viral protease. Again, observing this effect required the PTAP-motif to be mutated, and the effect could be suppressed by overexpression of ALIX. The ability of ALIX overexpression to suppress the two effects of clathrin binding site mutation on SIVmac, namely; (i) an assembly defect in the absence of the viral protease and (ii) Gag instability in the presence of the viral protease, suggests that they are different manifestations of the same defect in clathrin recruitment. This apparent ability to affect both protease-dependent and protease-independent processes influences the generalized models that can be invoked to explain the role of clathrin in retrovirus morphogenesis (see below).
It is interesting that the D51L52L53 motif in SIVmac Gag overlaps with the ALIX binding site, raising the possibility that ALIX and clathrin might compete with each other. Moreover, inspection of multiple viral strains of the SIVsm/SIVmac/HIV-2 lineage reveals that all p6 sequences contain 1, 2 or 3 copies of a DLL motif. In those that contain a single DLL motif, it appears that one copy has been displaced by a PTAP motif. These observations are suggestive of interplay between clathrin and the ESCRT machinery that merits future investigation.
A caveat to the aforementioned conclusions is that mutations, particularly in HIV-1 or SIVmac Pol might have pleiotropic effects, and it therefore was difficult to assign cause and effect in situations where, for example, Pol mutations both blocked clathrin incorporation and caused decreases in the levels of Pol proteins. Nonetheless, depletion or sequestration of clathrin could, in several cases, at least partly recapitulate the effects of mutations that blocked clathrin recruitment. The pleiotropic effect of HIV-1 Pol mutations likely arises from the fact that they cause premature protease activation, and because protease has multiple substrates in Gag and Pol, a variety of effects on the accuracy of particle assembly and Pol protein incorporation, and therefore particle infectiousness, are predictable consequences. It is possible that the pleiotropic effect of these mutations is a consequence of their effects on clathrin recruitment. Indeed, overexpression of clathrin, or fusion of a clathrin binding site in cis could increase the levels of HIV-1 Pol proteins that were apparently destabilized by Pol mutations. However, it was not possible to restore the infectiousness of Pol-mutant particles using this approach.
One factor that must be considered in arriving at a proposal for general model for a role of clathrin in retroviral replication was that the magnitude of the effect on infectious virion yield clearly varied according to which retrovirus was examined. Additionally, clathrin was not found to be efficiently incorporated into all retroviruses that we tested. For instance, equine infectious anemia virus (EIAV) and human endogenous retrovirus K (HERV-K) did not efficiently incorporate clathrin (unpublished observations). Moreover, the apparent effect of clathrin on retrovirus replication usually varied according to whether the experimental manipulation was clathrin binding site mutation, clathrin depletion, or clathrin sequestration. In some cases (e.g. SIVmac Gag), the effect of mutations in defined motifs responsible for clathrin recruitment was relatively modest in a single cycle of replication, while in others (e.g. MLV) the effect was dramatic. In the case of HIV-1, SIVmac and MLV the effects of mutations that abolish clathrin incorporation were much greater than the effects of clathrin depletion or sequestration. While this may be due to pleiotropic effects of the mutations that were introduced, it is also true that clathrin depletion or sequestration using the techniques employed herein was incomplete and, at least in the case of MLV, a relatively small fraction of the viral Gag protein is needed to be capable of recruiting clathrin in order for virions to be nearly fully infectious. Conversely, in the case of M-PMV, the magnitude of the effect of clathrin binding site mutation on particle infectivity was nearly precisely recapitulated by clathrin depletion or sequestration. Moreover, the impaired infectivity of the clathrin recruitment-defective mutant was not further impaired reduced by clathrin depletion or sequestration. Thus, we can be quite confident in the case of M-PMV case that clathrin enhances infectivity primarily through the action of the DLISLD motif in Gag. Details of the mechanism by which clathrin enhances M-PMV infection remain to be investigated and are difficult at present due to the paucity of reagents for studying this virus, but M-PMV may make the most tractable system for future investigations.
It is challenging, based on these data, to make definitive general conclusions as to the role of clathrin in retrovirus replication, because the ultimate outcome, in terms of defined lesions that occurred as a consequence of blocking clathrin recruitment, differed in each retrovirus studied herein (Table S2). Moreover, in some retroviruses Pol was responsible for clathrin recruitment, while in others Gag was responsible. Although conceptually unsatisfying, it may simply be the case that clathrin plays a completely different role in each retrovirus. However, for HIV-1, SIVmac and MLV, a common theme was that clathrin appeared to influence proteolysis of the viral polyproteins. This would likely have different, potentially pleiotropic, consequences for different retroviruses, including infectivity and morphogenesis defects as well as apparent viral protein instability. We observed all of these outcomes in this study and each could be reasonably hypothesized to be the consequence of mis-regulation of viral polyprotein processing.
Models that could potentially explain the aforementioned phenomena include the idea that clathrin contributes to the spatial organization of Gag and Pol proteins during particle assembly. As a homotrimeric Gag or Pol binding protein, clathrin could bind to multiple Gag or GagPol molecules simultaneously. Indeed, the affinity of clathrin for cellular clathrin adaptors, and likely, therefore, assembling Gag or Pol proteins, is dependent on their multimerization/polyvalency [12], [13]. The association of clathrin with assembling Gag and Pol proteins could influence proteolysis of the viral proteins by positively or negatively influencing protease dimerization (which is required for protease activity), by influencing substrate (Gag or Pol) mobility and/or accessibility to the protease, or by helping to retain or remove the products of partial or complete proteolysis from a nascent virion as particle assembly and proteolysis proceeds. Such a model can also be reconciled with the findings that the effects of clathrin appear quite variable, from overt effects on particle morphogenesis and viral protein stability to more subtle effects on virion infectiousness.
The proviral HIV-1 plasmid used throughout was pNL4-3 (NIH AIDS Research and Reference Reagent Program, Catalog No. 114). A protease defective variant, and other mutants thereof, were constructed by generating PCR products harboring mutations in PR or RT or IN, that were digested with SphI and EcoRI before subcloning. The SIVmac proviral plasmid was previously described [37] and based on SIVmac239. An MLV proviral plasmid (pNCS) was a gift from Stephen Goff and the M-PMV proviral plasmid pSARM-4 was a gift from Eric Hunter.
Overlap-extension PCR was used to generate HIV-1, SIVmac, MLV, and MPMV mutants. For HIV-1 Gag, HIV-1 GagPol, SIVmac Gag, SIVmac GagProtease (Gag-PR) and SIVmac GagPol expression plasmids, wild-type or mutant encoding sequences were amplified by PCR, using primers that incorporated 5′EcoRI and 3′ NotI sites and inserted into the HIV-1-based expression plasmid pCRV1 [38]. In some cases an HA epitope was inserted, fused in frame at the C-terminus of the expressed protein into pCRV1 vector. The panel of SIVmac double mutants PTAP- (PTAP/LTAL) DLL-(DLL/ALL), PTAP- (PTAP/LTAL) PY-(PY/LS) and DLL-(DLL/ALL) IN-N184L and triple mutants PTAP- (PTAP/LTAL) PY-(PY/LS) DLL- (DLL/ALL) were constructed by introducing additional point-mutation into pre-constructed single or double mutants, respectively. To generate chimeric HIV-1/SIVmac Gag proteins (Figure 2B), overlap-extension PCR was performed with primers targeting the corresponding region of SIVmac or HIV-1 Gag and PCR products were inserted along with a C-terminal HA-epitope tag into pCRV1. To insert mutant SIVmac p6 at the C-terminus of mutant HIV-1 GagPol, NotI sites were incorporated at both 5′ and 3′ end of the SIVmac p6-encoding PCR product which was inserted between HIV-1 Gag-Pol and the HA epitope in pCRV1.
Plasmids expressing codon-optimized SIVmac Gag, pCR3.1/SIVmac-Gag, as well as codon-optimized HIV-1 Gag, codon-optimized HIV-1 Gag-Pol and codon optimized GagGFP, were previously described [38] and mutations were made by overlap-extension PCR. To construct pCR3.1/SIVmacGag-IRES-eGFP expressing both SIVmac Gag and GFP, the IRES-eGFP region was amplified, using pIRES2-EGFP(Clontech) as template, digested by NotI and subcloned at the C-terminus of SIVmac Gag in WT and mutant pCR3.1/SIVmac-Gag expression plasmids.
For MLV mutants, PCR products harboring D156LL- mutations (DLL/ALL or DLL/DAA) were digested using BsrGI and XhoI and inserted into pCAGGS-based MLV Gag or Gag-Pol expression vectors [39] or an N-tropic MLV GagPol expression plasmid [30]. For M-PMV mutants, PCR products bearing the DL129I130SLD/DAASLD mutation were digested by SmaI/SacI and inserted into pSARM-4 [40] or pSARM-X-eGFP as described in [41].
Plasmids expressing clathrin HC residues 1–363aa or 1–494aa with a C-terminal HA or Myc epitope were amplified by using full-length clathrin HC as a template and inserted into pCR3.1 (Invitrogen).
The plasmid expressing pCR3.1/ALIX, the HIV-1 proviral plasmid bearing YFP in the stalk region of matrix (pNL4-3 MA-YFP) and the XMRV (xenotropic murine leukemia-related virus) GagPol expression plasmid were described previously [42], [43], [44]. The plasmid expressing FLAG-AP180C was a gift from Lois Greene, a plasmid expressing DsRed-clathrin LC was a gift from Sanford Simon and pSIV-T1 was from Francois-Loic Cosset [45].
Monoclonal antibodies included anti-HA (Covance), anti- Clathrin HC (BD Transduction laboratories), anti-FLAG (Sigma), anti-HIV capsid p24 (183-H12-5C), anti-Env (1D6), anti-RT (MAb21) (all from NIH AIDS Research and Reference Reagent Program), anti-HIV IN (a gift from Michael Malim), anti-MLV capsid p30 (ATCC CRL-1912 R187), anti-MLV p12 (ATCC, CRL-1890 548), and anti-M-PMV Gag (a gift from Eric Hunter) [46]. In addition, secondary antibodies included goat anti-mouse or anti rabbit IgG conjugated to horseradish peroxidase, or to Alexa Fluor 488 (Invitrogen), IRDye® 800CW and IRDye® 680 (LI-COR Biosciences).
Adherent cell lines from human (293T, TE671, TZM-bl) were maintained in DMEM supplemented with 10% fetal calf serum and gentamycin. CD4+lymphoid cell lines CEMx174, MT2, MT4 and C8166 were grown in RPMI/10%FCS/antibiotics.
In general, for transfection experiments in 293T cells, cells were seeded at a concentration of 1.5×105 cells/well (24-well) or 3×105 cells/well (12-well) or 2×106(10-cm dish) and transfected the following day using polyethylenimine (PolySciences).
To identify proteins incorporated into HIV-1 VLPs, 10 µg of an HIV-1 GagPol and Gag expression plasmids were transfected into 293T cells in 10 cm dishes. Particles in 30 ml of filtered supernatant were pelleted through 20% sucrose, resuspended in PBS, treated with subtilisin (Sigma) and separated on Optiprep gradients. Particulate material in each of 8 fractions was recovered by diluting the fractions with PBS and ultracentrifugation and analyzed by SDS-PAGE followed by Coomassie blue or silver staining. Bands of interest were excised and protein identification was done by the Rockefeller University Proteomics Center. To determine domains responsible for clathrin incorporation, 293T cells were transfected with 5 µg of HIV-1 or SIVmac GagPol or Gag expression plasmid along with 3 µg of clathrin heavy chain 1–494aa or clathrin heavy chain 1–363aa or 0.5 µg GFP-HA expression plasmid. Analysis of clathrin incorporation into other viruses was performed by transfection of 8 µg of MLV GagPol expression plasmid or 8 µg of M-PMV proviral plasmid, with or without 2 µg of clathrin HC 1–494aa or 2 µg of clathrin HC 1–363aa expression plasmids followed by ultracentrifugation and subtilisin treatment of pelleted virions.
To measure SIVmac VLP release, cells were transfected with 150 ng of native sequence or codon-optimized SIVmac Gag expression plasmids, along with 300 ng of ALIX expression plasmid where indicated. In Figure 6H, 500 ng of codon optimized SIVmac Gag was cotransfected with 0 ng, 0.5 µg, 1.0 µg or 1.5 µg of AP180C expression vector. VLPs were harvested at 24 hrs post transfection, pelleted through 20% sucrose and subjected to Western blot analysis.
To define the ALIX binding motif on SIVmac Gag, 500 ng of plasmids expressing wild-type or mutant SIVmac Gag proteins were cotransfected with 500 ng of pCR3.1/HA-ALIX into 293T cells. Cell lysates and VLPs were harvested and analyzed by Western blotting after 40 hrs.
To generate infectious viruses bearing a GFP reporter, 293T cells were transfected with 200 ng of an HIV-1 GagPol expression plasmid and 200 ng of reporter vector pHRSIN-CSGW [47] (for HIV-1), 200 ng of an SIVmac GagPol expression plasmid and 200 ng of reporter vector pSIV-T1 (for SIVmac), 200 ng of an MLV GagPol expression plasmid and 200 ng of reporter vector pCNCG [30] (for MLV), or 400 ng of pSARM-X-eGFP (for M-PMV), along with 100 ng of plasmid expressing the VSV-G envelope. To measure the effect of AP180C on infectiousness of viruses, 500 ng of empty vector or AP180C expression plasmid was added to the transfection mixture. To generate the VSV-G pseudotyped noninfectious VLPs used in the TRIM5 saturation assays in Figure 8I and Figure S11B, 10 µg of WT or DLL- mutant N-tropic MLV and 2 µg of VSV-G expression plasmid were used to transfect a 10 cm dish of 293T cells.
To determine the effects of AP180C on HIV-1 or SIVmac viral protein expression level, 200 ng of HIV-1 proviral plasmid was cotransfected with 0 ng, 300 ng, 600 ng, 900 ng of AP180C expression plasmid (Figure 4F), or 200 ng of SIVmac GagPol expression plasmid was cotransfected with 0 ng, 250 ng, 500 ng or 1000 ng of AP180C expression plasmid (Figure 7C).
To assess the ability of the MLV protease to process WT or mutant Gag proteins (Figure S9), 50 ng of WT or DLL/DAA mutant MLV GagPol expression plasmid were contransfected with increasing amounts (0, 25 ng, 50 ng, 100 ng, 200 ng, 400 ng, 800 ng) of plasmid expressing WT or DLL/DAA mutant MLV Gag. Alternatively, in Figure S10, 6 µg of XMRV GagPol expression plasmid were cotransfected with 300 ng of WT or DLL/DAA MLV Gag expression plasmid. Cell lysates and VLPs were analyzed by Western blotting 48 hrs later.
In all transfection experiments, the total amount of DNA was held constant within the experiment by supplementing transfection mixtures where necessary with empty expression vector.
Transfected 293T cells were placed in fresh medium at 20 hrs post infection and virion containing cell supernatants were harvested and filtered (0.22 µm) at 40 hrs post transfection. Infectious virus release (for HIV-1 and SIVmac proviral plasmid derived virus) was determined by inoculating TZM cells seeded in 96 well plates at 1.2×104 cells/well. At 48 hrs post infection, β-galactosidase activity was determined using GalactoStar reagent as per the manufacturer's instructions.
Reporter viruses (HIV-1, SIVmac, Moloney MLV and M-PMV) bearing a GFP indicator gene were generated by transient transfection of 293T cells along with VSV G envelope protein, in the absence or presence of AP180C as indicated above. TE671 cells were seeded one day prior to infection, inoculated with GFP reporter viruses and FACS analysis was carried out using Guava EasyCyte instrument (Guava Technologies).
TE671 were seeded at 2×104 cells/well in 24-well plates one day before infection. Cells were inoculated with a fixed dose of N-MLV or B-MLV GFP reporter virus, selected so that infection with the restricted virus in the absence of VLPs gave low, but measurable, levels of infection (about 0.1% GFP-positive cells). Restriction-abrogating VLPs were serially diluted two-fold and added to the target cells simultaneously with the fixed dose of N-MLV in the presence of polybrene. Infection by the GFP reporter virus was measured 48 h later as described above.
293T cells were transfected with 60–100 pmol of a clathrin-specific RNA duplex (SMART pool, Dharmacon) or a control firefly luciferase duplex (Dharmacon) using Lipofectamine2000 (Invitrogen) according to manufacturer's instructions. Twenty-four hours post-transfection cells were harvested and replated. Forty-eight hours after the first transfection, cells were co-transfected with siRNA (as above) and 200 ng of proviral plasmid or, in the case of GFP-reporter viruses, 200 ng of GagPol expression plasmid, 200 ng of corresponding retroviral reporter plasmid along with 100 ng of VSV-G expression plasmid using Lipofectamine2000 (Invitrogen). Virions and cells were harvested 48 hours later.
Cell lysates and pelleted virions or VLPs (recovered by centrifugation through 20% sucrose) were separated on NuPage Novex 4–12% Bis-Tris Mini Gels (Invitrogen). Proteins were blotted onto nitrocellulose membranes. Thereafter, the blots were probed with primary antibodies and a corresponding peroxidase conjugated secondary antibody and were developed with WestPico chemiluminescent detection reagents (Pierce). Alternatively, blots were probed with antibodies as above, followed by secondary antibodies conjugated to IRDye 800CW or IRDye 680. Fluorescent signals were detected and quantitated using Odyssey (LI-COR Biosciences).
293T cells stably expressing DsRed-Clathrin light chain were seeded on 3.5-cm, glass-bottomed dishes coated with poly-L-Lysine (Mattek). The following day, they were transfected with a plasmid expressing AP180C, using Lipofectamine 2000. Cells were fixed 24 hrs later and observed by deconvolution microscopy using an Olympus IX70-based Deltavision microscopy suite as described previously [48].
To generate fluorescent VLPs for microscopic analysis, 293T cells stably expressing DsRed-clathrin LC were cotransfected with plasmids expressing codon-optimized HIV-1 Gag, or Gag-Pol along with a plasmid expressing codon optimized Gag-GFP at a ratio of 4∶1 or 8∶1. Forty-eight hours post transfection the culture supernatants were pelleted through a 20% sucrose cushion. The resulting VLPs were resuspended in PBS, 0.22 µm filtered, and diluted 1∶1 with PBS containing 3% paraformaldehyde. They were fixed overnight onto poly-L-Lysine coated glass bottom dishes and, after permeabilization with 1% Trition X-100 and 0.5% SDS, the VLPs were stained with anti-HIV capsid antibody, and analyzed via microscopy (Deltavision, Applied Precision). Alternatively, proviral plasmids pNL4-3 and pNL4-3 MA-YFP were transfected in a 1∶1 ratio into 293T expressing DsRed-clathrin LC, subjected to the above protocol but were visualized without immunostaining.
For scanning electronic microscopy studies, 293T cells were transfected with plasmids expressing codon optimized pCR3.1/SIVmac Gag-IRES-eGFP cassette and inspected by fluorescent and scanning electron microscopy using a Hitachi S4700 field emission SEM (University of Missouri Electron Microscopy Core Facility).
For transmission electron microscopy, samples were fixed using paraformaldehyde and glutaraldehyde, postfixed using osmium tetroxide and then dehydrated and embedded in Embed-812 (EMS). Following polymerization, approximately 65 nm sections were cut with an ultramicrotome and mounted on copper grids. Sections were stained with uranyl acetate followed by lead citrate and imaged using a FEI TECNAI G2 transmission electron microscope.
|
10.1371/journal.ppat.1003706 | CXCR3-Dependent CD4+ T Cells Are Required to Activate Inflammatory Monocytes for Defense against Intestinal Infection | Chemokines and their receptors play a critical role in orchestrating immunity to microbial pathogens, including the orally acquired Th1-inducing protozoan parasite Toxoplasma gondii. Chemokine receptor CXCR3 is associated with Th1 responses, and here we use bicistronic CXCR3-eGFP knock-in reporter mice to demonstrate upregulation of this chemokine receptor on CD4+ and CD8+ T lymphocytes during Toxoplasma infection. We show a critical role for CXCR3 in resistance to the parasite in the intestinal mucosa. Absence of the receptor in Cxcr3−/− mice resulted in selective loss of ability to control T. gondii specifically in the lamina propria compartment. CD4+ T cells were impaired both in their recruitment to the intestinal lamina propria and in their ability to secrete IFN-γ upon stimulation. Local recruitment of CD11b+Ly6C/G+ inflammatory monocytes, recently reported to be major anti-Toxoplasma effectors in the intestine, was not impacted by loss of CXCR3. However, inflammatory monocyte activation status, as measured by dual production of TNF-α and IL-12, was severely impaired in Cxcr3−/− mice. Strikingly, adoptive transfer of wild-type but not Ifnγ−/− CD4+ T lymphocytes into Cxcr3−/− animals prior to infection corrected the defect in inflammatory macrophage activation, simultaneously reversing the susceptibility phenotype of the knockout animals. Our results establish a central role for CXCR3 in coordinating innate and adaptive immunity, ensuring generation of Th1 effectors and their trafficking to the frontline of infection to program microbial killing by inflammatory monocytes.
| Inflammatory monocytes have recently emerged as important effectors in intestinal defense against enteric pathogens, but requirements for their activation are poorly defined. Here we use the protozoan Toxoplasma gondii, an orally acquired Th1-inducing pathogen, to study the requirements for inflammatory macrophage activation in the intestinal mucosa. We find that CD4+ T lymphocytes, recruited in dependence upon their expression of the chemokine receptor CXCR3, mediate activation of intestinal mucosa inflammatory monocytes via secretion of IFN-γ, in turn resulting in control of infection. Thus, CXCR3 functions as a critical lynchpin coordinating anti-microbial communication between T lymphocytes and inflammatory monocyte effectors.
| The intestinal mucosa is a critical effector site for elimination of enteric pathogens. Toxoplasma gondii, a ubiquitous protozoan parasite, is a prime example of such a pathogen. Mammals are infected with T. gondii primarily by the ingestion of tissue cysts from undercooked meat or oocysts excreted in the feces of felines, which are the sole definitive hosts. Upon infection, the parasite induces a potent Th1 immune response that is characterized by high levels of IL-12 and IFN-γ [1], [2]. Initial IL-12 production is largely the result of MyD88-dependent Toll-like receptor (TLR) signaling in dendritic cells, and the parasite profilin molecule has been identified as a ligand for TLR11 and TLR12 [3]–[7]. IL-12 activates natural killer (NK) cells to initiate IFN-γ production and promotes T-cell differentiation towards a Th1 program. Ultimately IFN-γ is the critical cytokine involved in controlling Toxoplasma. While in vitro experiments suggest that macrophages activated by this cytokine acquire anti-Toxoplasma activity through upregulation of immunity-related GTPase (IRG) molecules that mediate destruction of the parasitophorous vacuole [8]–[10], the in vivo function of IFN-γ is less clear.
Inflammatory monocytes are an important component of defense against microbial pathogens, including Toxoplasma [11]. These cells express high levels of Ly6C/G (Gr-1) and are recruited from the bone marrow via chemokine (C-C motif) receptor 2 (CCR2) [12]. During Listeria monocytogenes infection, inflammatory monocytes are recruited from the bone marrow to the spleen and liver where they differentiate into TNF-α- and nitric oxide (NO)-producing DCs (Tip-DCs). There they are essential for bacterial clearance and mouse survival [13], [14]. Likewise, CCR2-dependent inflammatory monocytes are recruited to the lung during Mycobacteria tuberculosis infection where they protect mice from disease by recruiting and activating T cells and by producing NO [15], [16]. Mucosal defense against T. gondii has also recently been shown to require CCR2-dependent inflammatory monocytes [11]. Upon recruitment to the small intestine, these cells control the parasite either indirectly by production of IL-12 and TNF-α or directly through production of NO and IRG proteins [4]–[6], [8], [9], [11], [17]. While CCR2 enables recruitment of inflammatory monocytes to sites of infection, the factors that coordinate their activation and acquisition of effector function are not known.
CXCR3 is a Th1-associated chemokine receptor, and cells expressing this receptor respond to the IFN-γ-inducible chemokines CXCL9, 10, and 11 [18]. The receptor is expressed predominantly by T cells and NK cells and is rapidly upregulated upon cell activation. There is evidence that CXCR3 expression enables T-cell entry into sites of infection, although the outcome of recruitment varies among pathogens. In the case of Leishmania major, recruitment is protective as CXCR3-expressing T cells are required for the resolution of cutaneous lesions [19]. However, in the case of Plasmodium berghei ANKA, CXCR3 is pathogenic because it allows entry of proinflammatory cells into the CNS, resulting in cerebral malaria [20].
Here we determined the role of CXCR3 in the intestinal immune response to Toxoplasma. We found that loss of CXCR3 negatively affected host survival against oral infection. This was associated with diminished recruitment of CD4+ T cells to the lamina propria (LP), decreased T cell IFN-γ secretion, impaired inflammatory monocyte effector function, and inability to control the parasite in the intestinal mucosa. Reconstitution with CXCR3-competent CD4+ T cells restored inflammatory monocyte function, resulting in improved survival against the parasite. Protective effects of adoptively transferred CD4+ T lymphocytes depended upon their ability to produce IFN-γ, but occurred independently of CD4 expression of CD40L. Our data show that CXCR3 enables Th1 recruitment to the intestinal LP, where these cells instruct activation of CCR2-dependent inflammatory monocytes, in turn controlling infection. These results establish CXCR3 as a major determinant orchestrating communication between effectors of innate and adaptive immunity, enabling effective host defense against infection.
Because CXCR3 and its chemokine ligands are strongly associated with Th1 responses, we asked whether this proinflammatory axis was induced during Toxoplasma infection in the intestinal mucosa. Accordingly, mice were orally inoculated with cysts, and relative levels of CXCR3, CXCL9 and CXCL10 mRNA expression were measured over the course of acute infection. We found strong upregulation of CXCR3 and its specific chemokine ligands as early as Day 4 post-infection in both the ileum and mesenteric lymph nodes (MLN) (Fig. 1A). Overall, peak CXCR3 mRNA levels were attained by Day 6 post-inoculation.
In order to examine CXCR3 expression in more detail, we utilized Cxcr3 eGFP reporter (CIBER) mice, a bicistronic reporter strain in which cells expressing CXCR3 also express eGFP [21]. We found a large increase in CXCR3 populations of both CD4+ and CD8+ T lymphocytes in MLN, spleen (SPL) and LP compartments following infection (Fig. 1B). In general, CXCR3 upregulation was most pronounced in the CD4+ population. For example, in the MLN there was a 6-fold increase in CD4+CXCR3+ cells but only a 2-fold increase in CD8+CXCR3+ lymphocytes. NK cells are also known to express CXCR3 and are an important source of early IFN-γ during T. gondii infection [22], [23]. However, while CXCR3-GFP expression was relatively high on naïve NK cells, the GFP expression was in fact reduced during infection, suggesting lack of a role for CXCR3+ NK cells during intestinal infection (Fig. S1). We next examined expression of the activation marker CD27 on GFP+ and GFP− T lymphocytes in infected mice. CD27 was significantly lower in CXCR3−GFP− cells, suggesting an altered maturation state of the CXCR3− T cells (Fig. S2A and B). Likewise, there was a lower percentage of CD27+ cells amongst CXCR3-GFP− CD8+ T lymphocytes, although the decreases in expression were not as striking as with the CD4+ lymphocytes (Fig. S2C and D).
To further examine the role of CXCR3 during T. gondii infection, mice deficient in CXCR3 were orally inoculated with low virulence ME49 cysts, and the outcome of infection was monitored. While all wild-type (WT) mice survived acute infection with 30 cysts, Cxcr3−/− animals displayed increased susceptibility with nearly 75% of mice dying by 2 weeks post-infection (Fig. 2A). When the cyst dose was increased to 50, all CXCR3 knockout (KO) mice rapidly succumbed to infection, but some WT mice also died (Fig. 2B). Interestingly, when WT and KO mice were infected by intraperitoneal injection, lack of CXCR3 did not affect survival, indicating that the effect of CXCR3 is specific to the mucosal response (Fig. S3A). To further examine the overall response in orally infected mice, we examined the gross appearance of the small intestine of WT and Cxcr3−/− mice after 30-cyst infection. The small intestines of the KO mice were strikingly damaged as demonstrated by massive hemorrhage compared to WT (Fig. 2C). Consistent with intestinal shortening associated with increased damage [24]–[26], the length of the small intestine was reduced in the KO mice during infection (Fig. 2D). Increased damage was further confirmed by H&E staining of small intestinal sections. WT mice displayed minor villus blunting accompanied by moderate to severe inflammatory cell recruitment in the submucosa (Fig. 2E and G). In contrast, Cxcr3−/− mice displayed severe villus blunting, fusion, epithelial necrosis, sloughing of villus tips, and vascular congestion and hemorrhage (Fig. 2F and H). Blind scoring of H&E sections revealed a significant decrease in inflammation scores in the absence of CXCR3 (Fig. 2I), but when parameters of intestinal damage were quantitated, Cxcr3−/− mice scored significantly higher than WT counterparts (Fig. 2J). This damage was infection-dependent as intestines from non-infected WT and Cxcr3−/− mice both had normal architecture with few inflammatory cells (Fig. S3B). Increased epithelial damage in the absence of CXCR3 was further verified by loss of epithelial surface-associated Muc1 compared to infected WT animals, suggesting epithelial cell sloughing (Fig. S3C). Despite the overall decreased inflammatory score, Cxcr3−/− mice consistently displayed an influx of neutrophils into the LP compartment compared to WT mice, suggesting a role for these cells in causing damage, as argued by others [13], [14], [27] (Fig. S3D and E).
Genetic knockout of cytokines such as IFN-γ results in susceptibility to T. gondii through the inability to control parasite replication, whereas the deletion of anti-inflammatory mediators such as IL-10 results in susceptibility due to cytokine pathology [28], [29]. Based on decreased inflammation scores, we hypothesized that the Cxcr3−/− mice were more likely to be succumbing from uncontrolled parasite replication rather than immune-mediated damage. To examine this, intestinal tissues were stained for parasite antigen by immunohistochemistry. Sections from WT mice displayed minimal parasite infiltration within the LP (Fig. 3A). Conversely, Cxcr3−/− mice contained numerous large foci of parasite throughout the length of the small intestine that often coincided with areas of severe damage (Fig. 3B). Surprisingly, this difference was restricted to the LP and submucosa of the small intestine because Peyer's patches (PP) in WT and Cxcr3−/− mice contained similar levels of parasite antigen (Fig. 3C and D). Differences in parasite burden between WT and Cxcr3−/− mice in the MLN, spleen, and lung were also indiscernible by IHC analysis (data not shown). These results were further confirmed by quantitative PCR. Thus, while lung, liver, spleen, MLN and PP contained similar levels of parasite genomes regardless of CXCR3 expression, there was an approximately 50-fold increase in parasite levels in the absence of CXCR3 in intestinal tissues (Fig. 3E). These data suggest that increased susceptibility to Toxoplasma in Cxcr3−/− mice was due to a localized inability to control parasite replication within the LP of the small intestine.
The dominant effector cells required for elimination of T. gondii following oral infection are inflammatory monocytes. These cells express Ly6C/G (Gr-1), produce TNF-α, IL-12, and are likely to kill parasites via activation of IFN-γ-inducible p47 GTPases that assemble at the parasitophorous vacuole membrane and mediate its destruction [8], [11]. Consistent with others [11], we observed these cells in the LP of infected mice (Fig. 4A). Inflammatory monocytes are dependent upon CCR2 for exit from the bone marrow, but we wondered whether CXCR3 might be involved in recruiting these cells to the LP in response to T. gondii. Therefore, we examined CXCR3-GFP expression by intestinal inflammatory monocytes during infection. Inflammatory monocytes in the small intestinal LP of infected reporter mice did not express any GFP as compared to inflammatory monocytes isolated from infected Cxcr3−/− mice (Fig. 4B). In stark contrast, approximately 50% of LP CD4+ T cells expressed high levels of GFP (Fig. 4C). Furthermore, Cxcr3−/− mice displayed unaltered total numbers of LP inflammatory monocytes compared to wild-type controls (defined as CD11b+Ly6C+Ly6G−) (Fig. 4D). We next assessed the kinetics by which CD4+ T cells and inflammatory monocytes were recruited to the lamina propria during infection. Between days 4 and 7 of infection, there was a significant increase in the total numbers of CD4+CXCR3−GFP+ T cells and inflammatory monocytes (Fig. 4E). However, the total number of CD4+CXCR3−GFP− cells remained unchanged, further indicating that infection promotes the recruitment of CD4+CXCR3+ T cells (Fig. 4E). Few NK cells were observed in the lamina propria, but there was a small increase in their number during infection. This was attributable to an increase in CXCR3− NK cells (data not shown). Consistent with these results, there was an influx of CD4+ T cells in WT small intestines that was diminished in Cxcr3−/− mice (Fig. 4F–H). These findings demonstrate that CD4+ T cells fail to effectively traffic to the intestinal compartment in the absence of CXCR3, but the presence of LP inflammatory monocytes does not require this chemokine receptor.
We next asked if expression of CXCR3 affected the ability of T cells to secrete the Th1 cytokine IFN-γ. Initial experiments on bulk populations of splenocytes and mesenteric lymph node (MLN) cells from Day-11 infected WT and KO revealed no differences in the amount of IFN-γ, TNF-α or IL-10 produced during in vitro culture (Fig. S4). To specifically examine functional outcomes in intestinal cells, WT and Cxcr3−/− lamina propria leukocytes were harvested 4 and 6 days post-oral infection, stimulated ex vivo, and IFN-γ production was examined by flow cytometry. CD4+ T cells from both WT and Cxcr3−/− displayed enhanced IFN-γ production over time. However, in the absence of CXCR3, CD4+ T cells produced significantly less IFN-γ compared to WT cells at both examined time points (Fig. 5A–E). This was confirmed by measuring IFN-γ from the supernatants of Day-6 WT and Cxcr3−/− intestinal biopsy cultures, where IFN-γ was lower in the absence of CXCR3 (Fig. 5F). This effect was specific to the CD4+ T cell subset, as IFN-γ production by lamina propria CD8+ T cells and NK cells was unchanged between WT and knockout animals (Fig. S5A–F). Further confirming that this loss of IFN-γ production was specific to CD4+ T lymphocytes in the small intestine, and consistent with the bulk splenocyte culture experiments, splenocytes isolated from infected WT and Cxcr3−/− mice secreted equivalent levels of IFN-γ upon ex vivo stimulation with PMA and ionomycin (Fig. S5G–I). Together, these results indicate an intestine-specific defect in presence of CD4+ Th1 cells in the absence of CXCR3 as measured by the capacity to produce IFN-γ.
Although IFN-γ, IL-10, and TNF-α responses remained intact in the MLN and spleen late during infection of CXCR3-deficient mice, a significant decrease in IL-12 production was observed in the MLN (Fig. 6A) and spleen (Fig. 6B) of Cxcr3−/− mice. Defective IL-12 responses in the CXCR3 KO strain were infection dependent, because parasite antigen stimulated equivalent amounts of IL-12 in noninfected WT and KO splenocytes (Fig. 6C). This response, known to derive from resident splenic CD8α+ DC [30], may account for equivalent Th1 priming in secondary lymphoid organs, despite lower IL-12 levels during late acute infection.
Since IL-12 is also a characteristic cytokine of inflammatory monocytes, we investigated the impact of CXCR3 deletion on intestinal inflammatory monocyte function. Indeed, in vitro culture of intestinal biopsy samples revealed decreased production of IL-12 (Fig. 6D). To further identify the source of the defective IL-12, we examined the production of IL-12 from inflammatory monocytes. While the total numbers of LP inflammatory monocytes were equivalent in WT and CXCR3 KO mice (Fig. 4B), the population of CD11b+Gr-1+ cells co-expressing IL-12 and TNF-α was dependent upon CXCR3 (Fig. 6E and F). Further confirming impaired inflammatory monocyte function, iNOS expression was significantly decreased in inflammatory monocytes (Fig. 6G and H). These findings strongly suggest that inflammatory monocytes are functionally impaired in the absence of CXCR3. Interestingly, neutrophils in the LP of KO mice produced significantly higher levels of TNF-α compared to WT neutrophils (Figure S6A and B).
Given the data so far, we hypothesized that CD4+ T cells were unable to effectively home to the small intestine and prime inflammatory monocyte function in the absence of CXCR3, resulting in susceptibility to Toxoplasma. We therefore tested whether reconstitution with CXCR3-competent CD4+ T cells would allow Cxcr3−/− mice to overcome susceptibility and restore inflammatory monocyte function. Accordingly, CD4+ T cells from naïve WT spleens were enriched to 90–95% purity by magnetic bead separation (Fig. 7A) and injected i.v. into Cxcr3−/− recipients. Mice were orally challenged with T. gondii 24 hours post-adoptive transfer, and survival was monitored. Knockout mice that did not receive WT cells began to die 10 days post-challenge, while all KO mice that received CD4+ T cells and all WT controls survived the acute phase of infection (Fig. 7B). To confirm that the CD4-dependent survival was not an artifact of the transfer, Cxcr3−/− CD4+ T cells were adoptively transferred into KO recipients. The knockout cells were unable to protect against susceptibility, demonstrating that protection is dependent on CXCR3 expression by CD4+ T cells (Fig. S7A).
To assess the functional impact of WT CD4 adoptive transfer, parasite burden and cytokine production in the LP were assessed by flow cytometry in WT, Cxcr3−/−, and Cxcr3−/− +CD4 mice. While Cxcr3−/− mice had a clear increase in Toxoplasma infected cells relative to WT, upon adoptive transfer of WT CD4+ T cells, parasite levels were reduced to WT (Fig. 7C). Likewise, expression of IL-12/TNF-α by inflammatory monocytes was significantly reduced in KO mice, but these cytokines returned to WT levels upon transfer of WT CD4+ T cells (Fig. 7D and Fig. S7B). In addition to cytokine responses, intestinal damage was also alleviated by adoptive transfer of WT CD4+ T cells, as the intestinal lengths of the transferred mice were restored to WT (Fig. 7E). Possibly as a result of improved monocyte function and parasite clearance, neutrophil levels and neutrophil TNF-α secretion were also restored to WT levels following the transfer of WT CD4+ T cells (Fig. S7C and D).
To identify the mechanism behind the rescue of Cxcr3−/− susceptibility by WT CD4+ T cells, we performed adoptive transfer experiments utilizing T cells derived from knockout animals. Inasmuch as CXCR3 is a Th1 chemokine receptor, we began by asking whether reversal of susceptibility was dependent on CD4-derived IFN-γ Therefore, CD4+ T cells were isolated from Ifnγ mice and adoptively transferred into CXCR3-deficient recipients. Unlike IFN-γ-competent CD4+ T cells (Fig. 7B and Fig. 8B), transfer of IFN-γ KO CD4+ T lymphocytes failed to provide significant protection (Fig. 8A). It has been shown that CD40L contributes to inflammatory responses in the intestinal mucosa during oral Toxoplasma infection [31]. Therefore, we performed the adoptive transfer using Cd40l−/− CD4+ T cells and assessed survival. As expected, CXCR3-deficient animals were highly susceptible to infection. However, Cxcr3−/− mice receiving Cd40l−/− CD4+ T cells survived the infection, indicating that CD40L does not mediate CXCR3+CD4+-dependent protection (Fig. 8B).
Effective control of pathogens such as T. gondii requires the coordinated action of cells of innate and adaptive immunity. Orchestration of the response is governed by an underlying network of chemokines and chemokine receptor-expressing cells in both the hematopoietic and non-hematopoietic compartments. In this study, we demonstrate a central role for chemokine receptor CXCR3 in empowering Th1 trafficking to the small intestine, in turn enabling inflammatory monocyte activation and concomitant control of infection. While the importance of IFN-γ-secreting CD4+ T cells in resistance to Toxoplasma is well known to researchers in the field [28], [32], [33], and while the importance of anti-microbial inflammatory monocytes has recently become clear in the context of Toxoplasma and other infections [11], the present study is the first to reveal the functional link between CXCR3+ T-cell effectors, IFN-γ and inflammatory monocyte activation in tissues of the intestinal mucosa.
Although Th1 effectors depend upon CXCR3 to reach the site of infection, inflammatory monocytes require chemokine receptor CCR2 for optimal trafficking. In the latter case, inflammatory monocytes fail to exit the bone marrow in Ccr2−/− mice, resulting in a decreased level of this population in the periphery, in turn resulting in inability to control T. gondii infection [11]. Monocytes have also been suggested to promote the systemic dissemination of T. gondii to the brain [34]. While brain parasite loads were not examined in this study, it is unlikely that altered parasite shuttling is a mechanism by which the knockout animals are succumbing to Toxoplasma because peripheral parasite loads were not affected by absence of CXCR3 (Fig. 3B). Our data argue that the basis for increased parasites specifically in the intestine is the result of defective regional control by inflammatory monocytes that lack CXCR3-dependent activation signals. Recent data indicate that inflammatory monocyte expression of CCR1 enables a response to IL-15-dependent CCL3-secreting innate lymphoid cells, resulting in CCR1-dependent recruitment to the intestinal mucosa of Toxoplasma infected mice [35]. Taking these data and ours collectively, we propose that CXCR3, CCR2 and CCR1 act together as a control axis of innate and acquired immunity in intestinal immunity, ensuring coordinated recruitment of inflammatory monocytes and Th1 effectors to inflamed tissues.
It was recently demonstrated that NK cell-derived IFN-γ controls the differentiation of circulating monocytes into inflammatory dendritic cells during i. p. T. gondii infection, and is thus required for an optimal IL-12 response [36]. In our model we did not see a dependence on CXCR3 for NK cell recruitment, as NK cells appeared to lose CXCR3-GFP expression over the course of infection (Fig. S1), and production of NK cell IFN-γ was equivalent in WT and KO mice (Fig. S5D–F). This difference may be attributable to the alternative routes of infection used, as our model incorporates the intestinal response, while the i.p. route bypasses the intestinal mucosa. Consistent with this idea, absence of CXCR3 did not affect the ability of mice to survive i. p. infection (Fig. S3A).
While our study focuses on the early response to Toxoplasma infection in the intestinal mucosa, others have examined the role of CXCR3 and its ligands in additional tissues and at different stages of infection. Antibody-mediated depletion of CXCL10, a major CXCR3 ligand, increases susceptibility and blocks influx and expansion of T cells in the liver and spleen that accompanies T. gondii infection [37]. Additionally, a study of ocular toxoplasmosis revealed that T cells infiltrating the eye during infection express CXCR3 and produce IFN-γ. Depletion of CXCL10 in this model reduced the number of infiltrating T lymphocytes during chronic infection, resulting in increased parasite replication and ocular damage [38]. Recently, CXCL10 was shown to impact CD8+ T-cell mobility in the brain of chronically infected mice, enhancing their ability to control the parasite by increasing contact with infected cells [39]. Our results for the first time highlight the importance of CXCR3 and its impact on CCR2-dependent monocytes in the initial protective response to the parasite in the intestine.
It has been shown that TGF-β production by intestinal IEL protects against T. gondii-induced damage by down-modulating inflammation [40]. While we did not examine intraepithelial lymphocytes (IEL) in this study, it is possible that CXCR3 expression could also affect trafficking and function of this cell type, thereby contributing to resistance in this model. Future studies will allow us to determine whether CXCR3-expressing IEL play a role in immunity during intestinal T. gondii infection.
CXCR3 has also been assessed for its role in immunity to other protozoan pathogens, including Leishmania and Plasmodium. Interestingly, the function of CXCR3 differs depending upon the parasite, the route of infection and the site examined. For example, Cxcr3−/− mice exhibit impaired IFN-γ production and increased lesion development during cutaneous L. major infection, but the knockout mice are not more susceptible to hepatic L. donovani infection [19], [41]. Furthermore, CXCR3 and its chemokines promote cerebral inflammation and mortality during experimental malaria infection [20], [42]. Overall, CXCR3 and its chemokine ligands function as double-edged swords, inasmuch as they make an important contribution to protective immunity, but when dysregulated they are the cause of deleterious immunopathology.
In addition to its role in cell recruitment, CXCR3 has recently been suggested to be important for priming CD4+ T cells in the lymph node to become Th1 cells by promoting long-lasting interactions between T cells and CXCL10-expressing dendritic cells. In the absence of CXCR3, T cells fail to fully differentiate into IFN-γ-producing cells and are defective during subsequent lymphocytic choriomeningitis virus (LCMV) infection [43]. We found no evidence for defective Th1 responses in secondary lymphoid organs during Toxoplasma infection of Cxcr3−/− mice, a result that is supported by similar findings during L. major infection [19]. However, our results are consistent with impaired recruitment of Th1 cells in the absence of CXCR3, as lamina propria CD4+ T cells from Cxcr3−/− mice exhibited defective IFN-γ production. We conclude that the function of CXCR3 in promoting Th1 differentiation versus, or in addition to, homing to inflammatory sites is likely to be a context-dependent phenomenon.
Although we found no differences in T-cell activation in CXCR3 negative T cells of reporter mice or in CXCR3 KO animals, as measured by expression of CD25, CD69 and CD44 (data not shown), we observed a consistent decrease in CD27 expression amongst CXCR3-negative T cells of reporter mice. Overall levels of CD27 expression were also lower in T cells from Cxcr3−/− mice (data not shown). CD27 is a member of the TNF receptor super family that has been implicated as a T-cell costimulatory molecule [44]. CD27 expression is thought to characterize naïve or memory T cells, whereas loss of CD27 represents terminal differentiation [45]. Accordingly, it is possible that in addition to controlling T-cell recruitment to sites of infection, Cxcr3−/− T cells may undergo terminal differentiation and, possibly, premature Th1 effector death in tissues prior to mediating IFN-γ dependent inflammatory monocyte activation. This is consistent with studies showing that CD27low cells are more susceptible to apoptosis and that accumulation of influenza-specific T lymphocytes was impaired in the lungs of Cd27−/− mice during infection [44], [46]. In our study, CD27 expression was not rescued by adoptive transfer of WT CD4+ T cells (data not shown), which may suggest that increasing the T-cell pool allows the system to cross a certain threshold of T-cell levels in order to activate inflammatory monocytes without reversing CD27 expression.
In the absence of CXCR3, we found that lamina propria neutrophil levels were increased during infection, as was their activation status as measured by TNF-α expression. Furthermore, adoptive transfer of WT CD4+ T lymphocytes into the KO strain reversed increased levels of PMN as well as their production of TNF-α. Because this inflammatory cytokine has been linked to intestinal damage during Toxoplasma infection [47], it seems likely that CXCR3-dependent effects on neutrophils are likely to be secondary to loss of inflammatory monocyte function. In this scenario, defects in monocyte-mediated parasite killing would result in damage to the intestinal mucosa. Translocation of luminal gut flora, known to contribute to emergence of parasite-induced intestinal lesions [48]–[50], would in turn be expected to result in local neutrophil recruitment. Indeed, based on neutrophil depletion studies it has been suggested that these cells mediate damage to the intestinal mucosa during T. gondii infection [27]. The cellular and molecular basis for this effect is not at present known, but both the IL-17/IL-23 axis and CXCL8 have been shown to promote neutrophil accumulation in infected tissues suggesting involvement of one or both of these mediators [51]–[53].
The results of this study extend our understanding of immunity in the intestinal mucosa, which has become increasingly important as inflammatory bowel diseases (IBD), such as ulcerative colitis and Crohn's disease, become more common in developed regions of the world. In this regard, abnormally high levels of CXCR3 are associated with dysregulated intestinal responses in human IBD patients, underscoring the potential hazards of unbalanced inflammatory responses [54]–[56]. While CXCR3 can be pathogenic by recruiting effector cells to otherwise healthy tissue, as in IBD or cerebral malaria, we show here that CXCR3-expressing T cells play an essential protective role in host defense by enabling defense against pathogenic organisms. Antibodies against CXCL10 have been suggested as potential therapeutic agents against IBD [57], [58]. The results from this study, however, highlight the possible harm of inhibiting CXCR3-expressing cells into sites of inflammation during infection with an enteric pathogen.
As the first study to demonstrate a protective role for CXCR3+CD4+ T cells in the intestinal immune response, we have shown here that failure to appropriately recruit these T cells results in impaired inflammatory monocyte activation, accumulation of intestinal parasites, and subsequent recruitment of potentially pathogenic TNF-α-secreting neutrophils. Our results reveal CXCR3 as a critical chemokine receptor of the adaptive immune system that ensures appropriate placement of T cells in inflamed tissue, enabling inflammatory monocytes of innate immunity to acquire effector functions and mediate effective host defense.
All experiments in this study were performed strictly according to the recommendations of the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocols were approved by the Institutional Animal Care and Use Committee at Cornell University (permit number 1995–0057). All efforts were made to minimize animal suffering during the course of these studies.
Female Swiss Webster mice (6–8 weeks of age) were purchased from the Jackson Laboratory (Bar Harbor, ME). Cxcr3−/− and Cxcr3 eGFP knock-in reporter mice [21] were established as breeding colonies in the Transgenic Mouse Facility at the Cornell University College of Veterinary Medicine. The CXCR3 internal ribosomal entry site bicistronic eGFP reporter (CIBER) mice were generated as described [21]. This strain possesses a functional CXCR3 receptor, and all CXCR3 positive cells also express intracellular eGFP. Mouse infections were initiated by oral inoculation of cysts of the type II T. gondii ME49 strain. Cysts were isolated from chronically infected Swiss Webster mice by homogenization of whole brain in sterile PBS. Unless stated otherwise, mice were infected at 8–12 weeks of age with 30 cysts.
Intestines were excised, flushed with 10% neutral-buffered formaldehyde, and embedded in paraffin for sectioning. Sections were stained for hematoxylin and eosin for assessment of pathological changes. Sections were also stained for parasite antigen by immunohistochemistry at the Cornell Animal Health Diagnostic Center. Frozen sections were obtained by embedding 1 cm lengths of intestine in OCT. Sections of 6–8 µm were cut on a cryostat, fixed in ice-cold acetone, and blocked with PBS containing 2× casein and goat serum. To examine T-cell infiltration, sections were incubated with rat anti-CD4 antibody (GK1.5) (ATCC, Manassas, VA) or rat IgG at 4°C overnight followed by goat anti-rat Alexa-647 secondary antibody (Life Technologies, Grand Island, NY). Sections were then mounted in DAPI-Prolong antifade (Life Technologies) and imaged by confocal microscopy. ImageJ software was used to analyze fluorescence of independent channels.
Spleens and mesenteric lymph nodes were excised, crushed between sterile slides, and passed through a 70-µm filter (BD, Franklin Lakes, NJ). Red blood cells from splenocyte suspensions were lysed with ACK lysis buffer (Life Technologies). For LP leukocyte isolation, the small intestine was removed, cleaned of mesentery, flushed with sterile PBS, and cleared of Peyer's patches. The intestine was opened longitudinally and the mucosal layer was scraped with a blunt scalpel to remove epithelial cells. The tissue was cut into 5 mm sections and vigorously washed with Dulbecco's modified Eagle's media (Cellgro, Manassas, VA) and 5 mM EDTA (Life Technologies). Cells were liberated from the intestinal tissue by digestion with 10 mg/ml collagenase (Sigma, St. Louis, MO) at 37°C and subsequently passed through a 70-µm filter.
Secretion of IFN-γα, TNF-α, and IL-10 was assayed by ELISA (eBioscience, San Diego, CA) following manufacturer's instructions in the presence of soluble tachyzoite antigen (STAg) prepared as previously described [59]. IL-12p40 was quantitated using an in-house ELISA [60]. For ileum biopsy cultures, 1 cm intestinal sections were flushed with PBS, opened longitudinally, and cultured overnight in complete Dulbecco's modified Eagle's media supplemented with 10% bovine growth serum (Hyclone), 0.05 mM β-mercaptoethanol (Sigma), 1 mM sodium pyruvate, 0.1 mM nonessential amino acids, 10,000 U/ml penicillin, 10,000 µg/ml streptomycin, and 30 mM HEPES (reagents from Life Technologies). Supernatants were collected and assayed for cytokine by ELISA.
RNA was isolated from MLN and intestinal tissue from mice over a time course of infection. Tissue was initially disrupted with a tissue homogenizer and subjected to RNA isolation following manufacturer's instructions (Tissue RNA Kit, Omega Biotek, Norcross, GA). RNA was converted to cDNA (Quantas Biosciences, Gaithersburg, MD) and assayed for gene expression by SYBR green technology (Quanta Biosciences). Primers were designed to span exons by Integrated DNA Technologies. GAPDH was used as a housekeeping gene. Gene expression from each timepoint was normalized to uninfected control samples.
Single cell suspensions were pelleted and resuspended with primary antibodies (BioLegend, San Diego, CA: anti-CD4 PerCP, anti-CD8α APC-Cy7, anti-CD45 Alexa-488, anti-CD11b APC-Cy7, anti-CD11b APC, anti-Ly6G FITC; eBioscience, San Diego, CA: anti-CD25 PE, anti-CD45 APC or FITC, anti-CD69 PE, anti-Ly6C/G APC; BD Biosciences, San Jose, CA: anti-Ly6G PE-Cy7, anti-Ly6C V420, anti-CD45 PE, anti-Ly6C/G PerCP, anti-CD44 APC) in ice-cold FACS buffer (1% bovine serum albumin/0.01% NaN3 in PBS) for 30 min. For IFN-γ staining, cells were incubated for 6 hrs with Brefeldin-A (eBioscience; 10 ug/ml), PMA (Sigma; 10 ng/ml), and ionomycin (Sigma; 1 ug/ml), then fixed with 4% paraformaldehyde and subsequently incubated with primary antibodies resuspended in the FoxP3/transcription factor buffer staining set (eBioscience). For IL-12 and TNF-α staining, cells were incubated for 6 hrs with Brefeldin-A only (eBioscience). Intracellular staining experiments used 106 cells. Antibodies used for intracellular staining included anti-IFN-γ PE-Cy7, anti-TNF-α PE-Cy7 (Biolegend); anti-IL-12 PE, anti-TNF-α APC (BD Biosciences), and anti-Toxoplasma p30 (Argene, Shirley, NY). Cell fluorescence was measured using a FACS Canto (BD Biosciences). Data was analyzed using FlowJo software (FlowJo, Ashland, OR).
DNA was isolated from whole intestinal tissue using a Tissue DNA kit following manufacturer's instructions (Omega Biotek, Norcross, GA). DNA was amplified by quantitative PCR as described previously using primers against the T. gondii B1 gene and the host argininosuccinate lyase (ASL) gene [61]. Ten-fold serial dilutions of genome copy standard curves were created using known quantities of host (splenocytes) and parasite (tachyzoites) cells based on DNA quantity, Avogadro's number, and genome size. To quantify parasite burden, the generated values for host and parasite genome copies from the DNA preparations were expressed as a ratio of parasite (B1) to host (ASL) genomes.
Splenocytes from naïve mice were harvested and subjected to CD4 positive selection by magnetic bead sorting following manufacturer's instructions (Stem Cell Technologies, Vancouver, British Columbia). Cells were purified to ∼90–95% purity and were transferred by intravenous retro-orbital injection into Cxcr3−/− recipients at 5×106 CD4+ cells per mouse. Twenty-four hours post transfer, mice were challenged with 30 cysts of the T. gondii ME49 strain. In some experiments mice were left to assess survival following cell transfer. In other experiments intestinal tissue was collected at day 9 post-infection for intracellular cytokine analysis.
Swiss rolls of the intestines were histopathologically scored by an investigator that was double-blinded to sample identity. Intestines were scored on an ascending 0–4 scale as previously described [50], [62]. Briefly, scores of 0 were normal, scores of 1 indicated mild focal lesions, scores of 2 indicated moderate focal lesions, scores of 3 indicated moderate multifocal lesions, and scores of 4 indicated severe multifocal lesions. Histopathological features scored included: inflammation of the intestinal submucosa (lamina propria), inflammation extending throughout all histological layers of the intestine (transmural inflammation), sloughing of intestinal epithelium, intestinal villus fusion and blunting, and necrosis of villi. Sections of complete small intestines from 13 KO and 14 WT infected mice were scored. The data are plotted as mean scores of individual mice.
Differences between groups were analyzed using student's t-test. Differences between 3 or more groups were analyzed using one-way Anova with Newman-Keuls post-test. P-values less than 0.05 are considered significant and are designated by * p<0.05, ** p<0.01, or *** p<0.001. Pathology scores were analyzed using the Mann-Whitney t-test.
|
10.1371/journal.pntd.0005748 | Lymphatic filariasis patient identification in a large urban area of Tanzania: An application of a community-led mHealth system | Lymphatic filariasis (LF) is best known for the disabling and disfiguring clinical conditions that infected patients can develop; providing care for these individuals is a major goal of the Global Programme to Eliminate LF. Methods of locating these patients, knowing their true number and thus providing care for them, remains a challenge for national medical systems, particularly when the endemic zone is a large urban area.
A health community-led door-to-door survey approach using the SMS reporting tool MeasureSMS-Morbidity was used to rapidly collate and monitor data on LF patients in real-time (location, sex, age, clinical condition) in Dar es Salaam, Tanzania. Each stage of the phased study carried out in the three urban districts of city consisted of a training period, a patient identification and reporting period, and a data verification period, with refinements to the system being made after each phase. A total of 6889 patients were reported (133.6 per 100,000 population), of which 4169 were reported to have hydrocoele (80.9 per 100,000), 2251 lymphoedema-elephantiasis (LE) (43.7 per 100,000) and 469 with both conditions (9.1 per 100,000). Kinondoni had the highest number of reported patients in absolute terms (2846, 138.9 per 100,000), followed by Temeke (2550, 157.3 per 100,000) and Ilala (1493, 100.5 per 100,000). The number of hydrocoele patients was almost twice that of LE in all three districts. Severe LE patients accounted for approximately a quarter (26.9%) of those reported, with the number of acute attacks increasing with reported LE severity (1.34 in mild cases, 1.78 in moderate cases, 2.52 in severe). Verification checks supported these findings.
This system of identifying, recording and mapping patients affected by LF greatly assists in planning, locating and prioritising, as well as initiating, appropriate morbidity management and disability prevention (MMDP) activities. The approach is a feasible framework that could be used in other large urban environments in the LF endemic areas.
| Lymphatic filariasis (LF) can cause disabling conditions in infected patients including lymphoedema-elephantiasis (LE) and hydrocoele. Identifying the number and locations of these patients is the first step towards ensuring that these patients receive the care they require, however there is currently no standardised approach for this essential action. This paper presents a health community-led approach for rapidly identifying patients in urban areas using an SMS reporting system, MeasureSMS-Morbidity, that allows health workers to report individual-level patient information (age, sex, location, condition, severity), which can be then be viewed in real-time via a web browser. The quality of the data can be easily monitored during the data collection period, and there is instant availability of patient information. This system is used here in the large urban centre of Dar es Salaam, Tanzania. A total of 6889 patients were identified, equating to 80.9 hydrocele patients per 100,000 population, 43.7 LE patients per 100,000 people, and 9.1 patients with both conditions. This information is now enabling the national neglected tropical disease (NTD) program to provide the essential care facilities and training for LF healthcare in locations in the city where it is most needed.
| Lymphatic filariasis (LF) is a neglected tropical disease (NTD) that can have a devastating impact on affected individuals, with clinical symptoms such as acute dermatolymphangioadenitis (ADLA, “acute attacks”), lymphoedema and elephantiasis, and hydrocoele, causing physical, mental and economic distress [1–6]. In recognition of this, the World Health Organization’s (WHO) Global Programme to Eliminate Lymphatic Filariasis (GPELF) requires that countries wishing to be recognised as having eliminated LF are required not only to prove that disease transmission has been interrupted through mass drug administration (MDA), but that they are also alleviating the suffering of those affected by providing a minimum package of care to each person with lymphoedema/elephantiasis (LE) and hydrocoele in LF endemic areas [7]. This package includes (i) surgery for hydrocoele [8,9], (ii) support for episodes of ADLA [10,11], and (iii) management of LE to prevent disease progression. Countries wishing to complete the WHO dossier for certification of elimination as a public health problem are therefore required to firstly estimate the number of patients in endemic areas at the implementation unit (IU) level, and to provide information on the number of facilities able to provide the necessary care to these identified patients. Finally, an assessment of the readiness and quality of the care being provided at these services is also required [12,13].
Patient numbers at the IU level will enable national LF elimination programmes to appropriately forecast, plan and manage patient care, and to meet the requirements of the WHO dossier for programme success. At present, despite great progress towards interrupting transmission of disease with 63 of the 73 endemic countries having initiated MDA, only 18 of these are reported to monitor morbidity management and disability prevention (MMDP) at this geographical level, i.e. have identified the number of IUs with known cases, or the IUs where MMDP services are provided [12]. There are currently no specific guidelines on the methods for obtaining patient estimates, although an MMDP toolkit is currently under development by WHO to provide additional guidance for this. Current documented suggestions include collecting LE and hydrocoele information during LF baseline prevalence surveys, when enumerating households during MDA, or conducting separate surveys either independently or in collaboration with other organizations concerned with similar disabilities and their care [14]. Many countries are known to collate morbidity information whilst distributing MDA, however the quality of this information has been shown to be very variable, with national programmes lacking the necessary resources to validate their data [15,16]. Examples of bespoke patient enumeration surveys can also be found in the literature, however these tend to be on a small scale and are labour intensive and as such may be difficult to scale up to the required geographical level [16,17].
Dar es Salaam is a large and densely populated region on the coast of Tanzania which at the time of the survey (2015) comprised of three districts: Temeke, Kinondoni and Ilala. In this region LF is caused by the Wuchereria bancrofti parasite, transmitted by the Culex mosquito. Previous estimates of LF prevalence in Dar es Salaam measured using immunochromatographic tests (ICTs) includes 9.9% (Temeke = 13.2%, Kinondoni = 14.4%, Ilala = 4.7%), by Mwingira et al. (2017) [18], and 3.0% by Mwakitalu et al. (2013) [19]. Four rounds of MDA have been completed in the Dar es Salaam region since 2013, with varying therapeutic coverages due to the challenges associated with administering treatment within a large dynamic urban population. While it was anticipated that there was likely to be a significant number of LF clinical cases in Dar es Salaam, there was limited data on the scale of the problem. As such, there was a relative lack of evidence upon which to plan a suitable MMDP strategy [20].
The primary aim of this study was to improve our knowledge of the overall LF morbidity burden in Dar es Salaam, and further to gain an understanding of the geographical distribution of cases to guide the delivery of MMDP services to where they are most needed. Due to the association between the prevalence of clinical cases and of infection, knowing more about the locations of cases may also allow programmes to identify areas where MDA should be more focussed [21]. To undertake this task at this geographical scale in a time-effective manner, we implemented the mHealth short message service (SMS) reporting system, ‘MeasureSMS-Morbidity’, which enables the rapid collection, collation and dissemination of estimated LF patient numbers [22,23]. Prior to this study, MeasureSMS-Morbidity had been implemented in rural areas of Malawi and Ghana, but had yet to be trialled in a densely populated urban area [22]. The system relies upon a network of local health workers to each collect individual-level patient data over small geographical areas which they then submit via SMS using their own mobile phones. These data are then automatically collated into an online database which can be accessed and interrogated using a web browser. Mobile technology-driven, community-led approaches to conducting surveys such as this have proven to increase time efficiency, and consequently cost-efficiency, of data collection whilst further increasing the sense of local data ownership and accountability [23].
LF patient identification and reporting activities are part of routine programme activities conducted by the Ministry of Health and Social Welfare, Tanzania, and as such ethical clearance in Tanzania was not required, and a waiver was granted. Oral consent to record and report details relating to their condition was obtained from all identified LF patients, however this was not documented. Reported and verified patients were verbally informed of the purpose of the activity, and all resulting data were analysed anonymously. Data reporters did not record any information on patients who refused to disclose any information, and those who did not wish to be examined by the clinical officer were excluded from the verification survey. Ethical clearance was obtained from the Liverpool School of Tropical Medicine Research Ethics Committee (Research Protocol 12.22).
In order to achieve the primary aim of this study i.e. to assess the LF morbidity burden in Dar es Salaam, a patient identification survey was undertaken. This survey was undertaken in three phases between March and August 2015, covering each of Dar es Salaam’s three districts i.e. Temeke, Kinondoni and Ilala. To undertake this task efficiently with respect to both time and resources, a health community-led door-to-door survey approach was used and the MeasureSMS-Morbidity reporting tool was incorporated to rapidly collate and monitor the data. Detailed information on the MeasureSMS-Morbidity system can be found elsewhere [22–24]. In brief, this system allows basic information on identified patients such as their location, sex, and age, plus information on their clinical condition (LE or hydrocoele), severity of LE (mild, moderate, severe), and number of acute attacks experienced in the last 6 months, to be recorded by the health community staff and reported via SMS using their own phones. This information can then be viewed in real-time via a web browser. Each individual SMS report also generates a SMS response indicating either that the message had been sent in the correct format, or that there are reporting errors that required correcting. The severity of LE was classified as either mild = slight swelling, moderate = enlarged limb with shallow folds, and severe = greatly enlarges limb with deep folds as previously described [22]. These categories correspond to a Dreyer staging of 1–2 (mild), 3–4 (moderate) and 5+ (severe).
Prior to this current study, the MeasureSMS-Morbidity system had been implemented in rural settings only, and this was the first time at which the system had been implemented at such a large scale, and in an urban area. To facilitate this scale up, the implementation process was revised and refined after each phase. Each phase consisted of a training period, a patient identification and reporting period and a data verification period. This verification period was incorporated into the study to address the secondary objective of assessing data quality. Specifically, the verification period enabled the positive predictive value (PPV) of the identified patients to be estimated i.e. the proportion of patients identified during the survey who were confirmed to have lymphoedema or hydrocole. Phase 1 (Temeke) was initiated in March 2015, and Phases 2 (Kinondoni) and 3 (Ilala) was commenced in July and August 2015 respectively. The initial process and the subsequent refinements are described below.
Table 1 presents the number of patients with each condition reported by SMS. In absolute terms, Kinondoni had the highest number of reported patients (2846) comprised of 950 LE only patients, 1654 hydrocoele only patients, and 242 with both conditions. This was followed by Temeke (2550 total, 807 LE, 1560 hydrocoele, 183 with both), then Ilala (1493 total, 494 LE, 955 hydrocoele, 44 with both).
Prevalence estimates (per total population and per male population for hydrocoele) were calculated using the 2015 population estimates as the denominator, which were extrapolated from the 2012 census using annual growth rates of 5.8%, 4.9% and 6.5% for Temeke, Kinondoni and Ilala respectively, i.e. population estimates of 1,621,148, 2,048,976 and 1,485,308, totalling 5,155,432, with male population estimates of 707,861, 902,981 and 634,663 totalling 2,245,505 [25]. The overall reported morbidity prevalence for Dar es Salaam was 133.6 per 100,000 total population, with the highest prevalence seen in Temeke (157.3 per 100,000) and the lowest in Ilala (100.5 per 100,000). This pattern was consistent for both conditions, with LE only prevalence per 100,000 population ranging from 33.3 in Ilala to 49.8 in Temeke, and hydrocoele only prevalence per 100,000 males ranging from 150.3 in Ilala to 220.4 in Temeke. The ratio of LE to hydrocoele patients was consistent across all three districts, with the number of hydrocoele patients being almost twice that of LE cases.
Table 2 presents summaries of the reported patients by age and sex. District-level population data was obtained from the National Bureau of Statistics [26] for 2012, and the totals were projected to represent 2015 as described above. The prevalence of LE was approximately equal between males and females in all three districts (Temeke: 64.84 and 71.72 per 100,000 population; Kinondoni: 67.89 and 60.37; Ilala 39.86 and 42.84 for males and females, respectively). A positive relationship was observed between age and LE prevalence in all three districts and both sexes, with the highest prevalence being observed in the oldest age group in Temeke (>74 years, 708.44 patients per 100,000 population), and in the 60–74 age group in Kinondoni (474.30 patients per 100,000 population) and Ilala (353.16 per 100,000 population). For hydrocoele, a similar trend in age was observed (1,075.4 patients per 100,000 population aged >74 in Temeke, 727.8 patients per 100,000 population aged 60–74 in Kinondoni and 614.2 patients per 100,000 population aged 60–74 in Ilala).
Table 3 presents summaries of severity of reported LE (mild, moderate, severe) by district, including the mean and standard deviation of the number of reported number of acute attacks experienced in the previous six months. Overall, severe LE patients account for approximately a quarter of those reported (26.9% overall ranging from 25.0% in Temeke to 31.0% in Ilala), with the number of acute attacks increasing with reported severity (1.34 in mild cases, 1.78 in moderate cases, 2.52 in severe). The variability in the number of reported attacks also increases with severity (1.33 in mild cases, 1.62 in moderate, 1.84 in severe). Similar trends are seen in each of the three districts.
Maps of prevalence by ward level, using extrapolated 2015 population estimates as the denominator (http://ihi.eprints.org/2168/1/Village_Statistics.pdf) are presented in Fig 2. These figures highlight the high morbidity prevalence in the southern peri-urban wards of Temeke and the northern peri-urban wards of Kinondoni.
The MeasureSMS-Morbidity system stores every SMS received and automatically checks the SMS for formatting errors. Table 4 summarises the SMS acceptance rates (% of SMS without any formatting errors) for each of the three patient identification phases. These error rates reflect two processes. Firstly, they give an indication of the data reporters’ ability to report data using the MeasureSMS-Morbidity system, which may in turn reflect improvements in the training process. Secondly, they reflect the effectiveness of the data supervision process, as the incoming data were reviewed each day by the supervision team and data reporters were contacted when repeated errors or data inconsistencies were identified. During Phase 1 (Temeke), data supervision was largely undertaken by the donors whereas phases 2 (Kinondoni) and 3 (Ilala) was managed by the Ministry of Health staff, with support from the donor. In comparing the acceptance rates between these three period, no significant differenced were observed (p = 0.3788). However once duplicates were removed, significant differences were observed (p<0.001), with a greater proportion of accepted messages being kept in phase 1 (Temeke = 98.5%) in comparison to phases 2 and 3 (Kinondoni = 94.7 and Ilala = 94.0%). Fig 3 presents the acceptance rate by reporting day for each of the three districts. We again observe slightly higher acceptance rates by day in Phase 1 (Temeke). Table 4 indicates that in each district, data were reported over a period of 12–15 days, with most data (greater than 80%) being sent in the first 7 days of the reporting period. We therefore hypothesise that the decline in acceptance rates towards the end of the patient identification period seen in Fig 3 is likely due to there only being a small number of reports being submitted during this time, with these data reporters being those who had difficulties using the system.
The length of the verification surveys varied with each phase. Due to adverse flooding conditions experienced in March and May 2015, the simple (stratified) random sampling approach used, and the difficulties in locating selected patients as described in the methods section, the verification process in Temeke lasted over 13 months and was completed in May 2016. Over this period, 38 patients were randomly selected for verification and visited by the verification team. No issues were observed between paper forms and data sent via SMS when paper forms were reviewed as the initial part of the verification. When comparing the written records collected during this verification visit with the initial data reported during the patient identification survey it was possible to match 36 (15 reported LE, 21 reported hydrocoele) of the 38 patients by patient ID and gender. Of these 36, the main difference observed were in the initially reported age and the age reported during verification, with a median difference of two years. As inconsistencies in age recall is common in these settings, differences of up to 10–15 years were expected. Four matched patients in Temeke had an age difference of greater than 15 years.
Adaptations made to the data collected during patient identification i.e. the inclusion of the patient’s phone number, plus the refinements made to the sampling process resulted in the verification for Kinondoni and Ilala being completed much more efficiently. There were however still some delays due to the transience of the population in Dar es Salaam as well as some restructuring of the city resulting in large areas of houses being demolished between the time of patient identification and data verification period. In both Kinondoni and Ilala, 115 patients, reported by 16 data reporters in each district, were verified over a 6 month period. Of the 115 patients in Kinondoni it was not possible to match 17 patients with the reported data, hence only 98 verified patients (24 reported LE, 71 reported hydrocoele, 3 reported with both conditions) were included in the final dataset. The median difference in reported age for these 98 patients was two years, with 10 patients having age differences of greater than 15 years. In Ilala, only one patient could not be matched, hence 114 patients were included (30 reported LE, 75 reported hydrocoele, 9 reported with both conditions). The median difference in reported age of these patients was three years, with 12 patients having age differences greater than 15 years.
Table 5 presents the verification results by district. Of the 81 patients reported to have either LE or both conditions, 63 (77.8%) were diagnosed by the verification team as having LE, ranging from 71.8% (28/39) in Ilala to 86.7% (13/15) in Temeke. Of the 179 patients reported to have hydrocoele or both conditions, 165 (92.2%) were diagnosed as having hydrocoele by the verification team, ranging from 88.1% (74/84) in Ilala to 95.9% (71/74) in Kinondoni. Similar results were observed after removing the 26 patients whose reported age differed by more than 15 years between the two data sources (S1 Table).
Of the 18 people misdiagnosed with LE, 14 were confirmed to have hydrocoele, whereas 4 had other conditions. Of the 14 people misdiagnosed with hydrocoele, 5 had LE, 1 had a hernia, 2 had previously had hydrocoele but had been operated on, 5 had other conditions and 1 did not have any discernible illness. It is also worth noting that of the 165 confirmed to have hydrocoele, 27 (16.3%) had a concurrent hernia. Twelve patients with both LE and hydrocoele were included in the verification survey. Only 4 (33.3%) of these were confirmed to have both conditions, 5 had hydrocoele only, 2 had LE only and 1 had a non-LF associated condition.
Of the 63 patients correctly verified as having LE, 60 had reported the severity of the condition (11 mild, 26 moderate, 23 severe). In comparing these with the Dreyer score determined for each patient during the verification process (stage 1–2 = mild, stage 3–4 = moderate, stage 5+ = severe), 3 (27.3%) were correctly verified as mild, 16 (61.5%) were correctly identified as moderate and 6 (26.1%) were correctly identified as severe (Table 6). Overall, there was low agreement between the severity assigned by the patient identifiers and the verification team (Cohen’s kappa = 0.083).
With the prevalence of LF cases previously unrecorded, these results indicate a much higher burden of LE and hydrocoele in Dar es Salaam than anticipated for an urban centre, with 2251 patients reported to have LE, 4169 patients reported to have hydrocoele plus a further 469 patients having both conditions. Whilst it is recognised that there may be some false positives in these reports, and further that some patients may have been missed from the survey, the verification survey results give the national LF elimination programme confidence that these numbers are representative of the true burden. Further to these absolute numbers, the national programme now also has additional information on the geographical distribution of disease, with the district of Temeke being the most affected of Dar es Salaam’s three districts with the larger, less densely populated wards reporting in excess of 500 cases per 100,000 people.
This information is crucial in determining the resources needed to manage the conditions affecting these patients, and will assist in selection in locations within the urban area within which care services could be most effectively provided. As with many endemic countries, strategic partnerships have been developed in Tanzania between the national LF programme and international donors to address the immense burden of LF clinical disease within the country, leading to the development of plans to scale up hydrocele surgery throughout 2016. Further, plans to engage CHVs to provide home-based LE training and care for patients and their caregivers have already been developed with the region’s NTD and home-based care teams. These services are intended to be supplementary to the specialist filariasis clinic which has been operated by the national programme in Dar es Salaam since 2000 [20].
The data obtained in this study will also, in addition to guiding morbidity management activities, provide an insight into disease transmission within Dar es Salaam which is likely to aid the management of future MDA campaigns, or perhaps the development of alternative transmission breaking strategies such as vector control through improved drainage and Culex mosquito population reduction e.g. using polystyrene beads [27,28]. Difficulties in conducting MDA within an urban environment often arise due to the factors such as high population movement, and the resulting low compliance can result in MDA coverage levels that are inadequate to permanently break transmission. It is also not clear to what extent the distribution of patients reflects the transmission patterns in such a mobile urban population. However, the additional disease transmission information provided by patient mapping may be useful in raising awareness across the city in general, targeting potential high risk areas with more intensive community sensitisation campaigns, and/or increasing MDA distribution points to increase coverage [18,19,29,30]. It is likely that future MDAs will benefit from the active and successful morbidity management activities, as it has been shown that LE management programmes can be important for improving increasing MDA coverage [31].
In implementing the two-tiered patient identification and reporting process in an iterative manner, it was possible to refine the process with each phase of implementation. Relocating patients after the initial patient identification survey proved to be the most challenging aspect of this approach, with unavoidable contributing factors including adverse weather conditions and population displacement due to the demolition of many houses. Difficulties also arose due to the patient identifier being unfamiliar with the area to which they had been assigned. Whilst this problem lessened over the course of the exercise, primarily through ensuring the national programme could contact the patient directly by mobile phone, it may be beneficial to consider using locally-based patient identifiers should this exercise be repeated elsewhere. Additionally, by increasing the emphasis on the staging and severity of LE in the training, the accuracy of the reported data could be improved, thus providing more accurate information on the level of services required in any given area.
Despite the implementation challenges, there were many benefits to the patient identification and the reporting process adopted for this exercise. Notably, the ability to view and assess the quality of the patient identification data in real time using the MeasureSMS-Morbidity tool was a great asset. Enabling information on the morbidity burden to be known almost instantaneously, as opposed to having to wait a prolonged period for collation and digitization of the paper forms, also facilitates the provision of needed care to patients much more efficiently [20]. By the end of the study, the responsibility of monitoring the real-time data and ensuring its quality was led by the national programme team themselves, thus empowering the local teams to manage and control their own success. This approach of promoting the national programme to take full ownership of the data and the implementation of the process had the effect of building valuable data management and health surveillance capacity within the local team [32,33].
This exercise highlights how crucial patient estimates are for the provision of much needed care to LF patients and the information obtained greatly assists in planning, locating and prioritising, as well and initiating, appropriate MMDP activities. The approach to patient identification and reporting presented in this paper provide a feasible framework that could be adopted in other large urban environments [24], and thereby enable these endemic areas to achieve the MMDP components of the GPELF [12], or could be adapted to address patient identification issues for other diseases. The focus should now be to develop and implement strategies to meet the needs of the patients identified during this exercise, and thus ensuring no patient is left behind.
|
10.1371/journal.ppat.1002064 | Endemic Dengue Associated with the Co-Circulation of Multiple Viral Lineages and Localized Density-Dependent Transmission | Dengue is one of the most important infectious diseases of humans and has spread throughout much of the tropical and subtropical world. Despite this widespread dispersal, the determinants of dengue transmission in endemic populations are not well understood, although essential for virus control. To address this issue we performed a phylogeographic analysis of 751 complete genome sequences of dengue 1 virus (DENV-1) sampled from both rural (Dong Thap) and urban (Ho Chi Minh City) populations in southern Viet Nam during the period 2003–2008. We show that DENV-1 in Viet Nam exhibits strong spatial clustering, with likely importation from Cambodia on multiple occasions. Notably, multiple lineages of DENV-1 co-circulated in Ho Chi Minh City. That these lineages emerged at approximately the same time and dispersed over similar spatial regions suggests that they are of broadly equivalent fitness. We also observed an important relationship between the density of the human host population and the dispersion rate of dengue, such that DENV-1 tends to move from urban to rural populations, and that densely populated regions within Ho Chi Minh City act as major transmission foci. Despite these fluid dynamics, the dispersion rates of DENV-1 are relatively low, particularly in Ho Chi Minh City where the virus moves less than an average of 20 km/year. These low rates suggest a major role for mosquito-mediated dispersal, such that DENV-1 does not need to move great distances to infect a new host when there are abundant susceptibles, and imply that control measures should be directed toward the most densely populated urban environments.
| Although dengue is a major cause of morbidity in many tropical and subtropical regions of the world, little is known about how the causative virus (dengue virus, DENV) spreads through endemic populations. To address this issue we undertook a phylogeny-based analysis of 751 complete genome sequences of DENV-1 sampled from patients in southern Vietnam during 2003–2008. We show that multiple viral lineages co-circulate within the urban area of Ho Chi Minh City (HCMC), and spread at approximately equivalent rates through overlapping geographical areas, suggesting that they are of equivalent fitness. We also observed that DENV-1 within HCMC tended to disperse from more to less densely populated regions, and that this city was the source population for DENV-1 in the rural area of Dong Thap. Despite the high prevalence of DENV-1 in southern Vietnam, viral dispersion rates were relatively low, especially in HCMC where they averaged less then 20 km/year. Such a low rate is consistent with predominantly mosquito-borne spatial dispersal of DENV-1 in this urban setting containing a large number of susceptibles. Together, these results suggest that dengue control measures such as insecticide spraying should be directed toward the most densely populated regions of localities where the virus is endemic.
| Dengue is the most important mosquito-borne viral disease of humans, annually responsible for approximately 40 million cases and some 20,000 deaths in tropical and subtropical regions [1]. Dengue is caused by one of four single-stranded positive-sense RNA viruses (DENV-1 to DENV-4, also referred to as serotypes) of the genus Flavivirus (family Flaviviridae). Despite the large burden of dengue disease, and considerable research effort, there are currently no licensed vaccines or specific therapies. The challenge of effective and safe dengue vaccination is increased by the possibility that imperfect cross-protective vaccination could enhance DENV infection, or even virulence [2], and that lineages within individual DEN viruses, particularly different ‘genotypes’, may also differ in antigenicity [3]–[6]. In addition, the population dynamics of DENV within individual localities are complex, involving the birth-and-death of viral lineages that may also differ in both virulence and fitness [7]–[13], as well as the intricate patterns of gene flow, at both the local and international scales [7], [14], [15].
DENV transmission among humans is largely caused by the urban adapted and anthropophillic Aedes aegypti mosquito. Spatial and temporal patterns of dengue prevalence are likely driven by multiple factors including the immune status of human hosts [16], their age [17], [18], virus traits [13], [19], [20], the mosquito vector, and environmental variables including aspects of climate such as levels of precipitation [21], [22]. Human movement must also be an important, but poorly understood, contributor to viral transmission dynamics, and is obviously responsible for the increasingly widespread and complex distribution of the four DEN viruses at the global scale. On a local scale, how much DENV transmission within a specific population is due to the local movement of infected human hosts rather than of mosquitoes is unclear. Understanding the spatial and temporal dynamics of dengue transmission in endemic dengue populations is therefore central to the rational deployment of vector control activities and the design of intervention strategies. In this respect it is critical to determine the spatial structure of DENV within endemic populations, the rate at which DENV lineages diffuse through space (particularly in the face of a partially immune population), whether specific lineages are spreading more rapidly than others and indicative of enhanced fitness, and the likely contribution of mosquitoes and humans to local transmission patterns.
To address these questions we employed a fine-scale molecular approach to characterize the virus population dynamics of a recent DENV-1 outbreak in southern Viet Nam, a region of high dengue endemicity. Between 2006–2008 the estimated incidence of DENV-1 infection in the southern twenty provinces of Viet Nam ranged from 86–190 cases/100,000 [13], markedly higher than during the preceding six-year period when it ranged from 1–28 cases/100,000. The causes of this increased incidence are unknown.
To determine the patterns and dynamics of dengue transmission we utilized an expansive data set of DENV-1 whole genome sequences sampled prior to and during the peak in DENV-1 prevalence over a period of six years (2003–2008). We inferred the dynamics of viral transmission within individual communities, between communities, and between neighboring countries, using recently developed Bayesian phylogenetic methods that utilize both the temporal and spatial information of the sampled sequences. Uniquely, these time-calibrated phylogenetic methods provide the ability to reveal the complex interplay of spatial, genetic and epidemiological dynamics at the local, regional and global scales, and have the ability to consider individual viral lineages, whereas epidemiological approaches based on the analysis of incidence data are at best only able to distinguish among the four DEN viruses.
We determined the consensus DENV-1 genome sequence (minimum sequence from nt 70–10,400) in acute plasma samples collected from 751 hospitalized patients in urban Ho Chi Minh City (HCMC) (n = 575 sampled between 2003–2008) and rural Dong Thap Province in the Mekong Delta region (n = 176 sampled between 2006–2007). The majority of viruses were sampled from 2006 to 2008 during which DENV-1 was the most prevalent serotype in circulation (Figure S1).
To determine the evolutionary relationships of DENV-1 in Viet Nam in the context of surrounding countries we analyzed the envelope (E) gene sequences from these locations (Figure 1A). The 751 DENV-1 sequences sampled from Viet Nam fell into one of five clades within the broader Genotype I cluster of viruses [23]. Four of the five clades consistently clustered within the diversity of Cambodian viruses with good support (posterior probability ranging from 0.81 to 1.0). This phylogeographic evidence, coupled with Cambodia and Viet Nam's shared border, is compatible with Cambodia acting as the major source of Vietnamese DENV-1. A caveat to this is the lack of contemporaneous DENV-1 sequences from nearby Thailand, which has previously been shown to harbor substantial DENV diversity and importation into Viet Nam [13]. Clearly, wider sampling in both time and space is needed to test this hypothesis.
The majority of the clades largely comprised viruses from HCMC, with the exception of clade 1, which was found to be Dong Thap dominant. The timing of these inferred introductions were gauged from the age of the most recent common ancestor (TMRCA) of each clade (Table 1). The period in which these different viral clades emerged in southern Viet Nam ranged from late 2001 to mid-2005. Apart from clade 1, which was found to be the most recent introduction, the mean ages of clades 2–5 did not differ significantly, suggesting that different viral lineages were imported over short or similar time-scales, and then co-circulated. These clades were chosen for more detailed phylogeographic analysis. Finally, genome-wide rates of nucleotide substitution – at ∼1×10−3 nucleotide substitutions per site, per year (Table 1) – were the same among clades and highly consistent with those previously determined for DENV [24], [25].
For the clades identified as being within Viet Nam, a discrete spatial model [26] was employed to reveal the migration between the sampling locations. The results are shown in Figure 1B, in which branches are colored by the most probable state location. In four of the five clades HCMC was the most likely viral source, with viruses exported to the rural area of Dong Thap. The non-HCMC isolates in these clades were interspersed among the HCMC sampled isolates, which strongly suggested that the DENV-1 epidemics in southern Viet Nam mainly emerged first in HCMC. The exception was clade 1, which was dominated by Dong Thap viruses and where Dong Thap was inferred to be the most likely place of origin. Moreover, the HCMC viruses in clade 1 did not form a monophyletic group, supporting the view that clade 1 viruses were imported into HCMC on multiple occasions from Dong Thap.
To determine whether the viral migration rates varied between urban and rural epidemics, we compared the spatial dynamics between clades 1 and 4 (Table 2). When focusing on the number of transitions from the inferred source location, a symmetrical pattern was observed between the two clades. For instance, the transmission rate between HCMC and Dong Thap was higher in the HCMC dominant clade 4, while for the reverse direction (Dong Thap to HCMC) it was greater in Dong Thap dominant clade 1. Hence, once a virus became established in a location, rural or urban, the rate of viral exportation was found to be greater than the rate of viral importation.
The geographical coordinates of the patient's residential address in HCMC (n = 381) or Dong Thap Province (n = 175) was known for 556 cases and this information was employed to reconstruct the fine-scaled dispersion of the individual viral lineages within the sampling areas using a continuous spatial diffusion model with non-homogenous dispersion rates [27]. The average viral dispersion rate (km/year) was calculated for each clade, and separately for HCMC and/or Dong Thap data subsets, as if the epidemic in these regions derived from a single introduction (Table 2). We define virus dispersion rate as a measure of how quickly a virus lineage spreads geographically, given the inferred root location and final sampling locations. Even though we only had one estimate of the average dispersion rate of DENV-1 in Dong Thap, a clear disparity was observed when compared to the rates from HCMC lineages (Table 3). Specifically, the viral lineages from clade 1 in Dong Thap spread approximately 2–3 times faster than any lineage from HCMC. This is indicative of a fundamental difference in the epidemiological dynamics of DENV-1 in the two areas.
A further dissection of the dispersion rates through time in HCMC (clades 2, 4 and 5) and Dong Thap (clade 1) revealed interesting patterns in the rate of viral spread in the two locations. In HCMC (Figure 2A and B), the monthly incidence of DENV-1 showed a similar trend as in Dong Thap, with corresponding regular fluctuations and an increasing overall trend. However, there was no clear association between genetic diversity, incidence, and dispersion rate observed in the urban environment demonstrated by the roughly horizontal relationship in Figure 2B and the overlapping 95% HPD (highest posterior density) intervals. Hence, although the DENV-1 clades were introduced independently into HCMC, they had spread at similar and effectively constant rates. For Dong Thap, clade 1 was the only one clearly derived from a distinct single importation and of a sufficient size for analysis. The dispersion rate of DENV-1 appeared to be associated with the fluctuations in genetic diversity and monthly incidence in Dong Thap (Figure 2C and D). The two peaks in relative genetic diversity of clade 1 in Dong Thap coincided with the two major peaks in the monthly incidence, indicating that DENV-1 epidemic in Dong Thap is largely driven by this lineage.
To investigate whether these dispersion rate estimates in HCMC were simply a reflection of the geographic constraint of our samples, they were re-estimated by randomizing the tip location for each clade (Table 4). The results indicated what the maximum dispersion rate could be given the sampled locations, which were found to be 2–3 times greater than the empirical estimates, with wide HPD intervals (Table 4). The spatial reconstruction of the viral spread at different stages of the epidemics showed that these viral lineages had co-circulated in the same place at the same time (Figure 3). This observation is of fundamental importance as it suggests that the number of susceptible hosts to DENV-1 had not been saturated in HCMC, and could potentially have supported additional DENV-1 lineages in this area.
To determine whether transmission routes within HCMC varied according to population density, we employed a non-reversible discrete phylogeography model applied to district level data. Importantly, the more densely populated inner city districts (above 30,000 people per km2) were found to contribute significantly to DENV-1 transmission compared to the suburban districts (Figure 4). Moreover, the most densely populated region, District 5, had the highest number of connections, providing compelling evidence that this area might be a major hub in the city.
At the scale of South-East Asia, the observation that there is a strong clustering by country indicates that there is a far higher level of DENV-1 gene flow within than between countries. Such a phylogeographic pattern is compatible with relatively short transmission distances for DENV as a whole, including that meditated by mosquitoes. This rather limited spatial movement also sits in marked contrast to that observed in respiratory borne pathogens such as influenza, where there is relatively little clustering by place of isolation even on a global scale [28].
Each of the five clades of DENV-1 we identify has a very recent common ancestry, dating only shortly before the appearance of that clade. Given that dengue is endemic in southern Viet Nam, with DENV-1 circulating there for at least 23 years [29], such recent common ancestry suggests that there is a rapid and continual turnover of viral lineages, as has been increasingly described for this and other DEN viruses [8]–[11], [30], [31]. Less clear is whether these instances of lineage turnover are due to fitness differences between the lineages in question, such that natural selection is preferentially able to favor one lineage over another, or whether there is simply a stochastic die-off. That the three major clades we detect in HCMC co-circulate in the same spatial region with overlapping ranges, and possess broadly equivalent levels of relative genetic diversity, suggests that they are of similar fitness and hence that there is little, if any, competition between them. Consistent with this, we did not observe differences in early plasma viremia levels between patients infected with viruses belonging to the different clades (Figure S2). Indeed, we suggest below that HCMC is likely characterized by a large number of susceptible hosts, which would in turn reduce the extent of selective competition between lineages. More generally, these results indicate that although a specific viral serotype may appear to be endemic in a specific geographic region for an extended period, this does not mean that the same viral clades are involved throughout this period.
A striking result from this study is that the ‘virus dispersion rates’ we estimate appear to be very low, and particularly in HCMC where mean rates were universally <20 km/year. Such low rates are especially noteworthy given the rapidity and geographic scale with which DENV-1 re-emerged as the dominant serotype in southern Viet Nam [13]. We therefore interpret these low rates to mean that urban centers like HCMC are characterized by sufficiently high numbers of susceptible hosts such that the virus does not have to move very far to infect a new host. Such a notion is supported by the fact that higher virus dispersion rates are observed in Dong Thap, which is characterized by an approximately ten-fold lower population density (495 persons/km2) relative to HCMC (3024 persons/km2), although more estimates are clearly needed from this locality. In addition, the highest levels of viral movement were found in and out of the most densely populated region of HCMC (District 5), suggesting that this well-connected locality acts as a focal point for dengue dispersion within the city. Hence, it is not that DENV-1 moves slowly at a spatial scale in HCMC, but rather that it does not have to move far geographically to continue its transmission.
Although our sample of genome sequences is biased toward those from HCMC, our analysis indicates that DENV-1 generally diffuses from HCMC to Dong Thap. Again, this observation is suggestive of a gravity model of viral transmission, in which spatial diffusion occurs over a gradient of population density, and is compatible with our observation that dispersion rates are associated with the numbers of susceptible hosts. A similar gravity-dependent pattern of virus dispersion was recently suggested for DENV-2 in Viet Nam [14], although the use of a strictly reversible phylogeographic model in that case meant that directionality could not be ascertained with certainty. Combined, these studies strongly suggest that the density of the human host population plays a fundamental role in determining the transmission dynamics of endemic dengue.
Typically, adult A. aegypti mosquitoes travel short distances of less than ∼100 m during their average life-span of a few weeks [32]–[34]. The very short distances traveled by DENV-1, particularly in HCMC is consistent with mosquitoes, rather than humans, being responsible for the majority of the spatial spread in HCMC, which is again in part a function of the high density of susceptible hosts. A similarly limited movement of dengue has been reported by recent studies that focused on smaller geographic areas, reflecting the restricted spatial range of mosquito vectors, and corroborating the highly focal pattern of DENV transmission observed in HCMC [15], [35]. It is also notable that the geographical range of the three major clades in HCMC changed little from 2003–2008. As such, the full geographic range of these clades is established very early on as the virus is able to spread rapidly through a susceptible host population. Upon the introduction of a new dengue serotype into Iquitos, Peru, it was noted that early-confirmed cases were scattered throughout the city, suggesting a rapid establishment of the virus when entering a completely naïve population [36]. This observation gives added weight to our conclusion that the dispersion rates of DENV-1 in southern Viet Nam are largely a function of the availability of susceptible hosts.
These results have a number of important implications for the future control of dengue. Most generally, that DENV tends to spread relatively slowly on a spatial scale (such that DENV phylogenies exhibit a strong spatial structure both nationally and internationally) suggests that any future vaccine escape or drug resistance mutations would also spread relatively slowly. In addition, that the dispersion rates of DENV appear to largely reflect the density of human host population, including movement from Ho Chi Minh City to Dong Thap, suggests that future control measures, including mosquito spraying, should be directed toward the densest host populations and preferentially to urban over rural areas.
The dengue patients from whom DENV whole genome sequences were determined were enrolled in one of two prospective studies at the Hospital for Tropical Diseases in Ho Chi Minh City, Viet Nam or at Dong Thap Hospital, Dong Thap Province, Viet Nam. The median age of these patients was 12 years (interquartile range 7–17 years) and 51% were male. Serological investigations (IgM and IgG capture ELISAs) were performed using paired plasma samples using methods described previously [37]. DENV serotype and viraemia levels were determined using an internally-controlled real-time RT-PCR assay that has been described previously [38].
Viral genomes were sequenced using the Broad Institute's capillary sequencing (Applied Biosystems) directed amplification viral sequencing pipeline http://www.broadinstitute.org/scientific-community/science/projects/viral-genomics-initiative). This sequencing effort was part of the Broad Institute's Genome Resources in Dengue Consortium (GRID) project. Viral RNA was isolated from diagnostic plasma samples (QIAmp viral RNA mini kit, Qiagen) and the RNA genome reverse transcribed to cDNA with superscript III reverse transcriptase (Invitrogen), random hexamers (Roche) and a specific oligonucleotide targeting the 3′ end of the target genome sequences (nt 10868 to 10890, AGAACCTGTTGATTCAACAGCAC). cDNA was then amplified using a high fidelity DNA polymerase (pfu Ultra II, Stratagene) and a pool of specific primers to produce 14 overlapping amplicons of 1.5 to 2 kb in size for a physical coverage of 2-fold across the target genome (nt 40 to 10649). Amplicons were then sequenced in the forward and reverse direction using primer panels consisting of 96 specific primer pairs, tailed with M13 forward and reverse primer sequence, that produce 500–700 bp amplicons from the target viral genome. Amplicons were then sequenced in the forward and reverse direction using M13 primer. Total coverage delivered post amplification and sequencing was 8-fold. Resulting sequence reads were assembled de novo using the Broad Institute's AV454 assembly algorithm (Henn et al. 2011. in review) and a reference-based annotation algorithm.
All whole genome sequences newly determined here have been deposited in GenBank and assigned accession numbers (Table S1).
A data set of DENV-1 sequences was collated to include isolates from countries in Southeast Asia that were likely linked to Viet Nam via migration. An alignment of the envelope (E) gene (1485 nt) was assembled for the Southeast Asian and Vietnamese isolates (n = 134 and 751, respectively) to include the broadest range of locations. An initial neighbor-joining tree was constructed in PAUP* [39], using a HKY85 nucleotide substitution model with gamma-distributed rates. This allowed us to make an initial identification of the major clades of DENV-1 in Viet Nam. These Vietnamese isolates were then subsampled (n = 101) to explore their phylogeography in context of the South East Asian isolates. Isolation dates for the South East Asia data set were obtained from GenBank annotations and via personal communication. Where specific dates were not available in terms of day and month, a mid-point of the year of isolation was used.
The spatial dynamics of DENV-1 in Southeast Asia were investigated with a discrete diffusion model [26] using Bayesian Monte Carlo Markov Chain (MCMC) method implemented in BEAST [40]. The phylogeography analysis was executed with a codon-structured SDR06 substitution model [41], a relaxed uncorrelated lognormal clock [42] and a Gaussian Markov Random Field (GMRF) coalescent prior [43] over the unknown phylogeny. The discrete diffusion model used the country of isolation of the sampled sequences to reconstruct the ancestral location states of the internal nodes from the posterior time-scaled tree distribution. The MCMC was run for 50 million generations, sampling every 5000th state, and executed multiple times to ensure adequate mixing and stationarity had been achieved.
Major clades of Vietnamese DENV-1 identified from the broad-scale South East Asian analysis were selected for further study to examine the spatial and temporal variation in Viet Nam. In clades with appreciable numbers of sequences from Dong Thap and HCMC, isolates from these locations were analyzed independently to gauge the regional variation in viral transmission patterns. For the fine-scale analysis, a continuous diffusion model based on a lognormal relaxed random walk [27] was employed to model the DENV-1 spatial dynamics in Viet Nam. For each isolate, the specific sample date and location information in terms of the longitude and latitude of the patient's household were used. Isolates that were identical in sample date and location information were down-sampled so as to reduce the potentially biasing effect of over-sampling of epidemiologically-linked cases. The MCMC runs were evaluated as previously described, and the chain lengths ranged from 50 to 100 million generations, and were sampled regularly to yield 10,000 trees from the posterior distribution.
The viral dispersion rates (km/yr) for each data set were calculated across the tree (i.e. total straight-line distance travelled divided by the total time) and biannually to consider the spatial heterogeneity in a time-scaled framework. Plots of relative genetic diversity over time were reconstructed using the GMRF coalescent prior to reveal the association between the genetic diversity of each group in terms of their evolutionary history [43].
Further discrete phylogeography analyses were performed with the robust counting method [44], [45] to determine the extent of viral migration between Dong Thap and HCMC and whether this varied when the lineage originated in a rural or urban area. In this case, the discrete states were represented by either the isolate being sampled from HCMC, Dong-Thap or neither (non-Dong Thap or HCMC).
For the limiting case of a freely mixing (non-spatially structured) epidemic in HCMC, dispersion rates were estimated whilst randomizing the tip locations during the tree proposal in the MCMC, whilst co-estimating the rates for each independent lineage and the joint DENV-1 diffusion rate. To determine the viral transmission network within HCMC, a non-reversible discrete phylogeography model was applied to all the HCMC isolates, using the district of isolation for the discrete states. The analysis was performed and evaluated as described above with the addition of implementing Bayesian Stochastic Search Variable selection (BSSVS) to identify significant transition rates between locations [26]. The transition rates supported by a Bayes factor of at least 3 were examined further by looking at the number of in-degree and out-degree per district. The number of connections was normalized by the number of samples from the source location in order to reduce the bias from under-represented locations in our data set.
Patients (or their parents/guardians) gave written informed consent to participate in each of the studies. The study protocols were approved by the Hospital for Tropical Diseases and the Oxford University Tropical Research Ethical Committee.
|
10.1371/journal.pcbi.1004655 | Persistent Chaos of Measles Epidemics in the Prevaccination United States Caused by a Small Change in Seasonal Transmission Patterns | Epidemics of infectious diseases often occur in predictable limit cycles. Theory suggests these cycles can be disrupted by high amplitude seasonal fluctuations in transmission rates, resulting in deterministic chaos. However, persistent deterministic chaos has never been observed, in part because sufficiently large oscillations in transmission rates are uncommon. Where they do occur, the resulting deep epidemic troughs break the chain of transmission, leading to epidemic extinction, even in large cities. Here we demonstrate a new path to locally persistent chaotic epidemics via subtle shifts in seasonal patterns of transmission, rather than through high-amplitude fluctuations in transmission rates. We base our analysis on a comparison of measles incidence in 80 major cities in the prevaccination era United States and United Kingdom. Unlike the regular limit cycles seen in the UK, measles cycles in US cities consistently exhibit spontaneous shifts in epidemic periodicity resulting in chaotic patterns. We show that these patterns were driven by small systematic differences between countries in the duration of the summer period of low transmission. This example demonstrates empirically that small perturbations in disease transmission patterns can fundamentally alter the regularity and spatiotemporal coherence of epidemics.
| Measles epidemics continue to pose a significant public health risk wherever vaccination coverage is low. In such populations transmission rates tend to fluctuate seasonally, mirroring patterns of human aggregation, due to the timing of school terms, and/or the migration of workers and their families. Here we show empirically that slight changes in the seasonal pattern of measles transmission can lead to massive shifts in the complexity of measles dynamics, in some cases driving epidemic patterns that resemble deterministic chaos. Our analysis is based on a comparison of 20-year biweekly measles incidence time series in 80 major cities in the prevaccination era United States and United Kingdom. The results are important in two ways: first, in contrast to previous theory, we show that subtle shifts in seasonal patterns of transmission can cause deterministic chaos in the epidemic dynamics of acute immunizing infections; second, we demonstrate that this new route to deterministic chaos is significantly more robust to stochastic extinction compared with previous chaotic models, suggesting chaotic dynamics may be more common in natural populations than previously thought.
| Acute immunizing infections remain a leading cause of death worldwide, and have accounted for a significant portion of all morbidity and mortality throughout human history, especially among children and in countries without adequate vaccination coverage [1–4]. Understanding the processes that determine epidemic patterns in these diseases can aid in forecasting and improve the efficacy of public health interventions. Studying the epidemiological dynamics of these diseases also provides a unique window on population-level predictability and its limitations, in an important applied context.
Epidemics of acute immunizing infections often occur in predictable cycles[5–10]. The underlying drivers of measles epidemics are particularly well understood, consisting of the basic demographic clockwork of repeated depletion of the susceptible population by infection or vaccination, followed by susceptible recruitment through birth. Cycles of human aggregation from school holidays or the migration of workers and their families cause seasonal fluctuations in transmission to sustain recurrent epidemics [2,11]. This overall clockwork is modulated by secular variation in susceptible recruitment caused by changes in birth rate and vaccination uptake [12], and by demographic stochasticity and local extinction in small populations, which predisposes smaller towns and cities to be entrained to the dynamics of larger metropolitan centers [13,14]. Simple mathematical models that incorporate these drivers have in many cases successfully predicted incidence patterns, making measles a canonical system in the study of non-linear population dynamics and prime target for elimination [5,6,11,12,14].
The most intensively studied incidence patterns for measles are from Europe—notably the UK—during the prevaccination era and are characterized by stable limit cycles (regularly occurring seasonal epidemics) with annual or biennial periods [5,11,15,16]. In the biennial cycles, susceptible depletion in the major epidemic years is replenished by births throughout the following year, during which minor epidemics may occur[17]. Increasing birth rate in this context causes the susceptible population to replenish more rapidly, leading to a collapse from biennial to regular annual cycles, as observed during the post-World War II “baby boom” [5,12,18].
In contrast, recent analyses of measles in western Africa—notably Niger—have revealed complex dynamics, featuring episodic epidemics with highly variable amplitudes. These are primarily caused by sharp seasonal increases in population density driven by collective migration [2] resulting in deep epidemic troughs, through nonlinear resonant feedbacks [6,19]. Owing to the intense seasonality, the equations describing measles dynamics in Niger produce deterministically chaotic trajectories, as conjectured by previous theory [6,18]. However, the deep post-epidemic troughs invariably break the chains of transmission, precluding local persistence.
Previous case studies, therefore, suggest an impossible tension with respect to chaos in real world epidemics. Despite its mathematical plausibility [20–23], the large amplitude seasonal fluctuations in transmission rates that have been presumed a prerequisite for chaos [12,17,19], in practice result in so deep epidemic troughs that frequent stochastic extinctions are inevitable[19]. Thus exotic nonlinear dynamics and local persistence have been thought to be in opposition in nature [24]. We refute that hypothesis here by showing evidence of widespread persistent chaos in the epidemic dynamics of prevaccination measles in the United States, which emerged via a new route to chaos that is less prone to stochastic extinction.
We take a comparative approach, analyzing 20-year biweekly time series data on measles incidence in 80 major cities, 40 in the US and 40 in England and Wales (UK). To compare these contexts, we fit a Time-series Susceptible Infected Removed (TSIR) model to the measles incidence data for each of the 80 cities [5,6,11]. The TSIR model describes macroscale properties of the stochastic branching process of measles spread, focusing on the expectation for number of secondary cases arising from the current population of infected individuals (hereafter the “deterministic skeleton”) and the probability distribution describing variation around that expectation due to the stochastic nature of infectious disease spread.
Representing the number of infected individuals in generation t by It, the concurrent number of susceptible individuals by St, and the population size by Nt, the TSIR model is given by
E[It+1]=βtItαStNt−1
(1)
It+1∼Neg.Bin.(E[It+1],It)
(2)
where E[.] is the expected value and βt represents seasonally fluctuating transmission rate in each city. The mixing parameter α, usually set at slightly less than unity, accounts for latent inhomogeneity in contact patterns between susceptible and infected individuals [23], as well as compensating for instabilities arising from discretizing the underlying continuous-time process [25]. Following previous work [11,16], we use α = 0.975 for all cities, which leads to good performance under forward simulation of the model.
The TSIR model for measles operates at the characteristic two-week serial interval of infection. Eq 2 represents the birth-and-death stochasticity inherent in transmission dynamics resulting in a negative binomial distribution of new cases with mean E[It+1] and dispersion parameter It, so that the variance in It+1 is given by E[It+1] + E[It+1]2/It. To study the deterministic skeleton of the dynamics, we model It+1 = E[It+1] in place of Eq (2).
Susceptible dynamics are modeled as
St+1=St+Bt−It+1
(3)
where Bt is the observed time-varying birth rate in a given city (see below). Secular variation in susceptible recruitment is a well-known driver of variation in measles periodicity [12] that we account for by using data on birth rates for each city when fitting and doing forward simulations. The full procedure for fitting the TSIR model to data follows well established techniques [5,11] that also included here as Supporting Information (S1 Text).
We assembled biweekly time series of measles incidence in US cities using the Project Tycho database [26] and took biweekly measles incidence and demographic data for cities in the UK from previous work [11,27]. For US cities we took estimates of population size for each city over the period of the study from census data [28] and estimated effective birth rates by differencing biweekly time series of the number of children under one year old [29], adjusting for the rate at which children age out of this class. For total and infant population sizes in the US, biweekly time series were obtained by evaluating at each biweek a spline function fitted to the decennial data (see S1 Text). Variations in the approach to reconstructing US recruitment rate, including varying background infant mortality, and changing the degrees of freedom in spline fitting, did not affect the results.
We used data for the 40 US cities in the Project Tycho database with the most records of measles incidence, which included most major US cities. While the Project Tycho database has measles incidence data from 1903 to 1953, data coverage was uniformly high for these 40 cities between 1920 and 1940, so we used that period in the analysis. For the England and Wales measles data we used the city of London plus the largest 39 cities that were more than 50km from London to prevent a “borough effect” where UK cities in the greater London area are entrained to its dynamics.
Due to limitations on data availability, the US measles data we used extends from 1920–1940, whereas the England and Wales measles data extends from 1944 to 1964. Our analysis accounts for demographic differences associated with the changing time window between the US and the England and Wales data, including differences in birth rates over time among cities and countries. Consequently, the temporal mismatch between the US and UK data does not drive the observed epidemic patterns—evidence from other sources clearly shows that measles epidemics in London, UK and other major UK cities remained predominantly biennial and non-chaotic in the period covered by the US data (1920–1940; see S1 Text)[15,30].
Measles dynamics varied systematically among cities and countries in the prevaccination era, with US cities exhibiting more diverse and episodic epidemics than cities in the UK. Whereas measles dynamics in the UK were predominantly locked on a biennial cycle, as previously reported [5,6,11,31], a majority of US cities showed lower frequency, higher amplitude oscillations (Fig 1). Consequently, the mean periodicity of a city’s measles incidence (see S1 Text) varied more widely among US cities, and was higher on average, compared to cities in the UK.
We found a comparable systematic variation in the shapes of the underlying seasonal transmission patterns in each country, particularly a systematically lengthening of the summer period of low transmission in US relative to the UK (Fig 2A). Given the biology and demography of measles transmission, it is likely that this lengthening is associated with historical differences in timing and duration of school summer holidays between the two countries [11,32]. Corroborating historical data on the timing of school holidays in US cities is not currently available. Whatever their origins, we show that these systematic differences in transmission rates caused measles dynamics in the US to diverge from the stable annual or biennial limit cycles previously characterized in measles epidemics for the UK. US measles cycles exhibit higher and more variable mean periodicity (Fig 2B–2E) and are more sensitive to initial conditions (Fig 2E–2G), which are hallmarks of complex population dynamics[33]. This conclusion is supported by several lines of evidence, as follows.
First, twenty-year forward simulations of the deterministic skeleton of the TSIR model parameterized with the fitted seasonal transmission function for each city yielded close fits to the times series of measles incidence (Fig 2B–2D and Fig C in S1 Text), confirming that observed differences among cities and countries in epidemic complexity can be explained by systematic variation in seasonal transmission patterns. Occasional discrepancies between the data and the forward simulations (such as for London in 1964, where the epidemic was smaller than predicted) are due in part to the accumulation over many biweekly timesteps of measurement errors in the data on birth rates and case counts, which were imperfectly reported. In addition, latent processes not included in the TSIR model, such as variation in age structure, may cause discrepancies between the forward simulations and the data. However, previous work [12] has shown that the long-term impact of such latent structure may be encoded in the shape of the seasonal transmission function, which could explain how simple models can successfully capture the key features of epidemics in complex populations, as is the case with the measles periodicity described here.
As a second line of evidence for transmission-driven differential complexity in US measles epidemics, the deterministic TSIR simulations of measles in US cities were much more sensitive to initial conditions relative to UK cities. That is, slight changes to the initial proportion of the population that was susceptible or infected in US cities produced large differences in epidemic periodicity, but this was not the case for UK cities (Fig 2E–2G). The fine-scale dependence on initial conditions in US cities precludes long-range historical forecasts of measles epidemics in the prevaccination US because the outcome of such simulations depends on precise estimates of the initial proportion of the population that is susceptible and infected, which cannot be estimated without significant statistical uncertainty (Fig 2F and 2G). This is in contrast to the UK, where accurate forecasts of the prevaccination era incidence time series can be achieved from a wide range of starting conditions, making forecasts in the UK resilient to statistical uncertainty in the initial susceptibility of the population (Fig 2E).
Third, the stochastic model (Eq 2), which continually pushes epidemic trajectories away from the deterministic skeleton, provides further evidence of how cities with subtly different seasonal transmission patterns respond to perturbations in the number susceptible and infected. The stochastic simulations also show good correspondences between the model and the data, as measured by assessing whether the distribution of periodograms generated under repeated forward simulation of the stochastic model qualitatively matched the periodogram of the data. The distributions of periodograms for UK cities (Fig 2H) were less dispersed than in the US (Fig 2I and 2J). For US cities, stochastic simulations revealed multiple distinct periodic patterns. These distinct patterns coexist in the same parameter space, emerging as a result of stochastic variation in the simulation process alone (Fig 2I and 2J). In this case the data match a subset of the possible periodograms, while other periodograms suggested by the model for a given parameterization were not observed (Fig 2J). In the parlance of dynamical systems theory this suggests the presence of coexisting attractors[12] or important unstable manifolds [34,35]. In practical terms, the stochastic simulations imply that if time could be repeatedly wound back and played again from a similar starting point, biennial measles dynamics in UK cities would still be biennial, whereas US cities would display a diversity of possible trajectories. Where it happens that measles cycles in a US city are predominantly biennial (e.g. New York), the regular periodicity belies a sensitivity to initial conditions.
As further evidence for chaotic measles dynamics in the US we calculated dominant Lyapunov Exponents (LE; see S1 Text) for each city in the data. LEs measure the rate at which similar epidemic trajectories converge (LE < 0) or diverge (LE >0) [6,36], quantifying the sensitivity of a dynamical system to small changes in state. While all LEs for cities in England and Wales were negative, the majority of US cities had positive LEs (Fig 3A). This corroborates the results of the stochastic simulations, providing another line of evidence for sensitive dependence on initial conditions across US cities, due to a slight change in the shape of the seasonal transmission function relative to cities in the UK.
Surprisingly, however, the resulting complex dynamics in the US were as stable as those in the UK in terms of the risk of local extinction, with local extinction rare in all cities above around 300,000 inhabitants in both countries (Fig 3B). The highly irregular measles dynamics of the US are, thus, as robust to stochastic extinction as the clock-like regularity of the epidemics in the UK. This is surprising because previous models of chaotic epidemic dynamics for seasonally immunizing infections predicted that increased complexity is accompanied by increased risk of stochastic extinction, apparently precluding persistent deterministic chaos in real-world scenarios [12,19,24].
Analysis of the clockwork underlying these epidemic dynamics reveals two distinct routes to deterministic chaos for seasonally modulated immunizing infections (Fig 4). To demonstrate these routes we began with the TSIR model for Los Angeles, US, which has a mean periodicity of ~3 years and a positive LE, and systematically varied the amplitude of the seasonal transmission function, and/or the duration of the period of low transmission (see S1 Text), while holding susceptible recruitment constant. On one hand, increasing the amplitude of seasonal oscillations in transmission leads to chaotic dynamics through the previously well-characterized route [6], corresponding to the extinction-prone measles dynamics observed in Niger, where deep epidemic troughs frequently break local chains of transmission [19]. On the other, increasing the duration of the seasonal period of low transmission, while holding seasonal amplitude constant, also leads to chaos. In contrast to the chaotic measles epidemics previously described in Niger, the new route to chaos revealed in the prevaccination US is associated with local persistent chains of transmission (Fig 3B). Therefore, although these distinct routes to chaos yield equivalent levels of deterministic complexity, they are associated with contrasting properties of local persistence: only the new low-amplitude route to chaos exemplified by measles in US cities can sustain true chaotic fluctuations for a significant period of time (Fig 4B).
The existence of distinct routes to chaos with contrasting probabilities of local extinction explains both the complexity and persistence of measles epidemics in the prevaccination United States, as well as the systematic differences between measles epidemics in the prevaccination US, the prevaccination UK and in present-day sub-Saharan Africa. In particular, we found transitions to complex epidemic dynamics do not necessitate high-amplitude fluctuations in transmission rates nor broad secular changes in susceptible recruitment, as previously thought [12]. The US analysis shows that subtle shifts in seasonal transmission patterns can also lead to chaos. But the origins of dynamic complexity—whether through the canonical routes or the newly described low-amplitude route operating in the US—have important implications for the local persistence of the resulting epidemics.
Finally, we note that systematic differences in dynamics at the city level may have propagated to affect countrywide patterns in the spatiotemporal coherence of disease incidence patterns. While the annual or biennial predictable measles cycles in UK cities represented synchronous phase locked oscillations across the entire island[14,37], the locally persistent chaotic dynamics in the US appeared to break the phase-lock in US measles epidemics, both over the same spatial scale as the UK, and overall (Fig B in S1 Text). However, further, more detailed, spatial analyses are necessary to tackle systematic variation in the strength of these correlations at inter-city and regional scales. This may be an interesting avenue for future work.
The realization in the 1970s that simple models of population growth can have complex dynamics [20], spurred several decades of effort in ecology and epidemiology to explain highly variable time series using a few general equations [22,38], with hopes of emulating the success of Newtonian physics [24]. Although controlled laboratory experiments supported the hypothesis that complex dynamics in living populations can emerge from simple rules [39] applications to real-world scenarios were often stymied by the role of stochasticity—chance events play a significant role in the growth trajectories of many live populations, but such variation is minimized in deterministic models and controlled experiments.
A particular challenge was the fact that canonical chaotic models of seasonally forced epidemics carried a high risk of stochastic extinction. Specifically, the route to chaos described by these models involves broad-scale changes in susceptible recruitment, such as changes in birth rate or vaccination coverage, or significant structural changes in the seasonal pattern of transmission, such as changes in the amplitude of seasonal fluctuations in transmission rate [12,19]. But these structural changes result in deep epidemic troughs, where the chain of transmission is maintained by only a few individuals. This greatly increases the likelihood that an epidemic will fade out, due to random variation in the timing of infection and removal events [12,19,40]. This tradeoff, where achieving a realistic level complexity requires an unrealistic rate of stochastic fadeouts, apparently precluded persistent deterministic chaos as an explanation for capricious incidence time series. In contrast, we have shown that small shifts in the seasonal pattern of disease transmission can offer a new, more stable, route to persistent deterministic chaos (Fig 4).
The relative importance of noise and determinism in population dynamics varies with context: for instance, stochasticity appears to generate proportionally more of the amplitude in rubella cycles (as well as driving patterns in local extinction), because the deterministic skeleton of these dynamics falls on an attractor that is globally less stable [41]. Similarly, the intermittent 3–4 year periodic pertussis dynamics is thought to emerge from stochastic resonance around a deterministic skeleton with a dominant annual period [42]. For US measles dynamics in the prevaccination era, the effects of stochasticity and determinism are inextricably intertwined through highly nonlinear sensitive dependence on initial conditions.
Although not conclusive, analysis of cross correlation in measles incidence across cities suggests that US cities may have had less synchronized epidemics at regional and country-wide scales (Fig B in S1 Text). The level of synchrony among connected populations has been shown to influence patterns of disease persistence per se, but predictions for the impact of chaos on metapopulation dynamics have been somewhat equivocal, as epidemic chaos has its own complex relationship with persistence. Specifically, spatial decorrelation, such as that seen among US cities, can improve disease persistence in a metapopulation context, as subpopulations that experience local extinctions may be more likely to be rescued by connected subpopulations that have not experienced a fadeout—the metapopulation “rescue effect” [43–45]. However, spatial decorrelation specifically linked to chaos was previously thought to be an unlikely source of pathogen persistence, because deep seasonal troughs in transmission rates, thought to be a prerequisite to chaos, tend to synchronize the timing of fadeouts across the metapopulation, diminishing the rescue effect [40]. The new route to locally persistent but decorrelated dynamics may change this perspective.
Our analysis demonstrates the impacts of chaos on the metapopulation dynamics of cities, showing that the network consequences of complex epidemic patterns depend on the origins of the complexity. On one hand, if complex cycles emerge via high amplitude fluctuations in transmission rates, then local populations will be more likely to experience synchronous fadeouts, and metapopulation rescue effects will not considerably improve local persistence[19,40]. On the other hand, if complex epidemic cycles emerge from slight changes in the duration of the seasonal period of low transmission—as shown here for measles in the prevaccination era US—local populations will experience relatively higher rates of disease persistence, in addition to a plausibility of significant metapopulation rescue effects. The enhanced persistence is consistent with the data presented here, where the observed probability of measles fadeouts across US cities (Fig 3B) was still lower than that predicted in single-city simulations under the more stable route to chaos operating in the US, suggesting the presence of a rescue effect[44] (Fig 4B).
In conclusion, the emergence of persistent chaotic epidemics in the prevaccination US from small shifts in seasonal transmission patterns reveals a novel and potentially widespread route to chaos in population dynamics[24,46,47]. Moreover, these results show empirically that the viability of chaotic populations depends subtly on the route to chaos. In practice, this means that small perturbations in transmission rates, such as those caused by shifts in host behavior or the imposition of epidemic control measures, can lead to a rapid erosion of the capacity to forecast epidemic patterns, which can in turn reduce the efficacy of control strategies such as reactive vaccination[2,19,48,49]. Generally, population dynamics are deterministically more sensitive to perturbations than previously thought.
|
10.1371/journal.ppat.1004304 | EBNA3C Augments Pim-1 Mediated Phosphorylation and Degradation of p21 to Promote B-Cell Proliferation | Epstein–Barr virus (EBV), a ubiquitous human herpesvirus, can latently infect the human population. EBV is associated with several types of malignancies originating from lymphoid and epithelial cell types. EBV latent antigen 3C (EBNA3C) is essential for EBV-induced immortalization of B-cells. The Moloney murine leukemia provirus integration site (PIM-1), which encodes an oncogenic serine/threonine kinase, is linked to several cellular functions involving cell survival, proliferation, differentiation, and apoptosis. Notably, enhanced expression of Pim-1 kinase is associated with numerous hematological and non-hematological malignancies. A higher expression level of Pim-1 kinase is associated with EBV infection, suggesting a crucial role for Pim-1 in EBV-induced tumorigenesis. We now demonstrate a molecular mechanism which reveals a direct role for EBNA3C in enhancing Pim-1 expression in EBV-infected primary B-cells. We also showed that EBNA3C is physically associated with Pim-1 through its amino-terminal domain, and also forms a molecular complex in B-cells. EBNA3C can stabilize Pim-1 through abrogation of the proteasome/Ubiquitin pathway. Our results demonstrate that EBNA3C enhances Pim-1 mediated phosphorylation of p21 at the Thr145 residue. EBNA3C also facilitated the nuclear localization of Pim-1, and promoted EBV transformed cell proliferation by altering Pim-1 mediated regulation of the activity of the cell-cycle inhibitor p21/WAF1. Our study demonstrated that EBNA3C significantly induces Pim-1 mediated proteosomal degradation of p21. A significant reduction in cell proliferation of EBV-transformed LCLs was observed upon stable knockdown of Pim-1. This study describes a critical role for the oncoprotein Pim-1 in EBV-mediated oncogenesis, as well as provides novel insights into oncogenic kinase-targeted therapeutic intervention of EBV-associated cancers.
| The oncogenic serine/threonine kinase Pim-1 is upregulated in a number of human cancers including lymphomas, gastric, colorectal and prostate carcinomas. EBV nuclear antigen 3C (EBNA3C) is essential for EBV-induced transformation of human primary B-lymphocytes. Our current study revealed that EBNA3C significantly enhances Pim-1 kinase expression at both the transcript and protein levels. EBNA3C also interacts with Pim-1 and can form a complex in EBV-transformed cells. Moreover, EBNA3C increases nuclear localization of Pim-1 and stabilizes Pim-1 protein levels by inhibiting its poly-ubiquitination. Additionally, EBNA3C augments Pim-1 mediated phosphorylation of p21 and its proteosomal degradation. Stable knockdown of Pim-1 using si-RNA showed a significant decrease in proliferation of EBV transformed lymphoblastoid cell lines and subsequent induction of apoptosis by triggering the intrinsic apoptotic pathway. Therefore, our study demonstrated a new mechanism by which the oncogenic Pim-1 kinase targeted by EBV latent antigen 3C can inhibit p21 function, and is therefore a potential therapeutic target for the treatment of EBV-associated malignancies.
| Epstein-Barr virus (EBV), a ubiquitous lymphotropic herpesvirus, latently infects human populations worldwide [1]. EBV infection is typically asymptomatic and is an important etiological factor which contributes to different human malignancies [2]. EBV is consistently associated with nasopharyngeal carcinoma (NPC) [3], African Burkitt's lymphoma (BL) [4], post-transplantation lymphoproliferative disease (PTLD) [5], Hodgkin's disease (HD) [6], and AIDS-related non-Hodgkin's lymphomas (AIDS-NHL) [7]. Additionally, EBV is also found in a fraction of gastric carcinomas particularly in Asian and African countries [8]. EBV has the potential to transform human B-lymphocytes in vitro by maintaining a continuous proliferative state, known as “immortalization” which generates permanent lymphoblastoid cell lines (LCLs) [9]. The LCLs which are produced in culture carry the viral genome as extra-chromosomal episomes and express nine latent EBV proteins including, the six nuclear antigens (EBNA 1, 2, 3A, 3B, 3C & LP), an additional three membrane associated proteins (LMP1, LMP2A & 2B), and the two EBV-encoded small RNAs (EBERs) [10]. These viral factors help to activate the quiescent B-cells from G0 into the cell cycle, and to sustain proliferation and maintenance of the viral genome [11].
Among the potential EBV latent antigens, EBNA3A, EBNA3B, and EBNA3C are sequentially encoded in the EBV genome and generate protein products of approximately 1,000 aa. Moreover, the EBNA3A, EBNA3B, and EBNA3C amino- terminal homologous domains are associated with RBP-Jk which mediates the association of EBNA2 and Notch with DNA [12]. EBNA3C and EBNA3A are also essential for EBV to drive primary human B-lymphocytes into continuously proliferating LCLs and for maintaining LCL growth [13]. Notably, Epstein-Barr virus nuclear antigen 3C (EBNA3C) plays an intricate regulatory role in the transcription of several viral and cellular genes [14]. EBNA3C targeted RBP-Jκ antagonizes EBNA2-mediated transactivation [15], and cooperates with EBNA2 in activating the major viral LMP1 promoter [16]. EBNA3C was found to regulate chromatin remodeling by recruiting histone acetylase and deacetylase activities [17]. Moreover, EBNA3C modulates the transcriptional level of cellular genes which are involved in cell migration and invasion by targeting the metastasis suppressor Nm23-H1 [18]. In addition, EBNA3C can modulate diverse cellular functions, presumably mediated by direct protein–protein interactions [19]. EBNA3C also stabilizes c-Myc and interacts with Mdm2 to modulate p53 mediated transcription and apoptotic activities [20], [21]. Interestingly, EBNA3C was found to be crucial for regulating the activity of cellular kinases. Recently, we have shown that EBNA3C enhances the kinase activity of cell-cycle regulatory protein Cyclin D1 which allows for subsequent ubiquitination and degradation of the tumor suppressor pRb [22].
Provirus integration site for Moloney murine leukemia virus (Pim-1), a proto-oncogene encoding a serine/threonine kinase, is linked to several cellular functions involving cell survival, proliferation, differentiation, and apoptosis [23]. It was reported that overexpression of Pim-1 is associated with the development and progression of multiple hematopoietic malignancies such as B-cell lymphomas, erythroleukemias, and acute myelogenous leukemia, T-cell lymphomas, and non-hematological malignancies including, oral squamous cell carcinoma, and prostate cancer [24]. During the process of embryo development, Pim-1 is highly expressed in liver, spleen and bone marrow in typical hematopoietic progenitors [25], [26], neonatal heart [27], central nervous system [28], and mammary gland [29]. Surprisingly, at the adult stage, Pim-1 is only slightly expressed in circulating granulocytes [26]. Previous reports also indicated that heterologous expression of Pim-1 in transgenic mice leads to increased lymphoproliferation and inhibition of apoptosis [30]. Augmented expression of Pim-1 in lymphoid cells by transgenesis highlighted its potential for oncogenesis [31]. Being a potent serine/threonine kinase, Pim-1 is able to phosphorylate itself [32], [33], through an autophosphorylation site that diverges from its consensus phosphorylation motif [34]. Several Pim-1 substrates have been identified, including p21Cip1/WAF1 [35], [36], Cdc25A [37], PTPU2 [38], NuMA [39], C-TAK1 [40], and Cdc25C [41], indicating a crucial role for Pim-1 in cell proliferation through both the G1/S and G2/M phase transition. Pim-1 also possesses anti-apoptotic activity [42], and recent reports have demonstrated a role for Pim kinases in regulation of herpesviral oncogenesis. KSHV encoded LANA was found to be crucial for transcriptional activation of Pim-1 in KSHV-positive cells and it also acts as a Pim-1 substrate [43]. In the context of EBV infection, studies have shown that Pim-1 may be required for LMP1-induced cell survival [44]. Furthermore, the expression levels of Pim-1 and Pim-2 are up-regulated upon EBV infection and they in turn enhance the activity of the viral nuclear antigen EBNA2, suggesting a role in driving EBV-induced immortalization [45]. However, the molecular mechanism by which Pim-1 is activated through expression of viral antigens which creates a micro-environment for B-cell transformation is not fully elucidated.
In our current study, we demonstrated that EBNA3C is responsible for inducing Pim-1 expression in EBV transformed B-cells as well as in EBV-infected PBMCs. Further, we showed that EBNA3C interacts with Pim-1 through a small N-terminal domain (amino acids 130–159) and forms a complex in B-cells. Our results demonstrated that EBNA3C stabilized the Pim-1 protein by inhibiting its degradation by the ubiquitin/proteasome pathway. Interestingly, EBNA3C also facilitated the nuclear localization of Pim-1, and promotes EBV-induced cell proliferation by regulating Pim-1 mediated degradation of p21/WAF1. We observed that deregulation of p21 ultimately resulted in higher cellular proliferation. Lentivirus mediated stable knockdown of Pim-1 resulted in a significant reduction of EBV transformed cells and induction of apoptosis. Cumulatively, these findings demonstrate a vital role for Pim-1 in EBV-mediated oncogenesis and also support the conclusions that Pim-1 kinase is a potential target for therapeutic intervention strategies against EBV associated malignancies.
Pim-1 expression was found upregulated in different hematological and non-hematological malignancies [23]. To determine whether EBV latent antigen 3C modulates Pim-1 expression, 10 million human peripheral blood mononuclear cells (PBMC) were infected by wild type and mutant ΔEBNA3C BAC-GFP-EBV for 4 hrs at 37°C described previously [46]. The mRNA and protein levels of Pim-1 were detected after 0, 2, 4, 7 days of infection. Our results showed upregulation of both the transcript and protein levels of Pim-1 with wild type EBV infection (Fig. 1A). Interestingly, infection with ΔEBNA3C BAC-GFP-EBV resulted in low Pim-1 expression at 2 days post-infection and returned to the levels seen for infected cells at 0 day post-infection (Fig. 1B). The results indicated that Pim-1 expression was induced by wild-type EBV infection. Therefore, we wanted to determine the expression pattern of Pim-1 in EBV transformed Lymphoblastoid cells LCL1, LCL2, and EBNA3C stably expressing BJAB7 and BJAB10 cells when compared to EBV negative BJAB. Our results showed that Pim-1 expression was highly upregulated in LCL1, LCL2, BJAB7 and BJAB10 cells (Fig. 1C). To investigate the role of EBNA3C on Pim-1, we monitored the Pim-1 protein expression levels with a dose dependent increase of EBNA3C in EBV negative DG75 as well as in HEK-293 cells. The results showed a steady increase in Pim-1 expression levels in both cell lines (Fig. 1D and 1E). Moreover, Real-time PCR analysis showed upregulation of Pim-1 mRNA expression in BJAB7 and LCL1 cells when compared to EBV negative BJAB cells (Fig. 1F, left panel). To further investigate the role of EBNA3C in inducing Pim-1 expression, we performed Real-time PCR as well as Western blot analysis on EBNA3C stable knock-down LCL1 cells. The results demonstrated a substantial reduction of Pim-1 expression in both mRNA and protein levels as compared with sh-control LCL1 cells (Fig. 1F, right panel and Fig. 1G). Moreover, to check the role of other EBV antigens including EBNA2, EBNA3A, and EBNA3B on Pim-1 expression, we performed si-RNA mediated knockdown of EBNA2, EBNA3A, EBNA3B and EBNA3C in LCL1 cells. Our Real-time PCR analysis demonstrated that Pim-1 mRNA level is significantly reduced upon EBNA3C knockdown but no significant change was observed in Pim-1 mRNA expression level with EBNA2, EBNA3A, EBNA3B knockdown further suggesting a major role for EBNA3C in upregulating Pim-1 expression (Fig. S1A and S1B). Additionally, we performed Western blot analysis to determine whether knock down of EBNA3C may have an effect on other EBNAs expression levels. The results demonstrated that expression levels of other EBNAs were not affected with EBNA3C knockdown (Fig. S2).
To determine whether EBNA3C interacted with Pim-1, we performed co-immunoprecipitation experiments in HEK-293 cells by expressing Myc-tagged Pim-1, Flag-EBNA3C, or Myc-EBNA3C. Immunoprecipitation was performed using A10 (Fig. 2A) or 9E10 antibody (Fig. 2B). The results clearly demonstrated that EBNA3C strongly associated with Pim-1 (Fig. 2A and B). We further supported our results by GST-pull down assays using EBV negative BJAB, EBNA3C expressing BJAB10 and EBV transformed LCL1 cell lysates incubated with bacterially purified GST-Pim-1 protein. EBNA3C was detected by A10 antibody [47] which showed a substantial level of association between Pim-1 and EBNA3C in the EBNA3C stable cell lines as well as in an LCL (Fig. 2C). Coomassie staining of bacterially purified GST and GST-Pim-1 proteins are shown in Fig. 2D. We also observed the association between EBNA3C and Pim-1 in BJAB7, BJAB10, LCL1, LCL2 cells compared with BJAB in separate co-immunoprecipitation experiments by using Pim-1 specific antibody (Fig. 2E and 2F).
To determine the specific domain of EBNA3C associated with Pim-1, we performed co-immunoprecipitation experiments expressing GFP-tagged Pim-1 with Myc-tagged full length (residues 1–992) and different truncated mutants (residues 1–365, 366–620 and 621–992) of EBNA3C in HEK-293 cells. Immunoprecipitation (IP) was performed by using either 9E10 or GFP-specific antibodies. The results indicated that Pim-1 strongly associated with full length as well as the N-terminal domain (residues 1–365) of EBNA3C (Fig. 3A and 3B). We extended the binding experiments by performing in vitro GST-pulldown assay with in vitro translated full length and truncated mutants of EBNA3C including fragments within the N-terminal domain (residues 1–992, 1–365, 366–620, 621–992, 1–100, 100–200, 200–300, 366–992, 1–129, 1–159, 1–250, 130–300). Our results demonstrated that EBNA3C residues 100–200, 1–159, 1–250, 130–300 associated strongly with full length Pim-1 (Fig. 3C). To further map the specific binding residues, we performed additional in vitro GST-pulldown assays by using in vitro translated full length Pim-1 incubated with bacterially expressed N-terminal truncated mutants of GST-EBNA3C fused to residues 90–129, 130–159, 130–190, 160–190. The results indicated that Pim-1 strongly bound to residues 130–159 of EBNA3C (Fig. 3D, 3E and 3F).
In the context of cancer progression, the significance of different subcellular localization patterns of Pim-1 has not been fully elucidated. Previous studies suggested that irradiation can promote nuclear translocation of Pim-1 in radio-resistant squamocellular malignancies of head and neck [48]. Importantly, nuclear localization of Pim-1 may correlate with the proliferating cells and may also contribute to a survival response upon pathologic injury [49]. In our study, we transfected Myc-tagged Pim-1 with or without GFP-tagged EBNA3C expression vectors in HEK-293 cells. Cellular localization of Pim-1 was examined by immunofluorescence analysis using specific antibodies against the Myc-epitope. Interestingly, our results showed that the localization of Pim-1 was predominantly in the cytoplasm without EBNA3C and was translocated to the nucleus in the presence of EBNA3C. Also, strong co-localization with Pim-1 and EBNA3C was observed (Fig. 4A and 4C). To further validate these results, we performed nuclear and cytosolic fractionation assays using transiently transfected HEK-293 cells with Myc-tagged Pim-1 with or without Flag-tagged EBNA3C expression vectors. Our Western blot analysis with nuclear and cytosolic fractions showed that in the presence of EBNA3C, the level of Pim-1 substantially increased in the nuclear fraction (Fig. 4B). Moreover, we corroborated the above observations in EBV negative BJAB, EBNA3C stably expressing BJAB10 and EBV transformed LCL1 cells using specific antibodies against Pim-1 and EBNA3C. The results showed that Pim-1 was mostly localized in the nucleus in both EBNA3C expressing BJAB10 and EBV transformed LCL1 cells, but was almost entirely cytoplasmic in EBV negative BJAB cells (Fig. 4D).
Recent reports suggested that expression of EBNA3C is responsible for the stabilization of different oncoproteins, transcription factors and cellular kinases [19], [22], [50], [51], and also plays an important role in modulating the ubiquitin (Ub)-proteasome machinery [52]. Our results so far showed that EBNA3C is important for enhanced protein expression of Pim-1. To determine, if this induced expression is related to EBNA3C-mediated stabilization of Pim-1 by the inhibition of Ub-proteosome machinery, we co-transfected Myc-tagged Pim-1 with or without Flag-tagged EBNA3C expression plasmids in HEK-293 cells which were treated with or without the proteasome inhibitor MG132. The results showed a substantial accumulation of Pim-1 protein levels in MG132 treated cells in the presence of EBNA3C compared with mock treatment and control vector (Fig. 5A). Next, we performed the stability assay of Pim-1 by transfecting Myc-tagged Pim-1 with or without Flag-tagged EBNA3C in HEK-293 cells. After 36 hours of post-transfection, cells were treated with the protein synthesis inhibitor cyclohexamide and harvested at 0, 3, and 6 hours intervals. The Western blot results clearly demonstrated that Pim-1 levels were stabilized with co-expression of EBNA3C whereas, the Pim-1 expression levels were markedly reduced with cyclohexamide treatment by 3 to 6 hours in the absence of EBNA3C (Fig. 5B). To further corroborate our results, we extended the stability assays with EBV negative BJAB, EBNA3C stably expressing BJAB10 and EBV transformed LCL1, control vector transfected and EBNA3C stably knockdown LCL1 cells. As anticipated, our results showed that Pim-1 protein levels were stabilized in BJAB10, LCL1 and sh-Ctrl LCL1 cells as well but significantly reduced in BJAB, sh-E3C LCL1 cells over time with the treatment of cyclohexamide (Fig. 5C and 5D).
The enhanced stability of Pim-1 in the presence of EBNA3C encouraged us to investigate the role of EBNA3C for regulating Pim-1 poly-ubiquitination. Therefore, we performed in vivo poly-ubiquitination assays in cells by co-transfecting with control vector, Myc-tagged Pim-1, HA-Ubiquitin, with or without Flag-tagged EBNA3C in HEK-293 cells. The results demonstrated a significant reduction of Pim-1 poly-ubiquitination levels in the presence of EBNA3C (Fig. 6A). To further validate the role of EBNA3C, we performed poly-ubiquitination assays by using the wild type Myc-tagged EBNA3C and its specific mutant EBNA3C (Myc-EBNA3C-C143N) expression vectors. We observed higher poly-ubiquitination levels of Pim-1 in the presence of the EBNA3C-C143N mutant compared with wild type (Fig. 6B). We also performed the ubiquitination assays in a B-cell background by using EBV-negative BJAB, EBNA3C stably expressing BJAB10 and EBV transformed lymphoblastoid LCL1, as well as the sh-Ctrl and sh-EBNA3C LCL1 cell lines. Our result showed that the status of Pim-1 ubiquitination was much lower in BJAB10 and LCL1 cells compared with BJAB (Fig. 6C) and somewhat enhanced upon EBNA3C knockdown (Fig. 6D).
The serine/threonine-protein kinase Pim-1 is upregulated in a number of hematological malignancies such as leukemia [26], mantle-cell lymphoma [53], and diffuse large B-cell lymphoma (DLBCL) [54]. A wide range of Pim-1 substrates were identified including, BAD [55], NuMa [39], Socs [56], Cdc25A [37], C-TAK1 [40], NFATc [57], HP-1 [58], PAP-1 [59], and cyclin-dependent kinase inhibitor p21 or p21Cip1/WAF1 [35], which suggested that Pim-1 can function at different cellular events, such as cell proliferation, differentiation, and cell survival [60]. Earlier reports showed that p21 suppresses tumors by promoting cell cycle arrest in response to various stimuli. Furthermore, considerable evidence from biochemical and genetic studies have demonstrated that p21 can act as a master effector molecule of multiple tumor suppressor pathways for promoting anti-proliferative activities which are independent of classical p53 tumor suppressor pathway [61]. Studies have also shown that enhanced levels of Pim-1 kinase phosphorylates Thr145 residue, and regulates the activity of p21Cip1/WAF1 [35]. Therefore, we checked the kinase activity of Pim-1 towards its substrate p21 with or without EBNA3C to investigate whether EBNA3C can modulate the phosphorylation status of p21. HEK-293 cells were transiently transfected with control vector, with and without Myc-tagged Pim-1 and increasing doses of Flag-tagged EBNA3C expression vectors. Immunoprecipitation was performed using anti-Myc 9E10 antibody and immunoprecipitated complexes were further examined for in vitro kinase activity as determined by GST-p21 phosphorylation. Interestingly, the results demonstrated that the ability of Pim-1 kinase to phosphorylate p21 was substantially and proportionally augmented by a dose-dependent increase in EBNA3C expression (Fig. 7A). We further extended the kinase assay using a kinase-dead (KD) mutant of Pim-1 (Fig. 7B). As anticipated, there was no kinase activity observed with the kinase-dead (KD) mutant of Pim-1 when compared with wild type. Next, we performed in vitro kinase assay for Pim-1 in the presence or absence of EBNA3C by using wild type and mutant (T145A) GST-p21 as substrate. The results showed no phosphorylation with mutant (T145A) p21 in comparison with wild-type, even in the presence of EBNA3C (Fig. 7C). This suggested that the Thr145 residue is important for EBNA3C mediated enhancement of p21 phosphorylation by Pim-1 kinase.
Earlier reports suggested the potential of a complex containing Pim-1 and p21 in cells [36]. We have now confirmed a strong association between Pim-1 and EBNA3C above. We then performed competitive binding assays in HEK-293 cells by co-transfecting increasing doses of EBNA3C-expression construct and a constant amount of Myc-tagged Pim-1 and Flag-tagged p21. Immunoprecipitation (IP) was performed with anti-Myc antibody for immunoprecipitation of complex with Pim-1. Our results demonstrated that increasing doses of EBNA3C can result in reduced association between Pim-1 and p21 (Fig. 8A). Previous studies showed that p21 is a prime target for ubiquitination in gliomas [62], and was dependent on the ubiquitin ligase APC/CCdc20 for its proteolytic degradation by the proteasome [63]. To explore the modulation of p21 protein levels by EBNA3C through regulation of the Ub-proteasome machinery, HEK-293 cells were co-transfected with Myc-Pim-1, Flag-p21, and increasing amounts of untagged-EBNA3C then treated with the proteasome inhibitor, MG132. The results indicated that the level of p21 was significantly reduced in the mock treated cells. However, with MG132 drug treatment, the level of p21 was further enhanced in the presence of EBNA3C (Fig. 8B).
Previous reports suggested that p21 regulates fundamental cellular processes, including cell cycle progression, apoptosis, and transcription on DNA damage response [64], [65]. Interestingly, involvement of p21 in all these major signaling pathways may occur not only after DNA damage response, but also depends on physiological conditions [66], [67]. To determine whether EBNA3C alone or an EBNA3C/Pim-1 complex had a role in p21 stabilization in DNA damage response, we performed stability assays using cyclohexamide treated HEK-293 cells co-transfected with different combinations of untagged-EBNA3C, Myc-Pim-1, Myc-Pim-1 KD (kinase dead) mutant, and Flag-p21 expression constructs. The experiments were performed with or without DNA damage response signal (reduction of serum with etoposide drug treatment). Interestingly, the results demonstrated that p21 expression levels were substantially reduced with co-expression of wild type Pim-1 and EBNA3C. However, p21 expression levels remained unchanged with EBNA3C, wild type Pim-1 or kinase dead Pim-1 alone with or without DNA damage (Fig. 9A, upper and lower panels). Therefore, EBNA3C contributes to the process of p21 degradation in co-operation with wild type Pim-1. We also extended our stability assays in EBV negative BJAB, EBNA3C stably expressing BJAB10, EBV transformed lymphoblastoid LCL1, sh-Ctrl and sh-EBNA3C LCL1 cells with cyclohexamide treatment in the presence or absence of etoposide induced DNA damage. P21 protein expression was found significantly reduced in LCL1, BJAB10 compared with BJAB even both with or without DNA damage response (Fig. 9B, upper and lower panels). Importantly, the expression levels were found augmented with or without DNA damage in EBNA3C stable knockdown LCL1 cells (Fig. 9C, upper and lower panels) suggesting a role for EBNA3C in deregulating p21 stability independent of etoposide induced DNA damage response.
To examine, whether EBNA3C has a vital role in p21 degradation alone or in collaboration with Pim-1, we performed poly-ubiquitination assays by expressing Flag-p21 and Myc-tagged EBNA3C in HEK-293 cells. The results demonstrated there was no significant change in the level of poly-ubiquitination (Fig. 10A). Our study also revealed a strong association with p21 and EBNA3C by co-immunoprecipitation experiments (Fig. S3A). Next, we attempted to examine the potential changes in p21 protein levels by expressing Flag-p21, Myc-Pim-1, along with increasing amounts of EBNA3C in HEK-293 cells. Interestingly, we observed reduced levels of p21 with a dose dependent increase of EBNA3C in the presence of Pim-1 (Fig. S3B). Our poly-ubiquitination assay results for p21, with wild-type Pim-1 and kinase-dead Pim-1 mutant clearly showed that the level of poly-ubiquitination was much higher with wild-type Pim-1 compared with kinase-dead mutant in the presence of EBNA3C (Fig. 10B). This supported an important role for EBNA3C in enhancing Pim-1 kinase activity and is likely to be required for p21 degradation. Moreover, we extended the poly-ubiquitination assays using the P21T145A mutant to determine whether the p21 Thr145 phosphorylation was related to its degradation. The results showed that levels of poly-ubiquitination remained unchanged both with wild-type and the kinase-dead mutant of Pim-1 in the presence of EBNA3C (Fig. 10C). Additionally, we performed poly-ubiquitination assay using EBV negative BJAB, EBNA3C stably expressing BJAB10, EBV transformed lymphoblastoid LCL1, as well as sh-Ctrl and sh-EBNA3C LCL1 cells to monitor the poly-ubiquitination status of p21. The results clearly indicated higher poly-ubiquitinated levels of p21 in EBNA3C expressing BJAB10, and LCL1 cells compared with EBNA3C negative BJAB (Fig. 10D). The levels were also reduced in EBNA3C stable knockdown LCL1 cells (Fig. 10E). To determine if the degradation of p21 is Pim-1 dependent, we performed poly-ubiquitination assays with Pim-1 stable knockdown LCL1 cells. The results indicated that upon Pim-1 knockdown, the level of p21 degradation was reduced (Fig. 10F).
Earlier reports demonstrated that Pim-1 kinase activity is linked to enhanced cellular proliferation in neoplastic cell types [68]. To determine the effect of EBNA3C on Pim-1 mediated cell proliferation, HEK-293 cells were transfected with control vector, Flag-tagged EBNA3C, Myc-Pim-1 expression vector, and Myc-Pim-1 with Flag-EBNA3C. Colony formation assays were performed after G418 selection for 2 weeks. The results demonstrated a significant increase in the colony numbers in EBNA3C and the Pim-1 co-transfected set compared with control vector or only Pim-1 transfected sets (Fig. S4A and S4B). Additionally, cell proliferation assays were performed by cell counting using Trypan blue dye exclusion technique up to 6 days (Fig. S4C). Previous studies suggested that Pim-1 expression accelerated the process of lymphoproliferation and inhibits apoptosis [30]. Also, depletion of Pim-1 by RNA interference in mouse and human prostate cancer cells reduced cellular proliferation and survival [69]. To validate these studies, we used Lentivirus mediated delivery of sh-RNA vectors to knock down Pim-1 in LCL1 cells. Wild type LCL1, puromycin selected stable Ctrl-vector and Pim-1 knocked down cells with GFP fluorescence were monitored (Fig. 11A). Also, the expression levels of Pim-1 in different clones were examined by performing Western blot analysis (Fig. 11B). In order to determine whether Pim-1 knockdown in an LCL background has some implications in apoptotic cell death, we performed apoptosis assays using stable sh-Ctrl LCL1, sh-Pim-1 LCL1 cells with or without serum starvation. Cells were stained with Propidium iodide for FACS analysis. The results showed substantial increase in apoptotic cell death in stable Pim-1 knockdown LCL1 with serum starvation (Fig. 11C, D).
Programmed cell death or apoptosis is considered as a major regulator of cellular growth control and tissue homeostasis [70]. Previous reports suggested that caspases activation can be triggered through the induction of the extrinsic apoptotic pathway or at the mitochondria by stimulating the intrinsic apoptotic pathway in response with anticancer chemotherapy [71]. In order to determine whether the inhibition of Pim-1 had some effect on apoptotic event in LCLs, we performed Western blot analysis to monitor the levels of PARP-1 cleavage. Our result showed that Pim-1 knock-down EBV transformed cells showed higher signals for the PARP-1 cleavage (Fig. 11E). Moreover, we detected higher expression levels of Caspase-3, Caspase-9, and Apaf-1 in Pim-1 stable knockdown LCL1 in comparison with sh-Ctrl LCL1 cells which indicates that Pim-1 knockdown induced the intrinsic apoptotic pathway in EBV transformed cells (Fig. 11F). We performed cell proliferation assays in the context of Pim-1 knock-down. Interestingly, the result showed that the rate of proliferation of Pim-1 stable knock-down LCL1 cells was lower compared with LCL1 and sh-Ctrl LCL1 cells (Fig. 11G).
As an inhibitor of cyclin-dependent kinases, p21Waf1/Cip1 is required for proper cell-cycle progression [72]. Earlier reports suggested that p21 suppresses tumors by promoting cell cycle arrest in response to various stimuli [61]. In addition, substantial evidence from biochemical and genetic studies shows that p21 acts as a potential effector of multiple tumor suppressor pathways to promote its anti-proliferative activities independent of p53 [61]. Interestingly, several studies demonstrated that ubiquitin-mediated degradation of p21 can also promote cancer cell proliferation [73]. Our results above indicated that p21 is targeted by EBNA3C through Pim-1 dependent degradation. We next attempted to examine whether EBNA3C has a role in modulating p21-mediated inhibition of cell proliferation involving Pim-1 kinase with DNA damage response. HEK-293 and MEF cells were transfected with different combinations of Flag-tagged p21 (wild type and the T145A mutant), Myc-Pim-1 (wild type and the kinase dead mutant), EBNA3C expression vectors. Cell proliferation assays were performed without serum and with etoposide treatment after 2 weeks of G418 antibiotic selection. The results demonstrated that EBNA3C together with wild type Pim-1 effectively reduced the growth suppressive effect of p21. The cell proliferation rate in p21 expressing cells either with EBNA3C or wild type Pim-1 was shown to be enhanced compared to control vector alone. Interestingly, the lower rate of cell proliferation was observed with kinase dead Pim-1 mutant or P21T145A mutant even in the presence of EBNA3C (Fig. 12A, B). To check the expression levels of these proteins, we performed Western blot analysis with these G418 selected cells (Fig. S5A and S5B). Moreover, our immunofluorescence studies for BrdU incorporation with DNA damage showed an increased number of BrdU foci with wild type Pim-1 and p21, co-expressed cells with EBNA3C (Fig. 12C and 12D). To determine the possible contribution of different molecules which are involved with intrinsic apoptotic signaling in the context of Pim-1 mediated p21 downregulation in the presence of EBNA3C, we checked the protein expression profiles of Caspase-3, Caspase-9, Apaf-1, and Bcl2 in sh-Ctrl-vector transfected and Pim-1 stable knockdown EBV negative Ramos cells. These cells were transfected with p21 and an increasing dose of EBNA3C. Our results demonstrated that the expression levels of Caspase-3, Caspase-9, Apaf-1 were unchanged with P21 transfection in Pim-1 knockdown Ramos cells, in the presence of EBNA3C compared with sh-Ctrl-Ramos cells where the expression of these proteins were reduced (Fig. 12E, compare left and right panels). Interestingly, Bcl2 levels were found to be upregulated in sh-Ctrl cells (Fig. 12E). These results support our hypothesis that EBNA3C can potentiate Pim-1 kinase activities for inhibiting cell growth suppressive property of p21 which occurs through the intrinsic apoptotic pathway.
Pim-1 was identified in murine leukaemia virus (MuLV)-induced lymphomas that frequently contains proviral insertions which were associated with the transcriptional activation of the Pim-1 gene frequently associated with enhanced tumorigenesis [74]. Overexpression of Pim kinases have been found in various lymphomas and leukemias [75]. Different reports have suggested a role for Pim-1 kinase in progression of Burkitt's lymphoma [76], primary cutaneous large B-cell lymphoma [77] and prostate cancer [78]. Pim-1 kinase also performs multiple cellular functions related to cell survival, proliferation, differentiation, apoptosis, and progression of tumors [79].
Previous studies showed upregulation of Pim kinases during Epstein-Barr virus infection [45]. Epstein–Barr virus (EBV) was found potentially involved in the pathogenesis of different B-cell lymphoproliferative disease and all three EBV nuclear antigen 3 proteins can manipulate the expression of a wide range of cellular genes and they often act co-operatively to induce epigenetic chromatin modifications [80]. Another report demonstrated that the involvement of EBNA3A and EBNA3C expression with polycomb complexes for the covalent K27me3 modification of histone H3 at the p16INK4A promoter to repress the transcription [81]. Also, BIM expression was regulated in latently infected EBV cells through epigenetic modification and CpG methylation [11]. EBNA3C, regulates transcription of a wide range of viral and cellular genes [14]. EBNA3C was found to be associated with Nm23-H1 to regulate the transcription process of cellular genes which are critically involved in cell migration and invasion [82]. Recent reports also demonstrated that EBNA3C can physically interact and stabilize different host oncoproteins, including c-Myc and IRF4 [20], [51], and has a major role in regulation of the cell cycle regulatory protein complex Cyclin D1/CDK6 to drive B-cell malignancies [22]. Interestingly, some reports showed that Pim-1 levels are tightly controlled at many steps from the transcriptional to translational levels [60]. Our study now demonstrated upregulation of Pim-1 expression at the mRNA and protein levels with EBV infection in primary B-cells as well as EBV positive cancer cell lines. Interestingly, infection with the ΔEBNA3C BAC-GFP-EBV showed a much lower Pim-1 expression at 2 days post-infection. However, the expression patterns remain unchanged at later time points. Our results from the primary infection studies suggested a major contributory role of EBNA3C in inducing Pim-1 expression. We also observed a substantial reduction in Pim-1 expression levels only after siRNA mediated knockdown of EBNA3C but not with the knockdown of EBNA2, EBNA3A and EBNA3B which further confirmed a direct role of EBNA3C in regulating Pim-1 expression levels in EBV-transformed cells. Moreover, our studies showed that EBNA3C has a strong physical association with Pim-1, and that Pim-1 binds to the N-terminal 130–159 residues of EBNA3C. Interestingly, several other studies from our Lab also demonstrated that this region of EBNA3C specifically interacts with different important cellular proteins such as cyclin A, p53, E2F1, c-Myc, IRF4/IRF8 etc [20], [21], [46], [51], [83]. Therefore, this 130–159 aa residues of EBNA-3C have particular significance in deregulating major cellular process in EBV-infected cells. Further detailed investigation is needed to evaluate the functional role of this domain in connection with EBNA3C mediated oncogenesis. Our co-immunoprecipitation experiments in EBNA3C expressing Burkitt's lymphoma cells and EBV transformed Lymphoblastoid cells also demonstrated that Pim-1 forms a strong molecular complex with EBNA3C in infected cells.
Previous studies suggested that nuclear localization of Pim-1 is essential for the regulation of its cellular substrates as well as additional biological activities of this kinase [76]. Importantly, our co-localization studies showed that in the absence of EBNA3C, localization of Pim-1 was mostly in the cytoplasm and predominantly in the nucleus in the presence of EBNA3C. Our immunofluorescence assay therefore revealed a strong co-localization with Pim-1 and EBNA3C in the nucleus. Previously it was shown that the Hsp90 protein is responsible for correct folding and stabilization of Pim-1 [84]. Further, studies from our group demonstrated an important role of EBNA3C in stabilizing different oncoproteins such as Gemin3, Cyclin D1, and IRF4 [19], [22], [51] to deregulate normal cellular functions which can drive development of neoplastic events. Our study clearly demonstrated that Pim-1 protein stabilization by EBNA3C can result in increased levels of Pim-1 in EBV infected cells. Additionally, the stability of the Pim-1 kinase is largely regulated by the ubiquitin/proteasome pathway [85]. Several reports suggested the important role of EBNA3C for deregulating the functions of different cellular proteins by manipulation of ubiquitin/proteasome pathways [86]. Our Lab previously demonstrated the interaction between EBNA3C with SCFSkp2 E3 ligase complex [87]. Also, the N-terminal domain of EBNA3C physically associated with the acidic domain of Mdm2 which is a known E3 ubiquitin-protein ligase [50]. Other studies also suggested that EBNA3C associates with the α-subunit of the 20S proteasome and is degraded in-vitro by purified 20S proteasomes [88]. EBNA3C was found to facilitate the degradation of E2F1 by targeting ubiquitin-proteasome pathways [46]. Recently, we have shown that EBNA3C deregulates total H2AX levels through involvement of the ubiquitin/proteasome degradation pathway [89]. Interestingly, other reports suggested the potential involvement of Pim-1 with ubiquitin/proteasome pathways as enhanced expression of Pim-1 increases the level of SCFSkp2 ubiquitin ligase through the direct binding and phosphorylation of multiple sites on this protein [90]. A previous study showed the role of heat shock proteins and the ubiquitin-proteasome pathway for regulating the stability of Pim-1 kinase [85]. Our poly-ubiquitination experiments clearly suggested that Pim-1 poly-ubiquitination was significantly inhibited by EBNA3C and so resulted in increased Pim-1 levels. Since enhanced levels of Pim-1 is linked to different hematological or non-hematological malignancies, it reveals the intricate mechanisms that are linked to ubiquitin-proteasome-mediated degradation of Pim-1 and is important for designing therapeutic interventions. Additionally, this approach could enhance new therapeutic avenues by targeting Pim-1 kinase and so enhance the efficiency of conventional therapeutic strategies against EBV mediated oncogenesis.
Being a potent serine/threonine kinase, Pim-1 plays important roles in a number of cellular events. Most notably, Pim-1 can synergize with c-Myc to drive the rapid progression of B-cell lymphomas [91]. This synergism is likely to originate from the anti-apoptotic activity promoted by Pim-1 [92]. Among other Pim-1 substrates, the Cyclin-dependent kinase inhibitor 1 or p21 is important in the context of viral pathogenesis. Interestingly, p21 stability has been exploited by different tumor viruses. A number of viral proteins can affect the post-transcriptional regulation of p21, thereby affecting cellular proliferation. The human papilloma virus E6 proteins can downregulate p21 independently of p53 [93]. Also, the hepatitis C virus and K-cyclin encoded by the human herpesvirus 8 stimulates p21 phosphorylation at the Ser130 residue by CDK6 without affecting its stability [94]. These findings further establish that targeting p21 is likely to be a common strategy for viruses to regulate cell cycle progression and apoptosis. Previous reports also showed that EBV acts downstream of the p53 and appears to prevent the inactivation of cyclin-dependent kinase CDK2 by p21WAF1/CIP1 by targeting p21 for degradation by the proteasome pathway [95]. Basically, participation of p21 in multiple cellular functions emphasizes its importance and that its precise regulation is crucial for maintenance of the normal cellular function. Importantly, its phosphorylation and interaction with other cellular proteins are crucial to p21 stability at the post-translational level. Previous reports suggested that the Thr145 residue of p21 is preferentially phosphorylated by Pim-1 [36]. Our current study clearly demonstrated a role for EBNA3C in enhancing Pim-1 kinase activity to phosphorylate p21. We also observed that Pim-1 was not able to phosphorylate mutant p21 (T145A) even in the presence of EBNA3C. Interestingly, we identified a molecular association between Pim-1 and p21, with EBNA3C and our competitive binding assay demonstrated that increasing doses of EBNA3C resulted in reduced association between Pim-1 and p21, causing in destabilization of p21 by enhancing its proteasome-mediated degradation independent of etoposide induced DNA damage response.
Several reports indicated that Pim-1 expression is associated with cell proliferation and survival [96]. Pim-1 also induces anti-cancer drug resistance by inhibiting the intrinsic mitochondrial apoptosis pathway [97]. In our studies, siRNA mediated knock down of Pim-1 showed reduced proliferation of EBV transformed cells. Moreover, Pim-1 silencing potentially activated the intrinsic apoptotic signaling in EBV transformed cells. Recent studies showed that upregulation of p21 activated the intrinsic apoptotic pathway [98]. Our results also support this finding showing that EBNA3C induced Pim-1 mediated downregulation of p21 which is also related to the inhibition of intrinsic apoptotic pathway in EBV transformed cells. Recent evidence suggested a role for the RNF126 E3 ubiquitin ligase in promoting cancer cell proliferation by p21 degradation [73]. Our results strongly suggested an important role for EBNA3C to effectively inhibit the growth suppressive effects of p21 in the presence of Pim-1. Interestingly, we observed a lower rate of cell proliferation with the kinase dead Pim-1 mutant or P21T145A mutant even in the presence of EBNA3C. This supports a role for Pim-1-mediated phosphorylation of the Thr145 residue of p21 in cell proliferation.
In summary, our current work demonstrated an important molecular mechanism which revealed a direct role for the EBV latent antigen 3C in enhancing expression of the oncoprotein Pim-1 in EBV transformed B-cells as well as in EBV-infected PBMCs. We also showed the physical interaction between EBNA3C and Pim-1 and further mapped the binding to the Amino-terminal domain of EBNA3C. Moreover, our study demonstrated that EBNA3C mediated stabilization of Pim-1 through abrogation of the proteasome/ubiquitin pathway. EBNA3C also facilitated the nuclear export of Pim-1 and promoted EBV transformed cell proliferation by altering Pim-1-mediated regulation of the cell-cycle inhibitor p21/WAF1 activity. Our study now demonstrated that EBNA3C directly contributes to Pim-1 mediated phosphorylation of p21 which facilitates its proteosomal degradation. In addition, significant reduction of EBV transformed cell proliferation as well as a substantial induction of apoptotic cell death was also observed upon stable knockdown of Pim-1. Our study now provides a novel insight into the precise role of oncogenic Pim-1 in EBV-mediated oncogenesis (Fig. 13). Moreover, siRNA mediated knockdown of Pim-1 triggers the intrinsic apoptotic signaling pathway in LCLs and repressed proteasome-mediated degradation of p21. Pim-1 knockdown further demonstrated a vital role in EBV-mediated proliferation of B-cells by impeding the process of apoptosis. Our findings thus contribute to a more indepth understanding of the role of EBNA3C expressed in EBV-infected B-cells and its interaction with the critical cellular kinase which leads to EBV induced B-cell transformation.
PBMC were obtained from University of Pennsylvania Human Immunology Core (HIC) and donated by the healthy donors. This study was approved by University of Pennsylvania Human Immunology Core (HIC) which maintains University of Pennsylvania IRB protocol. In this IRB approved protocol the declarations of Helsinki protocols were followed and each donor gave written, informed consent. There is no link between donors and their information with this study.
Full length and truncated mutants of GST, Myc, Flag, and GFP tagged EBNA3C expression vectors were described previously [22], [51]. Myc-tagged EBNA3C with C143N point mutation was generated by using standard PCR primer mutagenesis method [50]. Constructs for Myc-tagged Pim-1, kinase dead (KD) version of Pim-1 as Pim-1 K67M (mutated at the ATP binding Pocket), pGEX2T-Pim-1 were mentioned previously [43]. Wild type pGEX-P21 construct was generated by using Flag-P21 construct as template. The PCR amplified insert was subjected for EcoRI/NotI restriction enzyme digestion and ligated into pGEX2T vector. pGEX-P21T145A and Flag-P21T145A constructs were cloned by using PBK/CMV/LacZ P21T145A (kindly provided by Dr. Nancy Magnuson) as a template for the PCR amplification. pCDNA3-HA-Ub construct was kindly provided by George Mosialos (Aristotle University of Thessaloniki, Greece) and pGIPZ was used as the sh-RNA vector described previously [51]. Constructs used for lentiviral packaging were previously described [99].
Antibodies of Pim-1 (E-16), Ub (FL-76), PARP-1 (F-2), and GFP (I-16), Caspase-3 (E-8), Caspase-9 p10 (H-83), Apaf-1 (H-324), Bcl2 (C-2) were purchased from Santa Cruz Biotechnology, Inc (Santa Cruz, CA). P21 (ab7960) antibody was purchased from Abcam (Cambridge, MA). GAPDH antibody was procured from US-Biological Corp. (Swampscott, MA). Flag (M2)-epitope, anti-mouse antibody was purchased from Sigma-Aldrich Corp. (St. Louis, MO). Hybridomas for mouse anti-Myc (9E10), anti-Hemaggutinin (12CA5), A10 were previously described [99].
HEK-293 (human embryonic kidney cell line) was kindly provided by Jon Aster (Brigham and Woman's Hospital, Boston, MA, USA). HEK-293, and MEF cells were grown in Dulbeccoo's modified Eagle's medium (DMEM). EBV negative Burkitt's lymphoma cells BJAB, DG75, Ramos were kindly provided by Elliot Kieff (Harvard Medical School, Boston, MA). BJAB stably expressing EBNA3C (BJAB7, BJAB10) were previously described [76]. EBV transformed lymphoblastoid cell lines (LCL1, LCL2) were maintained in RPMI 1640 media
Transfection in HEK-293, MEF and B-cells were performed by electroporation system with Bio-Rad Gene Pulser II electroporator. Cells were electroporated at 210 V and 975 µF (for HEK-293, MEF cells) or 220 V and 975 µF (for DG75, Ramos, LCL1 cells).
PBMCs (Peripheral blood mononuclear cells) were obtained from healthy donors from University of Pennsylvania Immunology Core as mentioned previously [100]. As defined earlier [46], 10 million PBMCs were mixed with wild type and EBNA3C knockout mutant (BAC-GFP-ΔE3C-EBV) virus supernatant in 1 ml of RPMI 1640 media containing 10% FBS for 4 hrs at 37°C and 5% co2 environment. Next, cells were centrifuged at 500×g for 5 minutes and pelleted cells were again re-suspended in 2 ml of complete medium. EBV-GFP expression was checked by fluorescence microscopy and used to evaluate the infection. Infected cells were harvested at specific time intervals to determine the Pim-1 protein and mRNA levels.
For GST pull-down assays, cell lysates from BJAB, BJAB7, BJAB10, LCL1, LCL2 cells were incubated with bacterially purified control GST protein and GST fusion proteins. Protein samples were washed using Binding Buffer (1× PBS, 0.1% NP-40, 0.5 mM DTT, 10% glycerol, with protease inhibitors) and resolved by 10% SDS-PAGE. A10 antibody was used for Western blot analysis.
Cells were harvested and washed with 1× Phosphate Buffered Saline (PBS). For the preparation of cell lysates, RIPA buffer (0.5% NP-40, 10 mM Tris pH 7.5, 2 mM EDTA, 150 mM NaCl, supplemented with 1 mM PMSF, and protease inhibitors) was used. Cell lysates were then pre-cleared with normal mouse/rabbit serum rotating with 30 µl of Protein-A and Protein-G (1∶1 mixture)-conjugated Sepharose beads for 1 hr at 4°C. 5% of the protein lysate was saved as input sample. Appropriate antibody (1 µg/ml) was used to capture the specific protein of interest by rotating the sample overnight at 4°C. The immune-precipitated samples were washed with RIPA buffer. Protein samples were boiled in laemmli buffer [101], resolved by SDS-PAGE and Western blotting was performed. The membranes were probed with appropriate antibodies and scanned using the Odyssey imager (LiCor Inc., Lincoln, NE).
Ice-cold PBS was used to wash the cells prior to RNA isolation. Trizol reagent (Invitrogen, Inc., Carlsbad, CA) was used for RNA extraction according to manufacturer's protocol. Next, Superscript II reverse transcriptase kit (Invitrogen, Inc., Carlsbad, CA) was used for cDNA preparation according to the manufacturer's instructions. The primers for Pim-1, EBNA3C, EBNA2, EBNA3A, EBNA3B were 5′-CGAGCATGACGAAGAGATCAT-3′ and 5′-TCGAAGGTTGGCCTATCTGA-3′ [102], 5′-AGAAGGGGAGCGTGTGTTGT-3′ and 5′-GGCTCGTTTTTGACGTCGGC-3′, 5′- GAACTTCAACCCACACCATC-3′ and 5′- CGTGGTTCTGGACTATCTGG-3′, 5′- GGTGAAACGCGAGAAGAAAG-3′ and 5′- CTCTCATCAGCAGCAACCTG-3′, 5′- AGAAGAGGCCCTTGTGTCTT-3′ and 5′- GGATTTCAAGAGGGTCAGGT-3′ respectively. GAPDH primers were used as 5′-TGCACCACCAACTGCTTAG-3′ and 5′-GATGCAGGGATGATGTTC-3′ [22]. SYBER green Real-time master mix (MJ Research Inc., Waltham, MA) was used for quantitative real-time PCR analysis. To check the specificity of the products a melting curve analysis was performed and the relative quantitation values were calculated by threshold cycle method. Triplicate sets were used to examine each sample.
300 thousand HEK-293 cells were transfected with different expression plasmids by Lipofectamine 2000 transfection reagents (Invitrogen, Carlsbad, CA). Cells were fixed with 3% paraformaldehyde (PFA) with 0.1% Triton X-100 and 1% BSA was used for blocking purpose. Myc-tagged Pim-1 was detected by using anti-Myc (9E10) antibody and the expression of GFP-tagged EBNA3C was detected by GFP-fluorescence. BJAB, BJAB10, and LCL1 cells were semi-air-dried on slides and fixed as mentioned above. Specific antibodies were used to check endogenous expressions of EBNA3C and Pim-1. Nuclear staining was performed by using DAPI (4′,6′,-diamidino-2-phenylindole; Pierce, Rockford, IL). After secondary antibody and DAPI incubation, cells were washed in 1× PBS and mounted with antifade mounting medium. The images were taken by Fluoview FV300 confocal microscope and FLUOVIEW software (Olympus Inc., Melville, NY) was used for image analysis.
HEK-293 cells were transfected with combinations of expression vectors. After 36 hrs of post-transfection, cells were washed with PBS and re-suspended into hypotonic buffer [5 mM Pipes (KOH) pH 8.0, 85 mM KCl, 0.5% NP-40 supplemented with protease inhibitors). Cells were incubated on ice, and Dounce homogenizer was used to homogenize the cells with 20 strokes. Nuclei were pelleted down (2300×g for 5 min) and the cytosolic fraction was collected. Nuclear pellets were washed again with PBS, re-suspended in nuclear lysis buffer (50 mM Tris, pH 8.0, 2 mM EDTA, 150 mM NaCl, 1% NP-40, and protease inhibitors) and lysed by vortexing intermittently for 1 h. The soluble nuclear fraction was separated by centrifugation at 21,000×g for 10 min. To determine the efficiency of nuclear and cytoplasmic fractionation, Western blot analysis was done against the nuclear transcription factor SP1 and cytoplasmic protein GAPDH using specific antibodies.
Expression vectors were transfected by electroporation in HEK-293 cells. Transfected cells were incubated for 36 hrs in fresh DMEM and treated with 20 µM MG132 (Enzo Life Sciences International, Inc., Plymouth Meeting, PA) for another 6 hrs. Protein samples were immunoprecipitated with appropriate antibodies and resolved by SDS-PAGE. The level of ubiquitination was detected by HA-specific antibody (12CA5).
Myc-tagged-Pim-1 (wild type or kinase dead mutant) (5 mg), Flag-EBNA3C (5 mg) expression constructs were transfected in HEK-293 cells. After 36 hrs of post-transfection, cell lysates were prepared and protein complexes were immunoprecipitated (IP) by using 9E10 ascites fluid. IP complexes were then washed with buffer A (containing 25 mM Tris [pH 7.5], 70 mM NaCl, 10 mM MgCl2, 1 mM EGTA, 1 mM DTT, with protease and phosphatase inhibitors) and incubated in 30 ml of kinase buffer B (containing buffer A plus 10 mM cold ATP, and 0.2 mCi of [c-32P]-ATP/ml) supplemented with bacterially purified GST-P21 (wild type or T145A mutant) for 30 min at 30°C. 2× laemmli buffer was added to stop the reaction and heated at 95°C for 10 min. 10% SDS-PAGE was used for resolving the labeled proteins. Quantitation of the band was performed by using Image Quant software (GE Healthcare Biosciences, Pittsburgh, PA).
Transient transfection was performed in 10 million HEK-293 cells using electroporation system with combinations of plasmids as mentioned in the text. After 36 hours transfection, transfected cells as well as B-cells were treated with 40 µg/ml cyclohexamide in specific time periods with DNA damage response and cell lysates were prepared with RIPA buffer. Protein samples were subjected to Western blot analysis. Odyssey 3.0 software was used to quantify the band intensities.
Short-hairpin (sh) oligonucleotides directed against EBNA3C were described previously [22]. The Pim-1 target sequence 5′-GUGUACUUUAGGCAAAGGG-3′ was described previously [43]. sh-oligonucleotides used for EBNA2, EBNA3A and EBNA3B knockdown were 5′- TCGAGTTGTTGACACGGATAGTCTTTCAAGAGAAGACTATCCGTGTCAACAATTTTTTA-3′ and 5′- CGCGTAAAAAATTGTTGACACGGATAGTCTTCTCTTGAAAGACTATCCGTGTCAACAAC-3′, 5′- TCGAGGAACACTTCTTCAAGCGATTTCAAGAGAATCGCTTGAAGAAGTGTTCTTTTTTA-3′ and 5′- CGCGTAAAAAAGAACACTTCTTCAAGCGATTCTCTTGAAATCGCTTGAAGAAGTGTTCC-3′, 5′- TCGAGTTGATTGTCATTGGTTTCATTCAAGAGATGAAACCAATGACAATCAATTTTTTA-3′ and 5′- CGCGTAAAAAATTGATTGTCATTGGTTTCATCTCTTGAATGAAACCAATGACAATCAAC-3′ respectively. EBNA3C and Pim-1 specific oligonucleotides were cloned into pGIPZ vector at XhoI and MluI restriction sites. Control shRNA sequence (Dharmacon Research, Chicago, IL) was used as 5′-TCTCGCTTGGGCGAGAGTAAG-3′ which lacks complementary sequences in the human genome, also cloned in pGIPZ vector. Lentivirus production and transduction of EBV-transformed LCL1 were described previously [51].
10 million Human kidney embryonic cells were subjected to transient transfection with Ctrl-vector, Myc-Pim-1, and Flag-tagged-EBNA3C by electroporation system. Transfected cells were allowed to grow in DMEM containing G418 as 1 mg/ml concentration. After selecting the cells up to 2-weeks, selected cells were fixed with 4% formaldehyde and stained with 0.1% crystal violet solution (Sigma-Aldrich Corp., St. Louis, MO). The area of the colonies was calculated by using Image J software (Adobe Inc., San Jose, CA). The data shown here are average of three independent experiments.
HEK-293, MEF cells were transfected with different combinations of expression vectors by electroporation as described in the text. Transfected cells were grown in DMEM and were selected with 1000 µg/ml G418 antibiotic for 2-weeks. After selection, Cells were incubated without serum and with etoposide (MP Biomedicals, LLC) treatment for 12 hrs. Cell lysates were prepared by RIPA buffer and protein expression was examined by Western blotting. From each transfected and selected set, 0.1×106 cells were plated and allowed to grow them for 6 days. Also, LCL1, sh-Ctrl LCL1 and sh-Pim-1 LCL1 cells were plated and grown in RPMI media. Counting of viable cells at specific time points was performed by using Trypan Blue dye exclusion method. All experiments were performed in triplicates.
HEK-293 cells were transfected with specific plasmid vectors as indicated in the text. After 36 hrs of post-transfection, BrdU was added and incubated cells for 2 hours in the presence of DNA damaging agents. Cells were fixed with 4% paraformaldehyde (PFA) for 15 min in room temperature. Cells were washed with PBS. 2 M HCl was then added and incubated for 20 min at room temperature. Next, 0.1 M sodium borate (Na2B4O7) pH 8.5 was added and incubated for 2 min at room temperature. Cells were washed with PBS and incubated with 0.2% Triton X100, 3% BSA in 1×PBS for 5 min at room temperature. Cells were washed three times with PBS/BSA for 10 min each. Cells were incubated with anti BrdU antibody in PBS/BSA solution. After washing three times with PBS/BSA solution, cells were incubated for 1 hr with secondary antibody. DAPI was added at the final washing steps to stain DNA. The images were observed by Fluoview FV300 confocal microscope.
Data represented here are as the mean values with standard errors of means (SEM). The significance of differences in the mean values was calculated by performing 2-tailed student's t-test. P-value of <0.05 was considered here as statistically significant.
Homo sapiens pim-1- GenBank: M16750.1, Homo sapiens cyclin-dependent kinase inhibitor 1A (p21, Cip1)- GenBank: BC001935.1, Epstein-Barr virus (EBV) genome, strain B95-8- GenBank: V01555.2, human Pim-1 protein- UniProtKB/Swiss-Prot: P11309, human P21 protein- UniProtKB/Swiss-Prot: P38936, EBNA3C protein- UniProtKB/Swiss-Prot: P03204.1.
|
10.1371/journal.pntd.0004704 | Using Rapid Diagnostic Tests as a Source of Viral RNA for Dengue Serotyping by RT-PCR - A Novel Epidemiological Tool | Dengue virus infection causes major public health problems in tropical and subtropical areas. In many endemic areas, including the Lao PDR, inadequate access to laboratory facilities is a major obstacle to surveillance and study of dengue epidemiology. Filter paper is widely used for blood collection for subsequent laboratory testing for antibody and nucleic acid detection. For the first time, we demonstrate that dengue viral RNA can be extracted from dengue rapid diagnostic tests (RDT) and then submitted to real-time RT-PCR for serotyping.
We evaluated the Standard Diagnostics (SD) Bioline Dengue Duo RDT, a commonly used test in dengue endemic areas. First, using the QIAamp RNA kit, dengue RNA was purified from the sample pad of the NS1 RDT loaded with virus isolates of the four serotypes, then quantified by RT-PCR. We observed greater recovery of virus, with a mean of 27 times more RNA recovered from RDT, than from filter paper. Second, we evaluated dengue NS1 RDTs from patients at Mahosot Hospital, Vientiane, (99 patients) and from rural Salavan Provincial Hospital (362 patients). There was good agreement between dengue RT-PCR from NS1 RDT with RT-PCR performed on RNA extracted from patient sera, either using RDT loaded with blood (82.8% and 91.4%, in Vientiane and Salavan, respectively) or serum (91.9% and 93.9%). There was 100% concordance between RDT and serum RT-PCR of infecting dengue serotype.
Therefore, the collection of NS1 positive RDTs, which do not require cold storage, may be a novel approach for dengue serotyping by RT-PCR and offers promising prospects for the collection of epidemiological data from previously inaccessible tropical areas to aid surveillance and public health interventions.
| Dengue fever, caused by a virus transmitted by mosquitoes, is a public health problem in tropical and subtropical regions. Dengue Rapid Diagnostic Tests, in which a drop of blood is loaded onto a paper strip in a plastic cassette, are simple to use and have good diagnostic accuracy. They are becoming the test of choice for the management of dengue epidemics, especially in rural areas without laboratory facilities. However, four types of dengue virus circulate in most tropical areas and their patterns of circulation are of epidemiological importance since they play a role in the severity and propagation of the disease. We show, for the first time, that molecular amplification permitting dengue virus detection and typing can be performed directly from used positive RDTs, that can easily be transported. This novel approach has promising prospects for the collection of dengue epidemiological data from previously inaccessible tropical areas to aid surveillance and public health interventions.
| The dengue virus (DENV) is an enveloped ssRNA flavivirus transmitted by Aedes mosquitoes [1]. Dengue infections are clinically classified by the World Health Organization (WHO) as dengue with or without warning signs and severe dengue [2]. It is an important public health problem affecting the tropical and subtropical world; Bhatt et al. estimate 390 million infections per year, of which 96 million present with clinical symptoms [3]. Approximately 2.4 billion people are currently at risk of dengue infection globally and most live in tropical and urban regions where the four dengue serotypes (DENV-1, 2, 3 and 4) circulate [4]. Secondary infections, which have been reported to be more severe than primary infections, occur when patients are sequentially infected with more than one serotype [5].
The combination of dengue NS1 antigen and anti-dengue IgM detection by ELISA is one of the standard diagnosis strategies, providing high sensitivity and high specificity, covering the viremic phase at the early course of the disease and a later phase when viral RNA is no longer detectable in blood, respectively [6,7]. Gene segment amplification by reverse transcription followed by polymerase chain reaction (RT-PCR) is widely applied for the detection of dengue virus during the viraemic phase, with the advantage of permitting dengue serotyping.
In Lao PDR (Laos), dengue infection is a major cause of morbidity with a rising case fatality rate [8]. Approximately 3.9 million residents are presently at risk of dengue infection [9]. It is usually regarded as an urban disease but recent studies in Laos suggest that it is also an important rural disease [10–12]. However, only limited data are available on dengue epidemiology in Laos as only few institutions, located in the capital city (Vientiane), have access to laboratory facilities required to perform ELISA and RT-PCR and the transportation of frozen specimens is very difficult.
Immunochromatographic Rapid Diagnostic Tests (RDTs), of which a variety of different brands are available for diagnosing dengue, are alternatives for diagnosis in rural areas. They are rapid, accurate, easy to use and do not require advanced technical knowledge or equipment. The Standard Diagnostics (SD) Bioline Dengue Duo RDT (SD dengue RDT; Standard Diagnostics, Kyonggi-do, Korea) permits the concomitant detection of dengue NS1 antigen and anti-dengue IgM and IgG antibodies with overall sensitivity and specificity greater than 80% [13,14]. This dengue RDT remained stable at elevated temperature over 2 years storage in Laos [15]. Dengue RDTs are now used in provincial hospitals and in a few health centers in southern Laos, and are likely to be extended in rural areas.
However, such RDTs do not give information on the infecting serotype, important for both public health surveillance and dengue epidemiology research. We therefore hypothesized that dengue virus could be extracted from NS1 positive RDTs for serotype determination and that such a system could be used for dengue serotype surveillance by the sending of positive RDTs to a central laboratory for RT-PCR. RNA detection from dried blood spot (DBS) for measles, HIV-1, Hepatitis C, dengue and Chikungunya viruses have been described [16–21], but detection of pathogen nucleic acid by PCR from RDTs has only been described for Salmonella enterica serovar Typhi and Plasmodium falciparum [22,23].
We therefore compared techniques for dengue RNA extractions for the four dengue serotypes, followed by RT-PCR, from SD dengue RDTs, filter papers and neat samples. Evaluation was then performed in two clinical cohorts in Laos, in a central and a rural hospital, of patients with suspected dengue.
The SD dengue RDT is an in vitro immunochromatographic assay for the detection of dengue virus NS1 Ag and anti-dengue IgM/IgG antibodies in human serum, plasma, or whole blood, from finger-prick or venous blood. This test comprises a pair of test devices, a dengue NS1 Ag test on the left side, and a dengue IgM/IgG antibody (Ab) test on the right side. Each device contains a strip, enclosed in a plastic cassette. The strip is made of three compartments; i) an absorptive pad where the patient sample (serum, blood or plasma) is applied and then moves along the strip, ii) a conjugate or reagent pad which contains antibodies specific to the target analytic conjugated to colored particles, iii) a nitrocellulose membrane on which the immunocomplexes move until the zone of reaction where they are immobilized and appear as a colored band. The test is easy to perform—three drops (using dropper provided with the kit, ~100 μL) and 10 μL (using a capillary provided with the kit) of sample are applied into the two small wells on the NS1 and Ab cassettes, respectively. Four drops of diluent (provided with the kit) are then applied on the Ab cassette. The test results are obtained in 15 minutes.
Samples were collected at two sites: Mahosot Hospital, a central hospital in Vientiane Capital, and Salavan Provincial Hospital, in a rural area of southern Laos 679 km to the south-east (15.72 N, 106.42 E).
At Mahosot Hospital, 99 consenting patients admitted with symptoms meeting the WHO criteria [2] for dengue infection were enrolled from August to November 2013. At Salavan hospital, 362 consenting patients with undifferentiated fever who tested negative by malaria RDT (SD Bioline Malaria Ag P.f/P.v) were enrolled from July to October 2012. Patient information is displayed in supporting information (S1 Table).
Venous blood alone was collected from patients at Mahosot Hospital. SD dengue RDTs were performed according to manufacturer’s recommendations using whole blood and, after whole blood centrifugation, serum. The dropper provided with the RDT and a micropipette set at 100μl were used to load whole blood and serum, respectively, on NS1 cassettes. Two drops of whole blood and one hundred microliters of serum were loaded on filter paper (FP, Grade 0903 Whatman, GE Healthcare) and 0.2 ml of serum was kept at -80°C as a reference neat serum sample.
At Salavan Hospital, both venous whole blood and capillary whole blood from finger pricks were collected. SD dengue RDTs were performed according to manufacturer’s recommendations using capillary whole blood and, after capillary whole blood centrifugation, serum. Whole blood was directly dropped onto the NS1 cassette and a micropipette was used to load 100μl of serum. Two drops of whole capillary blood were loaded on filter paper. After venous blood centrifugation, 0.2 ml of serum was kept at -20°C as a reference neat serum sample for each patient.
FP and RDT were dried at room temperature for 2 hours, put in individual plastic zip lock bags with desiccant and stored, at room temperature in Salavan and directly into -80°C at Mahosot Hospital until analysis. Samples were shipped from Salavan to the Microbiology Laboratory, Mahosot Hospital, on dry ice for serum and in metals boxes at ambient temperature for FP and RDT once a month. All specimens arriving at Mahosot Hospital were immediately kept at -80°C until testing.
Written informed consent was obtained from all recruited patients or responsible guardians. The patients were recruited in the framework of two studies with ethical approval by the Lao National Ethics Committee for Health Research and the Oxford Tropical Research Ethics Committee (OXTREC).
Sera from dengue patients, infected with one of all four dengue serotypes, admitted at Mahosot Hospital (diagnosed and serotyped by serum RT-PCR) [24] were inoculated on Vero cells as described [10]. After seven days of incubation at 37°C in 5% CO2, virus isolates (DENV-1, 2, 3, and 4) were recovered from the supernatant after centrifugation of cell culture medium. Ten fold serial dilutions using minimum essential medium (MEM, Gibco) were performed for each isolate and aliquots were stored at -80°C for subsequent experiments. One hundred microliters of each dilution were loaded on RDT NS1 cassette and FP, then stored as described above.
RDT strips and FPs, loaded with samples, were cut just before RNA extraction. The RDT NS1 cassette was opened using forceps and the strip taken out. The strip was cut, with a sterile scalpel, into four sections of 7 mm length each, from sample pad (S), conjugate pad (C) and nitrocellulose membrane (two pieces, N1 and N2) (see Fig 1). Subsequently, to improve the quantity of dengue RNA recovered from RDTs, additional experiments were performed by cutting out the whole sample pad (WS), obtaining a section of 15 mm length (Fig 1). Two discs of 6 mm diameter were punched from FP at the middle of the sample spot, using a single hole puncher.
140 μl of each sample of virus isolates and sera were extracted using QIAamp Viral RNA Minikit (QIAGEN AG, Hombrechtikon, Switzerland), following manufacturer instructions (elution in 60μl).
RDT sections and FP discs were processed according to the procedure described for dried swabs in the EZ1 Virus Mini Kit v2.0 handbook (Qiagen). They were incubated for 15 minutes at 56°C with 200μl of ATL lysis buffer (Qiagen) and then 140 μl of the mixture was extracted using the QIAamp Viral RNA Minikit (Qiagen), following manufacturer’s instructions, with 60μl elution volume.
For the detection of dengue RNA after extraction, the pan-dengue Taqman real-time RT-PCR system (DENV All RT-PCR) developed by Leparc-Goffard et al. [24] was used with four serotype-specific RT-PCRs. The SuperScript III Platinum One-Step qRT-PCR kit (Invitrogen) with 200nM of each primer and 100nM of probe on 5μl of RNA extract was used. Synthetic RNA control was prepared as described by Ninove et al. [25]. Three serial dilutions, 2.5 x106, 2.5 x104 and 2.5 x102 copies/μl of positive control were prepared and aliquoted at -80°C and used as standards in each RT-PCR run. All samples and standards were tested by DENV All RT-PCR in duplicate. Means of Ct values of the duplicates were used for the quantification of dengue RNA copies in tested samples (supporting information, S2 Table).
For each of the four serotypes, three virus isolate dilutions (4.3 x104, 4.3 x105 and 4.3 x106 copies/ml) were used to assess the extraction technique. Each of the three dilutions, for the four serotypes, was loaded in triplicate on RDTs and on FP (supporting information, S1 Fig). All RDT and FP samples underwent separate extraction along with 140μl of the 12 virus dilutions as comparators. All extracts underwent DENV All RT-PCR for RNA quantification. The mean number of dengue RNA copies was calculated from triplicate extractions, with the relative standard deviation (RSD), to assess the reproducibility of the technique. Dengue RNA extractions from RDTs and FPs were compared to direct extraction by dividing the number of copies obtained from RDTs and FPs by the number of copies obtained by the direct extraction of virus solution. This was expressed as a percentage of RNA recovery (multiplying the ratio by 100); 100% indicating that the number of RNA copies obtained after RDT or FP extraction was the same as from direct extraction.
The techniques developed in this study for RNA preparation from RDT and filter paper were compared to the direct RNA extraction from neat serum. In the absence of gold standard, outcomes (dengue RT-PCR results) are presented in a 2x2 table and agreements (95% confidence intervals) were calculated, as recommended by US FDA [26], using Stata v10 [27]. The agreements of the different RNA preparation techniques were then compared using the z test.
Dengue RNA was detected after extraction from all the 4 sections of the NS1 RDT strip, even for the isolates with the lowest dengue copy concentration (4.3 x104 copies/ml). The recovery of dengue RNA from RDTs was less efficient than the direct extraction of the virus isolate (Table 1). RDT-S extraction permitted recovery of 7 to 49% of the quantity of RNA recovered by direct extraction and was much more efficient than the extraction from the other RDT sections (Fig 2). The C, N1 and N2 RDT sections permitted recovery of 1–16%, 1–12% and 1–14%, respectively, of the quantity of RNA recovered by direct extraction. Extraction from the S section permitted the best reproducibility with the lowest RSD from 3 to 86% whereas the RSD for C, N1 and N2 sections were 13–135%, 19–145% and 13–173%, respectively.
To improve dengue RNA extraction from NS1 RDTs, the Whole S pad (WS, 15mm) was tested. The quantities of DENV RNA copies recovered from RDT-WS were higher than from RDT-S for all 4 DENV serotypes for all DENV isolate dilutions. The extractions from RDT-WS permitted recovery of 34%-169% of the quantity of RNA recovered by the direct extraction, in contrast to 7% to 49% for RDT-S extraction (Table 2). Therefore, extraction from RDT-WS was selected for subsequent experiment using patient samples.
The RNA recovery from FP was much less efficient with only 2 to 6% of the quantity recovered by the direct extraction (Table 2, Fig 3). On average, 27 times less dengue RNA copies were recovered from FP than from RDT-WS.
These results suggest that dengue serotype can be determined by PCR of the NS1 pad of one brand of dengue RDT, which is an potentially useful tool for the large populations without access to laboratory facilities.
Prado et al. [17] and Matheus et al. [20] reported detection of dengue virus by RT-PCR from dried blood spots on filter paper. The former study tested dengue 2 and dengue 3 viruses from FP from 52 patients and the latter tested FP from 666 NS1 positive patients. Here we found similar results with good overall agreement between neat serum and FP extraction either loaded with serum or whole blood from patients at Mahosot Hospital. On evaluation under field conditions, 362 patient samples collected on filter paper from Salavan, stored for 1 month at room temperature (18 to 46°C) [15], also showed good overall agreement of 90.3% between filter paper extraction and neat serum extraction. The extraction technique used has the advantage of being a simple commercial kit, without phenol. However, the use of Trizol remains a good alternative for laboratories with limited resources.
Although, P. falciparum and S. Typhi DNA detection by PCR has been achieved from RDTs, [22,23,28,29] to the best of our knowledge virus detection by RT-PCR from RDTs has not been described. Studies on Plasmodium DNA detection from RDTs showed that different components of the strip demonstrated variable suitability for nucleic acid purification. Cnops et al. [29] found the nitrocellulose membrane to be the most suitable area whereas Veron and Carne [28] and Ishengoma et al. [22] found that it was the sample pad. For SD dengue RDT our data show that the sample pad area from NS1 cassette is the best section for dengue RNA detection. The efficiency in RNA recovery from the full WS part (15mm) was close to what was obtained by direct neat sample extraction (34% to 169% RNA recovery). This shows the importance of assessing the optimal RDT section to be used for PCR. Whether this varies between dengue NS1 RDT brands remains to be determined. Interestingly, the RNA recovery from RDTs was 27 times more efficient than from 2 discs (6mm) of filter paper. The evaluation on patient samples from a central and a provincial hospital showed good overall agreements between neat sera and RDT extraction for all conditions (82.8% to 93.9%) with no significant differences when using RDT or filter paper, loaded with blood or serum, for the detection of DENV by RT-PCR (all comparison of agreements p>0.05). However, for NS1 positive patients from Salavan, better agreement was observed for RDT in comparison to filter paper. Some patients were found negative by PCR from neat serum and positive from RDT or filter paper, 42 patients from Salavan and only 4 from Mahosot. This difference is probably due to the suboptimal storage at -20°C of sera in Salavan. In addition, RDTs and filter papers from individual patients were kept in individual zip lock bag to avoid contamination but we can not exclude that this process was not strictly followed and that contamination between bloody RDTs and filter papers could have happened.
Although the patient populations we tested were infected by all four serotypes, patients from Mahosot Hospital were mainly infected by DENV-3 (83.3%), reflecting the DENV-3 outbreak occurring at that time in Vientiane [30] and those from Salavan mainly by DENV-1 (88.9%). Therefore, it was not possible to evaluate potential differences in RDT and FP dengue detection according to serotype. Additional studies are needed to better assess the effect of temperature storage in the efficiency of RNA recovery from RDT. Moreover, this study was performed in Salavan Provincial Hospital where staff are familiar with sample collection for testing in the central hospital. Study at other provincial and district hospitals, and eventually health centers, over a longer period would be useful to assess the sustainability of this strategy. And finally it would be important to test this process using other dengue RDT brands to see if this technique could be generalized.
Dengue RDTs are becoming important diagnosis tools in dengue epidemic management and is the only diagnostic test, when any are available, in provincial hospitals and health centers in Laos. It is expected that their use will be extended into more remote areas [15]. Therefore, positive dengue RDTs could be, in Laos and elsewhere in rural Asia, appropriate devices for sample storage and easy transportation to higher-level facilities for dengue serotype RT-PCR determination. RDTs and dried blood spots (DBS) should be considered as potentially infectious and thus handled appropriately [31]. One might imagine that the collection of used RDT could become a standard procedure, however this would require additional experiments to assess if RDT could be modified, as by pre or post chemical treatment, to improve RNA preservation and recovery. Although filter paper is of low cost and easily distributed it would not be needed where dengue RDTs are used for routine diagnosis. Used NS1 RDTs could be collected and transported for batched RT-PCR by national surveillance programs.
This technique may also permit dengue envelope sequencing for deeper molecular epidemiology analysis from RNA purified from RDTs. This could greatly increase availability of dengue epidemiological data from previously inaccessible tropical areas by facilitating dengue confirmation tests and strain identification to aid surveillance and public health interventions. This will also be of considerable importance if dengue vaccines are introduced. In addition, negative dengue RDTs could also be evaluated for PCR for other viruses causing similar clinical syndromes, such as chikungunya and zika viruses, to aid in differential diagnosis. As RDTs become increasingly used for a diversity of diseases, further exploration to look at what ‘added value’ could be extracted could be important for public health.
|
10.1371/journal.pntd.0003601 | Exposure to Leishmania braziliensis Triggers Neutrophil Activation and Apoptosis | Neutrophils are the first line of defense against invading pathogens and are rapidly recruited to the sites of Leishmania inoculation. During Leishmania braziliensis infection, depletion of inflammatory cells significantly increases the parasite load whereas co-inoculation of neutrophils plus L. braziliensis had an opposite effect. Moreover, the co-culture of infected macrophages and neutrophils also induced parasite killing leading us to ask how neutrophils alone respond to an L. braziliensis exposure. Herein we focused on understanding the interaction between neutrophils and L. braziliensis, exploring cell activation and apoptotic fate.
Inoculation of serum-opsonized L. braziliensis promastigotes in mice induced neutrophil accumulation in vivo, peaking at 24 h. In vitro, exposure of thyoglycollate-elicited inflammatory or bone marrow neutrophils to L. braziliensis modulated the expression of surface molecules such as CD18 and CD62L, and induced the oxidative burst. Using mCherry-expressing L. braziliensis, we determined that such effects were mainly observed in infected and not in bystander cells. Neutrophil activation following contact with L. braziliensis was also confirmed by the release of TNF-α and neutrophil elastase. Lastly, neutrophils infected with L. braziliensis but not with L. major displayed markers of early apoptosis.
We show that L. braziliensis induces neutrophil recruitment in vivo and that neutrophils exposed to the parasite in vitro respond through activation and release of inflammatory mediators. This outcome may impact on parasite elimination, particularly at the early stages of infection.
| Leishmania is the parasite responsible for the disease leishmaniasis, present in all continents. Leishmania parasites are spread through infected sand-flies and, during transmission into the vertebrate host, neutrophils are among the first cells to arrive at the infection site. Since neutrophils are key players at the frontline of defense against invading organisms, we investigated their response to Leishmania braziliensis. Importantly, L. braziliensis causes both Cutaneous and Mucocutaneous Leishmaniasis, two clinical manifestations characterized by their chronic development and by the presence of skin lesions with tissue destruction. Upon inoculation of mice with L. braziliensis, neutrophils rapidly arrive at the site of infection. We then observed that culture of mouse neutrophils with L. braziliensis induced the expression of adhesion molecules, production of Reactive Oxygen Species and secretion of elastase and TNF-α, two important inflammatory mediators. Also, infection with L. braziliensis induced neutrophil apoptosis, a cell death mechanism key for regulating inflammation. Our results show that neutrophils respond to presence of the L. braziliensis parasites by becoming activated and undergoing apoptosis. We suggest that this outcome modifies the local environment at the site of parasite inoculation and thus contributes with parasite killing in the infected host.
| Neutrophils are essential components of the early inflammatory response, acting as the first line of defense against invading pathogens (rev. in [1]). Neutrophil recruitment to the infection site occurs in response to various stimuli and is followed by cell rolling and adhesion to the vasculature, processes mediated by interactions between selectins and integrins [2]. Pathogen phagocytosis subsequently elicits the production of superoxide, which is quickly dismutated into hydrogen peroxide and other secondary Reactive Oxygen Species (ROS), which are highly toxic to the invading pathogen [3]. Phagocytosis stimulates the secretion of additional antimicrobial molecules such as neutrophil elastase, into the phagosome further contributing with pathogen killing [4]. Resolution of inflammation requires efficient removal of apoptotic neutrophils by professional phagocytes such as resident macrophages [5]. Phagocytosis of apoptotic neutrophils prevents the release of potentially toxic molecules and, in parallel, regulates the inflammatory response [6].
During experimental Leishmania infection, neutrophils play distinct roles depending on the combination of mouse strain and parasite species. For L. donovani and L. infantum, neutrophils contributed to parasite killing [7,8]. For L. major, neutrophil and monocyte depletion enhanced disease in resistant mice [9–12] whereas, in susceptible mice, the absence of neutrophils inhibited Th2 cell development [12]. Neutrophil depletion led to faster lesion development in mice infected with L. amazonensis promastigotes [13] whereas amastigotes displayed resistance to the neutrophil microbicidal machinery [14]. Following phagocytosis, some Leishmania spp can be found within non-lytic compartments [15]. This evasion strategy suggests that Leishmania parasites may exploit neutrophils as to gain access to macrophages where, ultimately, infection is established [16,17]. In vivo, neutrophils readily arrive at the site of L. major [12,18] and L. infantum-chagasi inoculation [19] within minutes. Employing a natural transmission model, Peters et al. showed that neutrophils capture L. major parasites at the site of sand fly bite, but the parasites remain viable [20]. In this model, the absence of neutrophils was unfavourable to infection. More recently, Ribeiro-Gomes et al. showed that in experimental infection, the route of inoculation (intradermal, subcutaneous or intraperitoneal) also impacts on the capture of L. major parasites by neutrophils and on the establishment of infection [21].
Previously, we showed that L. braziliensis-infected macrophages co-cultured with live neutrophils display a reduced parasite load [22]. This outcome was dependent on the interaction between macrophages and neutrophils and was associated with the production of TNFα and superoxide. We suggested that clearance of neutrophils in L. braziliensis-infected mice promotes a pro-inflammatory environment, contributing with parasite clearance. Herein we investigated how exposure to L. braziliensis and internalization or not of the parasite impacts the neutrophil response.
Female BALB/c mice, 6–8 weeks of age, were obtained from CPqGM/FIOCRUZ animal facility where they were maintained under pathogen-free conditions. All animal work was conducted according to the Guidelines for Animal Experimentation of the Colégio Brasileiro de Experimentação Animal and of the Conselho Nacional de Controle de Experimentação Animal. The local Ethics Committee on Animal Care and Utilization (CEUA) approved all procedures involving animals (L-03/2011).
L. braziliensis promastigotes (strain MHOM/BR/01/BA788) [23] or transgenic L. braziliensis parasites expressing mCherry [24], kindly provided by Phillip Scott (University of Pennsylvania), were grown in Schneider’s insect medium (LGC) supplemented with 10% FBS, 2 mM glutamine, 100 U/ml penicillin, and 100 mg/ml streptomycin. Parasite cultures were seeded at 105 parasites/mL and were closely monitored to ascertain that parasites had reached the stationary phase (7 days). Before co-culture experiments with neutrophils, stationary-phase parasites were opsonized with 5% heat-inactivated fresh naïve serum for 30 min at 24°C. Metacyclic enriched promastigotes were obtained as described elsewhere [25]. In some experiments, we employed L. major (WHOM/IL/80/Friedlin) or dead parasites, prepared as described [26].
BALB/c mice were inoculated in the ear dermis with 106 stationary phase promastigotes, in 10μL, using a 27 1/2G needle. Control mice were injected with serum-free DMEM medium. After 6, 24 and 48h post-inoculation, mice were sacrificed and the dorsal and ventral ear sheets separated with forceps. The two leaflets were transferred to RPMI supplemented with 10% FCS and antibiotics. After 16h the cells emigrating out of the ear explants were collected, counted and stained for flow cytometry [27]. For cell surface molecules, mAb 24G2 was used to block FcRs and cells were stained using anti-Ly6G-APC/Cy7 (clone 1A8) (BioLegend) and anti-CD11b-eFluor 450 (clone M1/70) (eBioscience). All cell events were acquired on an LSRII flow cytometer (BD Biosciences) and analyzed using FlowJo (Tree Star, Inc.).
Peritoneal neutrophils (herein referred as inflammatory neutrophils) were obtained by i.p. injection of 10% thioglycollate (SIGMA), as described [28]. Cells were collected 18 h later, by peritoneal washings, counted, and were left to adhere for 1h at 37°C. Non-adherent cells were recovered, washed and examined for purity by both FACS and H&E staining of cytospin preparations. Neutrophils of purity >90% were used in experiments. Bone marrow neutrophils were obtained from the tibia and femur of mice; labeled with neutrophil-specific mAbs anti-Ly6G (clone NIMP-R14-FITC or clone-1A8-PE) (BD PharMingen) and purified by MACS-positive selection, using using anti-FITC or anti-PE magnetic beads (Miltenyi Biotech). Alternatively, neutrophils were labeled with anti-Ly6G (clone 1A8, conjugated to Biotin, Miltenyi Biotech) and purified using anti-Biotin magnetic beads. Purity of neutrophils following either NIMP-R14 or 1A8 positive MACS selection was >95%, as assessed by FACS. Control stainings with CD11b and Ly6C were performed following magnetic separation and neutrophils (inflammatory and bone marrow) were characterized as CD11b+1A8+Ly6Cint and Gr1high (S1 and S2 Figs, respectively).
Inflammatory neutrophils or bone marrow neutrophils were cultured for 2h, in RPMI medium supplemented with 10% FCS, 100 U/ml of penicillin and 100 μg/ml of streptomycin (all from Invitrogen), in the presence or absence of serum-opsonized L. braziliensis (at a 2:1 parasite:cell ratio). The infection rate of inflammatory or bone marrow neutrophils co-cultured with L. braziliensis-expressing mCherry was determined by flow cytometry. Data were acquired on a Fortessa or an LSRII flow cytometer (BD Biosciences) and analyzed using FlowJo (Tree Star Inc.).
Neutrophils were co-cultured L. braziliensis, as described, for 2 h. For cell surface staining, neutrophils were incubated with FcBlock (CD16/CD32) (BD Pharmingen) followed by anti-CD18-FITC (clone M18/2) or anti-CD62L-PE (clone MEL-14) (all from E-bioscience, including isotype control Rat IgG2a). For the detection of Reactive Oxygen Species, cells were co-cultured with L. braziliensis for 2 h and were later stained with Dihydroethidium (DHE) (Invitrogen), a superoxide indicator, at 3 μM for 30 minutes. As a positive control, neutrophils were incubated with phorbol 12-myristate 13-acetate (PMA) (100nM) (SIGMA) for 30 minutes. Data were acquired with a FACSAria or FACScan (BD Biosciences) and analyzed with FlowJo (Tree Star Inc.).
Neutrophils were co-cultured with L. braziliensis, as explained above, for 4h. Elastase enzymatic activity was measured as described [10]. Briefly, cell culture supernatants were harvested and added (20 μL) in triplicate to ELISA plates. Following addition of Elastase reaction buffer (55μL) (0.1 M HEPES, 0.5 M NaCl, 10% dimethylsulfoxide, pH 7.5) and 0.2 mM Elastase substrate I (MeOSuc-AAPV-pna; Calbiochem) (150 μL), samples were incubated at 37°C for 3 days. Elastase activity was determined by reading absorbance at 410 nm, using serial dilutions of human elastase (Calbiochem), as standards. For the detection of TNF-α, neutrophils were co-cultured with L. braziliensis for 24h. Cell culture supernatants were collected and TNF-α levels were determined by ELISA, using a commercial kit (R&D Systems).
Inflammatory neutrophils were co-cultured with L. braziliensis or L. major (at a 5:1 parasite:cell ratio) for 18 h. Neutrophils were then stained with Annexin V-FITC and PI (both from BD Biosciences). Bone marrow neutrophils were co-cultured with mCherry-L. braziliensis, as described above, for 18 h. Apoptotic neutrophils were obtained by ultraviolet irradiation exposure (245nm) for 10 minutes [10]. Cells were stained with Annexin V-FITC (Biolegend) and DAPI (SIGMA) and apoptosis was assessed by flow cytometry. Data were acquired with a FACSAria or FACScan (BD Biosciences) and analyzed with FlowJo (Tree Star. Inc.).
Inflammatory neutrophils were co-cultured with L. braziliensis or L. major (at a 5:1 parasite:cell ratio) for 18 h. Cells were fixed with 2% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.4, and post-fixed in 1% OsO4 and 0.8% potassium ferricyanide and 5 mM calcium chloride in the same buffer. Cells were dehydrated in a graded series of acetone and embedded in Poly/Bed 812 (Polysciences, Inc.) resin. Ultrathin sections were stained with uranyl acetante and lead citrate and examined on a Zeiss109 transmission electron microscope operating at 80 KV.
The significance of the results was calculated using non-parametrical statistical tests [Mann Whitney (two-sided t-test) or Kruskal-Wallis followed by Dunn’s post test]. Analyses were conducted using Prism (GraphPad software) and a p-value of <0.05 was considered significant.
Previously, we reported that neutrophils are present throughout the course of lesion development in BALB/c mice inoculated with L. braziliensis [22]. Herein, we initially evaluated the kinetics of neutrophil recruitment at the early moments following L. braziliensis infection. Mice were inoculated in the ear dermis with L. braziliensis parasites and recruited cells were selected based on size and granularity; within this population, we defined Ly6G+ neutrophils (Fig. 1A). Six hours following parasite inoculation, we did not see differences in the number of Ly6G+ neutrophils comparing mice inoculated with L. braziliensis and control mice, inoculated with saline (Fig. 1B). Twenty-four hours later, the number of Ly6G+ cells recruited to the inoculation site significantly increased in experimental mice and 48h later, this number decreased (p<0.05, compared to the 24 h time-point). These results show that neutrophil recruitment peaks one day after L. braziliensis inoculation.
Following the observation that neutrophils rapidly accumulate at the site of L. braziliensis inoculation (Fig. 1), we investigated the expression of molecules important for cell rolling such as CD62L (L-selectin) and adherence and transmigration such as CD18 (β2 integrin). We also employed serum-opsonized L. brazilensis since Leishmania is delivered into the host in a blood pool, where promastigotes likely encounter serum and the complement system. Inflammatory neutrophils were co-cultured with serum-opsonized mCherry L. brazilensis and neutrophils were selected by size and granularity and, subsequently, by expression of Ly6G (Fig. 2A). In parallel, we also compared infected neutrophils (Ly6G+mCherry+) and bystander neutrophils (Ly6G+mCherry-), the latter defined as neutrophils that remained uninfected in spite of exposure to L. braziliensis (Fig. 2A). Following co-culture of inflammatory neutrophils with L. braziliensis, the percentage of infected (mCherry+) neutrophils was approximately 36% (Fig. 2B) whereas 49% of cells remained uninfected (mCherry-). The percentage of CD18+ cells among infected (mCherry+) neutrophils was higher (p<0.01) in comparison to bystanders (mCherry-) (Fig. 2B) whereas in control cultures (not exposed to neutrophils) the percentage of CD18+ cells was very low (Fig. 2B). In non-exposed neutrophils, the percentage of CD62L+ cells was high (Fig. 2C) and differently from CD18, the percentage of CD62L+ cells was lower (p<0.05) among infected (mCherry+) neutrophils compared to bystanders (mCherry-) (Fig. 2C). Incubation of inflammatory neutrophils with Zymozan did not significantly alter the percentage of CD18+ cells (S3 Fig).
To expand on these findings, we performed experiments with bone marrow neutrophils, which, comparatively have an enhanced capacity to become primed [29]. Bone-marrow neutrophils were also selected by size, granularity and Ly6G expression (Fig. 3A) and following co-culture with L. braziliensis, the percentage of mCherry+ neutrophils was approximately 48% (Fig. 3A) whereas 43% of cells were mCherry-. As with inflammatory neutrophils (Fig. 2), the percentage of CD18+ cells was also very low in control non-exposed cultures and significantly higher (p<0.05) among infected (mCherry+) neutrophils compared to bystanders (mCherry-) (Fig. 3B). Also replicating our findings with inflammatory neutrophils (Fig. 2), the percentage of CD62L+ cells was highest in non-exposed neutrophils (Fig. 3C), and significantly higher (p<0.05) in bystanders (mCherry-) compared to infected (mCherry+) neutrophils. These data indicate that neutrophils infected with L. braziliensis upregulate CD18 and downregulate CD62L, regardless of their activation state.
Neutrophils produce Reactive Oxygen Species (ROS), which form a central component of the defense mechanism against foreign pathogens during infection. Inflammatory neutrophils co-cultured with L. braziliensis displayed a significant increase in superoxide production (Fig. 4A), which was attributed mostly to neutrophils harboring L. braziliensis-mCherry. With bone marrow neutrophils, superoxide production was also significantly higher in cells harboring mCherry, however, ROS was also observed in bystanders (mCherry-) (Fig. 4B). Additionally, superoxide production was similar upon co-culture of inflammatory neutrophils with either stationary phase or metacyclic L. braziliensis (Fig. 5A and B). Co-culture with dead parasites also did not change superoxide production in relation to neutrophils cultured alone (Fig. 5A and B). Similar results were obtained regarding the percentage of CD18+ cells (Fig. 5C and D).
Neutrophils display granules enriched with antimicrobial molecules, including serine proteases such as elastase [30]. Additionally, cytokines secreted by neutrophils, such as TNF-α, influence macrophage and dendritic cell function, with important effects on the adaptive immune response [28]. Herein, co-culture with L. braziliensis, triggered the release of elastase by both inflammatory (Fig. 6A) and bone marrow neutrophils (Fig. 6B). In the same manner, the presence of TNF-α was significantly higher in cultures of inflammatory (Fig. 6C) and bone marrow neutrophils (Fig. 6D) co-cultured with L. braziliensis. We did not detect IL-10 nor IL-12p40 in the culture supernatants.
At infection sites, cells dying by apoptosis express phosphatidylserine (PS) and PS exposure can be detected by Annexin V staining and quantified by flow cytometry. Inflammatory neutrophils were co-cultured for 18 h with L. braziliensis parasites and we investigated whether this interaction resulted in apoptosis. In these co-cultures, there was a significant (p<0.05) increase in the percentage of early apoptotic (Annexin V+/PI-) neutrophils (Fig. 7A and B), compared to neutrophils cultured alone, whereas the percentage of late apoptotic/necrotic (Annexin V+/PI+) neutrophils was similar (Fig. 7A and B). On the other hand, upon co-culture with L. major, we detected a lower percentage of both early (Annexin V+/PI-) and late apoptotic/necrotic (Annexin V+/PI+) neutrophils, compared to neutrophils cultured with L. braziliensis (Fig. 7A and B). Analysis of neutrophils by transmission electron microscopy confirmed apoptosis of L. braziliensis-infected neutrophils as seen by the presence of pyknosis, chromatin condensation as well as remnants of internalized degenerated parasites (Fig. 8). Moreover, internalized L. braziliensis parasites presented chromatin condensation, cytoplasmic disorganization and vacuolization (Fig. 8). In co-cultures performed with L. major, however, the neutrophils remained with a well preserved cytoplasm and viable parasite were observed inside the parasitophorous vacuole (Fig. 8), reinforcing the finding that L. major delays neutrophil apoptosis [31], differently from L. braziliensis.
Following the observation that L. braziliensis induced apoptosis in inflammatory neutrophils, we then examined whether this would also occur with bone marrow neutrophils. Indeed, upon co-culture with L. braziliensis, a significant increase in the percentage of late apoptotic/necrotic (Annexin V+/DAPI+) neutrophils (Fig. 9A and B) was observed. As a control of late apoptosis/necrosis, neutrophil exposure to UV increased the percentage of cells positive for Annexin V+/DAPI+. Importantly, late apoptosis/necrosis (AnnexinV+/DAPI+) was mostly detected in infected neutrophils (mCherry+) when compared with bystanders (mCherry-) (Fig. 9C). At this time point, the percentage of mCherry+ neutrophils was ~28% (Fig. 9D).
Numerous studies have demonstrated that neutrophils play a crucial role in immunity against bacterial, fungal [1] and intracellular pathogens [32]. Earlier on, we demonstrated that L. braziliensis inoculation into the ear dermis of BALB/c mice leads to the development of a cutaneous ulcer, which heals spontaneously after ten weeks of infection [23]. Additionally, co-inoculation of L. braziliensis and neutrophils decreased lesion size whereas depletion of neutrophils and monocytes had an opposing effect, significantly increasing parasite load and lesion size [22]. Given that neutrophils are among the first cells to encounter the parasite at the site of the sand fly bite [20] and, thus, will readily encounter Leishmania parasites, the purpose of the current study was to investigate how neutrophils respond to L. braziliensis exposure, evaluating neutrophil activation and downstream events such as apoptosis.
Sand flies probe the human host to obtain blood and, in this process, lacerate capillaries forming a blood pool into which Leishmania promastigotes are inoculated. Following this event, there is rapid accumulation of neutrophils [20] and it has been shown that the co-inoculated salivary molecules can modulate neutrophil function [33,34]. Herein, we confirmed neutrophil infiltration to the site of L. braziliensis inoculation by syringe and showed maximal accumulation at 24h. Of interest, syringe inoculation of L. amazonensis, also induced maximal neutrophil accumulation at 24 h [13], indicating a common kinetic for neutrophil recruitment for these two New World Leishmania species.
Following our observation that neutrophils are recruited in response to L. braziliensis inoculation, we then performed a series of in vitro experiments to investigate how neutrophils respond to this type of stimulation and, in addition, we compared the responses of inflammatory and bone marrow neutrophils. Initially, we evaluated the expression of adhesion molecules. β2 integrins are leukocyte-specific integrins required for neutrophil adhesion and transmigration across the activated endothelium [35] and CD18 is the common β2 integrin present in LFA1 (CD11aCD18), Mac-1/ CR3 (CD11bCD18) and p150/94/CR4 (CD11cCD18). In the presence of L. braziliensis we detected an increase in the percentage of neutrophils (inflammatory and bone marrow) expressing CD18 and this increase was associated with infected neutrophils (mCherry+), indicating that L. braziliensis were readily internalized. Indeed, Mac-1/CR3 (CD11bCD18) plays a major role in the phagocytosis of complement-opsonized L. major promastigotes by both macrophages [36–38] and human neutrophils [39].
L-selectin (CD62L) participates in neutrophil tethering and rolling [40] but it is cleaved from the leukocyte surface following cellular activation and exposure to inflammatory stimuli [41,42]. Upon co-culture with L. braziliensis, we detected a lower percentage of inflammatory CD62L+/mCherry+ neutrophils, compared to bystanders (mCherry-). Similar results were obtained with bone marrow neutrophils, indicating that L. braziliensis phagocytosis induced more CD62L shedding, marking neutrophil activation [43]. With regards to bystanders, the percentage of inflammatory CD62L+/mCherry- neutrophils was lower compared to bone marrow (CD62L+/mCherry-) neutrophils, possibly reflecting their already primed nature and their extravasation to the peritoneum following thyoglycollate stimulation. Such difference may also be related to the priming potential of bone marrow neutrophils vs. inflammatory, as shown by fMLP stimulation and induction of ROS [29]. Also, we cannot presently attribute CD62L shedding to the infection rate since mCherry staining was similar for both bone marrow (~48%) and inflammatory (~36%) neutrophils. Although we do not know which molecules may be activating bystander neutrophils, it has been shown that L. amazonensis LPG activates human neutrophils in levels similar to those observed with promastigotes [44].
In the presence of L. braziliensis, both inflammatory and bone marrow neutrophils displayed a significant increase in the production of superoxide, a hallmark of neutrophil activation, and ROS detection was significantly higher in infected (mCherry+) neutrophils. Similar results were obtained in experiments with other Leishmania spp. [14,44–46]. Neutrophils exposed to ROS also up-regulate the production of TNF-α and MIP-2 [47,48] and TNF-α primes murine neutrophils to become activated, an effect that is concomitant with the mobilization of CR3-containing granules to the plasma membrane [49]. Since TNF-α and CD18 expression were increased upon neutrophil-co-culture with L. braziliensis, we can suggest that ROS produced by infected cells contributed with TNF-α secretion and CD18 (a Mac1/CR3 component) expression. Furthermore, elastase production was also elevated in neutrophils cultured with L. braziliensis and, importantly, elastase was associated with the killing of intracellular Leishmania in macrophages cultured with neutrophils [10], a process dependent on TLR4 signaling [50]. IL-10 production, on the other hand, was not modulated in our experiments, as seen in previous studies [14,51].
Cell death and the subsequent clearance of apoptotic neutrophils is crucial for maintaining homeostasis and, at the same time, necessary for resolution of inflammation. At inflammatory sites, neutrophils can undergo spontaneous apoptosis [52] or apoptosis due to the recognition of cell-death mediators such as TNF-α and FasL [53]. Co-culture with L. braziliensis induced neutrophil apoptosis, findings that were confirmed by transmission electron microscopy analysis. Indeed, infected neutrophils displayed condensed chromatin and degraded intracellular parasites. Similar results were obtained with bone marrow neutrophils: Annexin+/DAPI+ staining was significantly higher in infected cells (mCherry+) compared to bystanders (mCherry-) and the percentage of infected neutrophils (mCherry+) was lower compared to bystanders (mCherry-). We can suggest that phagocytosis of L. braziliensis results in apoptosis and, in parallel, parasite destruction, hence the lower percentage of infected cells. In addition, ROS [54] and TNF-α [55] also trigger neutrophil apoptosis, two mediators that were produced upon culture with L. braziliensis.
Neutrophil apoptosis was also observed upon culture of neutrophils with L. amazonensis [14] but L. major, on the other hand, delays neutrophil apoptosis [31], enhancing cell lifespan [56]. Parasites survive within infected neutrophils [39,57] and viable parasites have been recovered by cell sorting [20]. Indeed, in our hands, the frequency of late apoptotic (Annexin+/PI+) staining was low in neutrophils cultured with L. major in contrast to neutrophils cultured alone and to neutrophils cultured with L. braziliensis, both of which were positive for Annexin/PI (Fig. 6). Electron microscopy confirmed the presence remnants of L. braziliensis parasites while in contrast, intact parasites were found within L. major infected neutrophils (Fig. 7). Clearance of apoptotic neutrophils by macrophages promotes parasite replication in vitro [58], indicating that L. major may exploit neutrophil apoptosis as means to ascertain infection. Moreover, the phagocytosis of apoptotic neutrophils inhibits the response to L. major [59]. Therefore, for L. major, current literature indicates that neutrophils are rapidly and massively recruited to the site of Leishmania inoculation, where they phagocytose the parasites. Depending on the source of neutrophils, species and strains of Leishmania, internalized parasites can survive and neutrophils would thus provide a transient safe shelter prior to parasite entry into macrophages, the definitive host cell (rev. in [17,60]). In experiments with L. braziliensis, however, co-culture of infected macrophages with UV-treated neutrophils did not modulate the parasite load [22], also suggesting that differences within Leishmania species may induce distinct outcomes regarding neutrophil apoptosis and downstream effects.
We showed that neutrophils are recruited to the site of L. brazilensis inoculation and upon contact with promastigotes, in vitro, neutrophils become activated producing superoxide, TNF-α and elastase. Later, we observed neutrophil apoptosis, particularly of infected cells. However, once amastigotes become predominant, a different scenario may ensue since this stage is more resistant to these same effector mechanisms, as recently described for L. amazonensis [14], impacting on disease development. Indeed, BALB/c mice infected with L. braziliensis develop cutaneous ulcers, despite the presence of neutrophils [23]. However, in this experimental model lesions heal spontaneously and parasites are eliminated from the infection site. Neutrophils could also play a role at the chronic stages of infection, through cooperation with L. braziliensis-infected macrophages, as previously shown in vitro [22]. Neutrophils have been shown to cross-talk with dendritic cells [27,61,62] and such cross talk may also be related to the development of the adaptive immune response to L. braziliensis. However this remains to be investigated. Thus, the strong impact of L. braziliensis on neutrophils phenotype and function reported here in vitro are likely to occur at the onset of infection with the parasite, suggesting that these cells are playing a crucial role following infection.
|
10.1371/journal.pgen.1007678 | A synonymous RET substitution enhances the oncogenic effect of an in-cis missense mutation by increasing constitutive splicing efficiency | Synonymous mutations continue to be filtered out from most large-scale cancer genome studies, but several lines of evidence suggest they can play driver roles in neoplastic disease. We investigated a case of an aggressive, apparently sporadic medullary thyroid carcinoma (MTC) harboring a somatic RET p.Cys634Arg mutation (a known MTC driver). A germ-line RET substitution (p.Cys630=) had also been found but was considered clinically irrelevant because of its synonymous nature. Next generation sequencing (NGS) of the tumor tissues revealed that the RET mutations were in cis. There was no evidence of gene amplification. Expression analysis found an increase of RET transcript in p.Cys630=;p.Cys634Arg patient compared with that found in 7 MTCs harboring p.Cys634 mutations. Minigene expression assays demonstrated that the presence of the synonymous RET mutation was sufficient to explain the increased RET mRNA level. In silico analyses and RNA immunoprecipitation experiments showed that the p.Cys630 = variant created new exonic splicing enhancer motifs that enhanced SRp55 recruitment to the mutant allele, leading to more efficient maturation of its pre-mRNA and an increased abundance of mature mRNA encoding a constitutively active RET receptor. These findings document a novel mechanism by which synonymous mutations can contribute to cancer progression.
| Synonymous mutations—once considered “silent” because they do not alter the gene product’s amino-acid sequence—are now emerging as potential drivers of cancer. Our recent investigation of an aggressive medullary thyroid carcinoma (MTC) revealed a novel mechanism that could underlie such effects. The MTC analyzed harbored a somatic p.Cys634Arg mutation of the RET protooncogene (a well-known MTC driver). A second RET substitution (p.Cys630=) discovered in the germline had been considered clinically irrelevant because of its synonymous nature. Our next-generation sequencing analysis of the patient’s cancer tissues revealed that the RET mutations were in cis. We also found that RET mRNA levels in patient’s MTC were significantly higher than those found in seven other MTCs with various amino-acid substitutions at position 634 in RET. Subsequent experiments demonstrated that the presence of the synonymous mutation created new exonic splicing enhancer motifs in the mutant allele, which led to more efficient maturation of its transcript and increased expression of a constitutively active RET receptor.
| It is now clear that synonymous mutations—single-nucleotide substitutions that do not alter the amino acid encoded by the affected codon—can play functionally relevant roles in human disease [1]. Over three decades of research have shown that these mutations can affect protein synthesis and/or function by interfering with a host of cellular mechanisms ranging from pre-mRNA splicing to protein folding [2]. Despite these advances, synonymous mutations are still largely ignored in most studies of cancer, and they continue to be filtered out from large-scale analyses of cancer genomes [3]. However, a growing body of evidence indicates that these mutations might actually play driver roles in human cancer [3,4]. Two recent studies provide evidence for the causal involvement of synonymous mutations in melanoma [5], as well as other cancer types [6].
This paper describes a series of molecular studies we conducted to explore the role played by a synonymous substitution affecting the RET proto-oncogene [MIM: 164761] in an aggressive case of medullary thyroid carcinoma (MTC [MIM: 155240]; http://www.omim.org). MTC is a relatively rare neuroendocrine calcitonin-secreting tumor derived from the parafollicular C-cells of the thyroid. Hereditary forms, including multiple endocrine neoplasia types 2A and 2B (MEN2A [MIM: 171400]; MEN2B [MIM: 162300]), in which MTC is associated with other endocrinopathies, are almost invariably associated with germ-line point mutations in RET. The membrane tyrosine kinase receptor encoded by this gene is activated by the binding of ligand—co-receptor complexes, which induces dimerization of RET proteins and autophosphorylation of intracellular tyrosine residues [7]. Gain-of-function point mutations frequently occur in RET’s cysteine-rich extracellular domain (e.g., those involving codon 634) or intracellular kinase domain (e.g., mutations at codon 918 or 804). Both result in inappropriate activation of the tyrosine kinase receptor—the former by inducing ligand-independent, disulfide-bonded homodimerization of the RET proteins, the latter by causing these proteins’ constitutive autophosphorylation [8].
Correlation has been observed between specific mutations and features of the hereditary MTC phenotype, including age at onset, tumor aggressiveness, and the presence of other endocrine tumors [9]. Somatic gain-of-function RET mutations appear to drive around half of all sporadic MTCs [10–12], but the genotype—phenotype correlations in these cases are less clear-cut [13].
The case we analyzed involved an apparently sporadic MTC harboring a somatic RET c.1900T>C mutation [p.Cys634Arg] in a previously healthy 29-year-old man (patient ID0110M). The postoperative course was characterized by rapid disease spread. Despite multiple surgical interventions and systemic vandetanib therapy [14], the patient died approximately 3 years after diagnosis (Supplemental Note). Several clinical features of this case (e.g., age at onset, aggressive disease behavior [15]) were difficult to explain solely on the basis of the somatic RET p.Cys634Arg driver mutation identified by Sanger sequencing. Shortly before the patient’s death, we therefore began re-analyzing DNA and RNA samples from his primary and metastatic tumors using next-generation sequencing (NGS) to identify any somatic mutations in other cancer-relevant genes. No other known somatic driver mutations were found, but the results redirected our attention to the germ-line RET c.1890C>T [p.Cys630=] substitution, which had been identified by the preoperative Sanger sequencing analysis and considered clinically inconsequential because of its synonymous nature. Our NGS data showed that the synonymous substitution was in cis with the somatic RET p.Cys634Arg driver mutation, and subsequent experiments demonstrated that it exerted appreciable effects on RET pre-mRNA splicing.
Using the Ion Torrent PGM NGS platform, we simultaneously sequenced 409 cancer-related genes in DNA from the patient’s preoperative blood specimen and formalin-fixed paraffin embedded (FFPE) samples of the primary and metastatic lesions. Coverage statistics and the variants identified in each tissue are reported in S1 and S2 Tables, respectively.
As shown in Table 1, only a somatic mutation with an allele frequency (AF) ≥ 10% was shared by all four neoplastic tissue samples analyzed. This was a missense substitution (c.1900T>C [p.Cys634Arg]) in RET exon 11. AFs approaching 50% in all tissues (primary tumor: 37.1%; level-IV lymph node metastasis: 43.4%; lung metastasis; 32.6%; cervical lymph node metastasis: 43.4%) were consistent with mutation clonality and the absence of gene amplifications. The latter conclusion was also supported by Sanger sequencing data, which confirmed the presence of the p.Cys634Arg mutation in all four tissues and documented complete overlap between the mutant and wild-type (WT) peaks (Fig 1A–1D). NGS and Sanger sequencing findings also corroborated the presence in these tissues of the synonymous RET c.1890C>T (p.Cys630=) substitution documented in the germ line preoperatively (Fig 1E). The discovery that this substitution was in cis with the somatic p.Cys634Arg mutation led us to re-consider its possible role in the disease.
Interrogation of the GnomAD database (http://gnomad.broadinstitute.org/) revealed the p.Cys630 = to be a very low-frequency single nucleotide variant (SNV), AF in the European population, 0.003% [16]. However, the c.1890C>T substitution involves codon 630 in RET exon 11, a common site of non-synonymous RET mutations known to be pathogenic (e.g., C630R, C630Y) [9]. Supek et al. maintain that synonymous mutations are significantly enriched in oncogenes, particularly those harboring activating nonsynonymous mutations, and they recurrently alter exonic splicing enhancer or silencer (ESE and ESS, respectively) motifs, causing intron retention, exon skipping, or aberrant forms of alternative splicing [6]. These considerations led us to explore the potential effects of the germ-line p.Cys630 = substitution on the quality of the RET transcript.
We used RT-PCR to analyze RET mRNA quality of the region that includes exons 10, 11, and 12 in an FFPE sample of patient ID0110M’s primary MTC and a fresh-frozen sample of the cervical lymph node metastasis. The resulting amplicons were identical in length to those observed in healthy control (WT pool) and in a fresh-frozen sample of a MTC harboring a different RET mutation (c.1832G>A [p.Cys611Phe]) (Fig 2A). These findings argue against the presence of macroscopic alterations involving exon 11 in the RET transcript (e.g., retention in the mature transcript of an intronic region between exons 10 and 11, exon 11 skipping), although we cannot exclude the possibility of alternatively spliced transcripts that had already undergone nonsense-mediated decay (NMD) [17]. Sanger sequencing revealed no sequence alterations within the exon 10–11 junction in cDNA from our patient’s tumors (Fig 2B).
As shown in Fig 3, RET mRNA levels in patient ID0110M’s MTC were significantly higher than those found in seven other MTCs with various amino-acid substitutions at position 634 in RET, including two with the missense p.Cys634Arg mutation present in patient ID0110M’s tumor. Sanger sequencing analysis of the patient’s cervical lymph node metastasis revealed no mutations within the 1000-bp region preceding the RET start codon or in the 3′ UTR of RET transcript (S3 Table). Biologically relevant alterations involving distant enhancer regions cannot be excluded (although published data linking such mutations to MTC are currently lacking).
On the whole, these findings suggest that the p.Cys630 = substitution may exert quantitative effects on RET mRNA.
To further explore if the p.Cys630=;Cys634Arg quantitative effect on RET mRNA was translated in a more abundant RET protein, we used Western blotting to assess RET protein levels associated with the p.Cys630=;Cys634Arg phenotype (Fig 4).
Compared with a nodal metastasis from a p.Cys634Arg MTC (control), the lymph node lesion with the p.Cys630=;Cys634Arg phenotype displayed increased expression of RET protein (Fig 4A), and the latter effect was associated with increased phosphorylation of RET’s downstream target ERK (Fig 4B). Patient ID0110M’s tumor cells (with the p.Cys630=;Cys634Arg phenotype) thus appeared to contain higher levels of RET mRNA, and this alteration resulted in more abundant activated RET protein than that found in a p.Cys634Arg MTC.
To explore the causal role of the p.Cys630 = substitution in the increased abundance of mRNA RET transcript, we used the minigene approach, an important tool for the identification and in vivo analysis of regulatory elements that affect precursor RNA maturation [18]. As shown in Fig 5A–5C, we created four RET minigenes, each consisting of exons 10, 11, and 12 plus flanking intronic sequences with fundamental consensus splicing motifs (donor, acceptor, branch sites). Minigenes containing the WT RET sequence or one of the mutant sequences (p.Cys630=, p.Cys634Arg, or p.Cys630=;Cys634Arg) were transfected into HeLa cells, which do not endogenously express RET (S1A Fig). Twenty-four hours later, cells were harvested, and levels of immature and mature RET minigene transcripts were measured using two different methods, RT-PCR and real-time PCR. Expression levels of immature transcript from the four minigenes were similar to one another, but substantial differences were noted in mature transcript levels (S1B Fig). We also compared the minigenes in terms of their mature/immature RET transcript ratios. For this analysis, we considered the exon 10–11 and exon 11–12 junctions in mature minigene transcripts separately. As shown in Fig 6A, mature exon 10–11 junction transcript levels for the p.Cys630 = minigene were ~3-fold higher than those of the WT minigene, and the increase was even more substantial for the p.Cys630=;Cys634Arg minigene (~9-fold higher than WT transcript levels, P = 0.03). Similar effects were observed when we analyzed mature transcript levels for the exon 11–12 junction (Fig 6B).
Transcript levels for both minigenes harboring the p.Cys630 = substitution were always higher than those for the minigenes without this substitution. While these findings do not allow us to draw any conclusions on the possible effects on RET transcript levels of the p.Cys634Arg mutation, they strongly suggest that the presence of the p.Cys630 = substitution alone is sufficient to explain the increased abundance of mature RET transcript seen in Fig 6A and 6B. Moreover, the results of both analyses suggest a synergic effect of p.Cys630 = and p.Cys634Arg variations.
Collectively, these data suggested that the increase might be mediated by effects exerted on the maturation of minigene pre-RNA or the stability of the minigene transcripts. To explore the latter possibility, we treated the minigene-transfected HeLa cells with the transcription inhibitor, actinomycin D and monitored immature and mature transcript levels over time. No differences were observed between the four minigenes in terms of their immature or mature transcript half-lives (Fig 6C and 6D). It therefore seems more likely that the p.Cys630 = variant impacts RET expression by influencing the efficiency of maturation of RET pre-mRNA.
The location of the synonymous p.Cys630 = substitution at the beginning of RET exon 11 (nucleotide 11) suggested that its effects on RET mRNA levels might be mediated by changes involving the splicing process. To explore this possibility, we used four in silico platforms to analyze the WT and p.Cys630=;Cys634Arg-mutated sequences of RET exon 11 for potential splice sites, potential branch points, and ESS and/or ESE sequences. ESEs are recognized by SR proteins, highly conserved, structurally related splicing factors that promote exon definition, directly (by increasing recruitment of the splicing machinery) or indirectly (by antagonizing the action of splicing silencers) [19]. As shown in Table 2, all four tools predicted that the p.Cys630 = substitution would create one or more new ESE motifs (recognized by SRp55 or less commonly by SF2/ASF) within exon 11 or enhance the SR protein-binding affinity of one or more ESEs already present in WT exon 11. No such changes were predicted for the region containing the p.Cys634Arg mutation (which affects nucleotide 21 of exon 11), and none of the tools predicted the creation or elimination of ESSs or any other alterations within the mutated sequence. These data suggest that the synonymous RET p.Cys630 = substitution increases the binding affinity of the SRp55 and SF2/ASF proteins for the mutant allele pre-mRNA, possibly enhancing recruitment of the splicing machinery to exon 11.
To assess the above hypothesis, we used RNA immunoprecipitation (RIP) to evaluate the SR protein-recruiting capacities of the immature RET minigene transcripts. Nuclear extracts from minigene-transfected HeLa cells were immunoprecipitated with an antibody specific for SRp55 (the SR protein most likely to recognize the new motifs, according to in silico predictions—see Table 2). Total SRp55-bound RNA was then isolated, reverse-transcribed, and analyzed by real-time PCR to quantify the presence of immature RET transcript. As shown in Fig 7, the immunoprecipitated RNA was strongly enriched for immature RET p.Cys630 = and RET p.Cys630=;Cys634Arg transcripts (488- and 561-fold, respectively, compared with the IgG-precipitated control). These findings do not directly demonstrate SRp55’s causal role in the increased levels of mature RET transcript. However, they are fully compatible with the hypothesis that the new SRp55-binding ESE motifs created by the synonymous p.Cys630 = substitution enhanced splicing machinery recruitment to exon 11.
Synonymous variants continue to be filtered out from pipelines used to analyze data from large-scale cancer genome/exome studies [20]. Consequently, they are under-reported, and their potential to drive cancer has been underestimated. Recent evidence shows, however, that these variants can indeed play functionally relevant roles in neoplastic disease [3,6,21]. Supek et al. [6] found them to be significantly over-represented in oncogenes, as compared with both cancer-unrelated genes and tumor-suppressor genes, and up to half of all oncogene-associated synonymous variants appear to be under selection. They also reported a strong association between synonymous variants in oncogenes and aberrancies involving alternative splicing, reflecting the ability of these substitutions to generate, alter, and/or eliminate exonic splicing regulatory motifs. To validate their in silico predictions, the authors used minigene expression assays to assess the effects of 12 synonymous oncogene substitutions on exon inclusion/skipping patterns in their mature transcripts. Six of the 12 were found to affect alternative splicing by creating new ESEs and/or eliminating ESSs. These changes led mainly to diminished exon skipping, which in some cases caused increased expression of full-length isoforms of the encoded protein and, consequently, enhanced oncogene activity [6].
The synonymous RET p.Cys630 = substitution found in our patient’s germ-line DNA involved a constitutive exon (exon 11), i.e., one fundamental for the functional integrity of the encoded protein. The complex multifactorial process regulating constitutive splicing has thus evolved to ensure that these essential exons are unambiguously recognized by the splicing machinery and included without fail in the mature transcript [22]. Not surprisingly then, mutations altering the sequences of cis-elements that contribute to this optimal regulatory set-up (including ESEs and ESSs) have consistently been found to cause pre-mRNA maturation that is dysfunctional [19,20,23,24].
These findings prompted our original search for qualitative alterations in RET mRNA in patient ID0110M’s tumor tissues. Instead, the tissues contained correctly spliced mature RET transcript but at higher levels than those found in control tumors. Consistently, in silico analyses did not predict that the presence of RET p.Cys630 = in exon 11 would reduce or eliminate ESE motifs in the pre-mRNA transcript from the mutant allele: instead, it was expected to create de novo ESEs and enhance the binding affinity of others for SR protein splicing regulators. These findings were also in line with our in vitro demonstration, compared with its WT counterpart, immature RET p.Cys630 = minigene transcript binds more SRp55, a protein implicated in spliceosome recruitment rather than in transcription regulation [25,26]. Therefore, instead of provoking exon-11 missplicing, the RET p.Cys630 = substitution increased levels of correctly spliced mature transcript from the mutant allele, possibly by increasing the number of molecules of pre-mRNA entering in the splicing pathway.
We used RET minigenes to investigate the synonymous substitution’s effect on exon 11 splicing at both its 5′ and 3’ ends. We chose to reduce intron size to more easily incorporate RET exons 10–12 into our splicing reporter construct. This design appears to have diminished the overall splicing efficiency compared with that achieved during in vivo transcript maturation, thereby facilitating our demonstration of the synonymous substitution’s effect on RET minigene maturation. Our in vitro findings led us to exclude the possibility that the p.Cys630 = substitution affects transcription or the stability of immature or mature forms of RNA. These data strengthen our suspicion that the variant’s effects were mediated by its impact on pre-RNA splicing, although they cannot exclude the involvement of other molecular mechanisms in vivo.
Patient ID0110M’s germ-line RET p.Cys630 = substitution occurred in cis with a somatic missense RET mutation (p.Cys634Arg), a known driver of MTC. It is difficult to speculate on the potential clinical impact of the p.Cys630 = substitution in the absence of p.Cys634Arg mutation (particularly in light of the substantial gaps in our patient’s family history [Supplementary Note]). It is reasonable to assume, however, that a genetic variant with no effects on the amino-acid sequence of the RET protein would not generate a constitutively activated RET receptor. In that case, RET signaling levels would remain dependent on ligand availability. In terms of its biological impact, the higher levels of mature transcript produced by the p.Cys630 = substitution-bearing allele may resemble those generated by gene amplification. Ciampi et al. reported RET amplifications in both familial and sporadic MTCs, where their presence was generally correlated with relatively poor clinical outcomes. However, with rare exceptions, these amplifications were found in tumors that also harbored a RET mutation. The authors therefore excluded the possibility that RET amplification is an alternative mechanism of RET gene activation, suggesting instead that it might exert potentiating effects during the transformation and/or progression of RET-mutated MTCs [27].
This conclusion is in line with the results of our analyses. The presence of the synonymous p.Cys630 = substitution led to more efficient maturation of transcript from the mutant RET allele, which also harbored the somatic p.Cys634Arg mutation. Expression of p.Cys630=;Cys634Arg allele was thereby increased (at both the transcript and protein levels) over levels that would be expected in the absence of p.Cys630=. The result was an excess of RET p.Cys634Arg monomers, which are capable of ligand-independent dimerization and autophosphorylation of the RET receptor. Such effects could conceivably augment the p.Cys634Arg mutation’s transforming capacity, but on the basis of currently available data, we cannot draw any meaningful conclusions on this possibility. Moreover, the tumor’s transforming potential could also be influenced by multiple other mechanisms, genetic (i.e., CNV or mutations in distant regulatory regions) and/or epigenetic, and we cannot exclude their involvement in our patient’s disease on the basis of the experiments conducted thus far.
Although the mechanism we hypothesize has not been previously described, it may be more common that it currently appears, given the high frequency of synonymous substitutions in oncogenes harboring activating non-synonymous mutations and their ability in this setting to increase the number of ESE motifs [3,6]. Further study is mandatory, but the analyses we have conducted thus far shed new and interesting light on the complex regulation of splicing.
All experiments performed included positive and negative controls, and commercial products were used according to the manufacturer’s instructions.
Research has been approved by our Institutional Review Board (Comitato Etico dell’azienda policlinico Umberto I). Study title: “Beyond RET: identification of the new genes involved in RET wild-type medullary thyroid cancer using Next Generation Sequencing based approach”, (Rif. 3367/25.09.14).
Written informed consent was obtained from all donors for use of their samples in genetic studies.
Neoplastic tissues were obtained with written informed consent from patients undergoing surgery for MTC at the Umberto I Hospital (Sapienza University of Rome). Two pathologists confirmed the neoplastic phenotype of all tissues based on standard histopathological criteria. Tumors were staged according to the criteria of American Joint Committee on Cancer [28].
Surgical samples of neoplastic tissue and peripheral venous blood were collected from patient ID0110M. The blood sample was used for the preoperative mutational analysis of RET. Sanger sequencing and NGS were used for postoperative mutational analysis of FFPE samples of the patient’s primary tumor and metastases (one to the lung, one to a level IV lymph node obtained before vandetanib treatment, one to a cervical node obtained after vandetanib treatment). FFPE samples of the primary tumor and cervical node metastasis were used to asses qualitative changes in RET mRNA. All FFPE were pathologist-identified samples of tumor (rather than stromal) tissue. Fresh-frozen tissue from the patient’s cervical node metastasis was used to assess qualitative changes in RET mRNA and protein expression. The results of both analyses were compared with those for fresh-frozen tissues from MTCs used in previous studies (21) and a non-neoplastic control pool of RNA from 64 normal thyroids (Clontech). RET protein data on patient ID0110M’s MTC were compared with those on an archived fresh-frozen sample of a lymph node metastasis from RET [p.Cys634Arg] MTC. Due to the paucity of the tissues, we performed a single protein extraction. All fresh-frozen neoplastic tissue samples had pathologist-confirmed tumor cell contents > 80%.
RET mutations were identified by Sanger sequencing of DNA from peripheral blood and FFPE samples of the primary and metastatic tumor tissues. Blood DNA was isolated using Qiagen’s QIAmp-Blood MIDI Kit; FFPE-tissue DNA using NucleoSpin Tissue Kit (Macherey-Nagel GmbH & Co.). RET mutational analysis was also performed on cDNA from a cervical lymph-node metastasis and from the primary tumor. Total RNA was isolated from a fresh-frozen tissue sample of the nodal metastasis using Trizol reagent (Thermo Fisher Scientific) and from FFPE tissue sample of the primary tumor using RecoverAll Total Nucleic Acid Isolation kit (Ambion). After DNase treatment, the RNA was used to generate cDNA with the High Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific). Sanger sequencing analyses of DNA and cDNA were performed as previously described [29], using the primers shown in S4 Table.
We performed NGS analysis on DNAs from peripheral blood and FFPE primary and metastatic tumor tissues. Using the Ion AmpliSeq Comprehensive Cancer Panel (Thermo Fisher Scientific), we sequenced 409 cancer driver genes. We amplified 40 ng of DNA by PCR using four premixed primer pools (Ion AmpliSeq Comprehensive Cancer Panel) and Ion AmpliSeq HiFi mix (Ion AmpliSeq Library Kit 2.0). The multiplexed amplicons were treated with a FuPa reagent to partially digest primer sequences and phosphorylate the amplicons. Adapters were ligated to digested amplicons using the Ion AmpliSeq Library Kit 2.0. To clonally amplify each DNA fragment onto the IonSphere Particles (ISPs), we performed emulsion PCR of each library on an Ion One Touch2 Instrument. Templated-ISPs were then isolated using the Ion One Touch Enrichment System. Sequencing was performed with the Ion PGM Sequencing 200 Kit version 2, using two 318 chips for each DNA sample.
We analyzed data using Variant Caller v4 (Thermo Fisher Scientific). Variant caller format files were annotated with Ion Reporter 4.0 (Thermo Fisher Scientific) and wANNOVAR. Variants were called when a position was covered at least 100 times. We set the clinical sensitivity of point mutations and indels at 10%. Variant prioritization was based on population frequency, quality values, and functional consequences. Variants were filtered based on their frequency among the European-descendent population from the 1000 Genomes Project (http://www.internationalgenome.org), ESP6500SI (evs.gs.washington.edu), and ExAC datasets (http://exac.broadinstitute.org) and on clinical associations (ClinVar database) (https://www.ncbi.nlm.nih.gov/clinvar). Rare variants were defined as those with a minor allele frequency (MAF) < 0.5%. Variants classified by ClinVar as “not-pathogenic,” “probable-not-pathogenic,” “drug response,” or “other” were excluded. High-quality variants were those with a depth of coverage (DP) of ≥100, genotype quality (GQ) scores of ≥30, a minimum alternate allele frequency of 10% (AF≥10%), and absence of both strand bias (SAF/SAR ≠ 0) and homopolymer regions (HRUN<6). Finally, variants were prioritized based on their genomic location, with exclusion of intronic, intergenic, ncRNA-intronic, and UTR variants. Synonymous variants were excluded by the default functional filter. Exonic, splicing, stop-gain, stop-loss, and frameshift insertion and deletion variants were retained for further evaluation.
Total RNA was isolated from FFPE slides with NucleoSpin tissue (Macherey Nagel), from fresh-frozen tissues with Trizol reagent (Thermo Fisher Scientific), and from transfected cells with RNeasy Mini Kit (Qiagen). After DNase treatment and purification (precipitation with ethanol and sodium-acetate), RNAs were quantified with Nanodrop 2000 (Thermo Fisher Scientific) and used to synthesize cDNA with the High Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific). Qualitative changes in RET mRNA and RET expression in tissue samples and minigene-transfected cells were assessed with RT-PCR (performed with AmpliTaq Gold [Thermo Fisher Scientific] according to a standard protocol) and real-time PCR. The RETcDNAF8 and RETcDNAR12 primers used for RT-PCR assays are reported in S4 Table. Minigene expression in transfected cells was assessed by RT-PCR with Immature F2 and Immature R1 primers for amplification of immature minigene transcript and MatureF1 and RET rs11 primers for mature transcript (S1A Fig). Amplification was repeated 25 times for the immature transcripts and 40 times for mature transcripts. Primer sequences are in S4 Table.
Real-time PCR analyses were done with a 7900HT Fast Real-Time PCR System, and SDS 2.3 software (both from Thermo Fisher Scientific) was used to calculate Ct values [30]. TaqMan Universal Master Mix was used for quantitative analyses with TaqMan Gene Expression Assays-on-Demand using a standard protocol. Results were calculated using the 2−ΔΔCt method and normalized to the calibrator sample. In particular, TaqMan Gene Expression Assays-on-Demand Hs01120021_m1 and Hs99999905_m1 (Thermo Fisher Scientific) were used to quantify RET and GADPH (reference control) expression, respectively, in MTC samples. RET expression in patient ID0110M was used as the calibrator. The TaqMan Gene Expression Assay-on-Demand Hs01120021_m1 (Thermo Fisher Scientific) was used to quantify mature minigene exon 11–12 junction transcript levels. Immature minigene transcript levels were quantified with Taqman Gene Expression Assay AIWR4LA (Thermo Fisher Scientific) designed with the Taqman Gene Expression design tool (https://www.thermofisher.com/order/custom-genomic-products/tools/gene-expression/) and the minigene sequence as a template. The assay targeted the ligation sites (i.e. XhoI restriction sites, see “RET minigene cloning strategy and plasmid isolation” paragraph). Thermo Fisher Scientific certifies the high amplification efficiency (of 90–110%) of all Taqman gene expression assays. Immature minigene transcripts encoded by minigenes were used as reference control and the WT sample (ratio mature/immature) as a calibrator.
The SensiMix SYBR kit (Bioline) was used with specific primer pairs to quantify minigene expression levels and the half-lives of immature and mature minigene transcripts using a standard protocol. A standard curve was used to assess primer efficiency. Slopes, Y-intercepts, and R^2 values were -3.836, 22.087, and 0.975 for Immature F2 and Immature R1 and -3.253, 111.48, and 0.645 for Mature F1 and RETrs11. Levels of immature and mature minigene transcripts in transfected HeLa cell extracts were expressed as nanograms of cDNA, the ratio of mature to immature transcript levels, or percentage of mature transcripts.
Custom Taqman Gene Expression Assay AIWR4LA (Thermo Fisher Scientific) was used to quantify immature minigene transcript levels in RNA immunoprecipitate and IgG-precipitated extracts (negative controls). Results were normalized to input samples and expressed relative to IgG control.
Total proteins were extracted on ice from fresh-frozen tissues and quantified using the Bradford method. The lysis buffer contained TrisHCl (pH 7.4, 50 mM), NaCl (150 mM), Triton (1% v/v), EDTA (20 mM), phenylmethylsulfonyl fluoride (2 mM), protease and phosphatase inhibitors (Pierce), leupeptin (2 μg/ml), and glycerol (10% v/v). For Western blot analysis, 30 μg of proteins separated by SDS-PAGE, transferred to polyvinylidene fluoride membranes, and probed with the following primary antibodies (all from Cell Signaling Technology and all used at 1:1000 dilution): anti-RET (C31B4), anti-Tubulin (sc-8035), anti-Erk1/2 (cat.# 137F5), and anti-phospho-ERK1/2 (Thr202-Tyr204) antibody (cat.# 4370). Secondary antibodies (anti-mouse SC-2005 and anti-rabbit SC-2004) were HRP-conjugated and used at 1:5000 dilution. Membranes were incubated with ECL reagent (Clarity Western, Bio-rad), and chemiluminescence was quantified with Bio-Rad’s Chemidoc MP Imaging system (Bio-rad). Band intensity was analyzed with the system’s Image Lab Software and the results normalized to that of controls (tubulin bands for RET; ERK bands for phospho-ERK bands). Experiments were repeated twice.
We used PCR to amplify three RET gene fragments containing exon 10, exon 11, or exon 12, each with flanking intronic sequences. Fragments were amplified using AmpliTaq Gold (Thermo Fisher Scientific). Primer sequences are shown in S4 Table (RETex10_F, RETex10_XhoI_R, RETex11_XhoI_F, RETex11_XhoI_R, RETex12_XhoI_F, RETex12_SalI_R). Fragment 1, which included exon 10 plus the first 23 nucleotides (nt) of intron 10, and fragment 3, which comprised the last 99 nt of intron 11 and all of exon 12, were both amplified from patient ID0110M’s DNA containing the WT RET allele. Fragment 2 contained the last 128 nt of intron 10, all of exon 11, and the first 20 nt of intron 11. Four versions were created: one with the WT sequence (amplified from the DNA used for fragments 1 and 2); the second with the p.Cys630 = substitution (amplified from germ-line DNA from patient ID0110M); the third with the p.Cys634Arg mutation (amplified from DNA isolated from an MTC harboring this mutation); and the fourth containing both of the latter variations (amplified from somatic DNA isolated from patient ID0110M). Amplified fragments were cloned into a pGEM-T Easy Vector (Promega) linked by the XhoI restriction enzyme sites. The sequences of all fragments were verified by Sanger sequencing. The construct was subcloned into a pcDNA3.3-TOPO TA vector (Thermo Fisher Scientific) and used to transform JM109 competent cells (Promega), using the heat-shock procedure. The transformed cells were plated on Luria broth (LB) agar containing 100 mg/mL ampicillin and incubated overnight at 37°C. Positive colonies were suspended in 2 ml of LB supplemented with 100 mg/mL ampicillin. After 16 hours of incubation at 37°C with shaking, the bacterial mixture was used to isolate plasmids, as previously described [31].
The day before transfection, HeLa CCL-2 cells purchased from ATCC were seeded (0.8 x 105) into 6-well culture plates containing Dulbecco’s modified Eagle medium (DMEM) (Gibco by Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS), penicillin (100 U/mL), and streptomycin (100 μg/mL) and incubated at 37°C in a 5% CO2 incubator. Two hours before transfection, the medium was replaced with antibiotic- and FBS-free DMEM. The starved cells were then transfected with RET minigenes (2 μg) using OPTI-MEM medium and Lipofectamine 3000 (both by Gibco, Thermo Fisher Scientific). After a 24-h incubation, the transfected cells were harvested, RNA was isolated, and cDNA was generated and assayed to quantify minigene expression, as described above (see “RNA analysis”).
For the treatment with Actinomycin D (Thermo Fisher Scientific), we plated HeLa cells (0.4x105) into 12-well culture plates and, as described above, four minigenes were transfected in Hela cells. After 24-h from transfection; we treated HeLa cells with 10 μg/ml [32] of Actinomycin D (Thermo Fisher scientific) for 30 minutes, 1-, 2- and 4-hours. Cells were harvested at established time point trough enzymatic method (0.5% trypsin) after two washes in PBS, in order to preserve only viable cells. Transfected and treated cells were pelleted at 1000 rpm for 5 minutes. Total RNA was isolated from transfected and treated cells with Qiagen’s RNeasy Mini Kit. RNA was treated with DNase and cDNA was generated using the High Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific). The half-lives of mature transcript minigenes were assessed as described in “RNA analysis” paragraph.
Four platforms were used to identify altered splicing motifs and splicing factor binding sites in exon 11 of RET: the Human Splicing Finder (http://www.umd.be/HSF3/); ESResearch (http://ibis.tau.ac.il/ssat/ESR.htm); Automated Splice Site And Exon Definition Analyses (ASSEDA) (http://splice.uwo.ca); and ESE Finder (http://krainer01.cshl.edu/cgi-bin/tools/ESE3/esefinder.cgi?process=home).
The results were obtained analyzing the entire sequence of exon 11 WT or with p.C630=;C634R variations in all queries of the four platforms.
We treated RET minigene-transfected HeLa CCL-2 cells with 1% formaldehyde for 10 minutes at room temperature to crosslink RNA-protein complexes. The cells were washed in glycine (125 mM) to stop the reaction, harvested, and resuspended in PBS diluted 1:1 with nuclear isolation buffer (NIB) consisting of sucrose (1.28 M), Tris-HCl pH 7.5 (40 mM), MgCl2 (20 mM), and Triton X-100 (4%). To lyse the nuclear membranes, we resuspended the nuclear pellet in RIP buffer containing KCl (150 mM), Tris pH 7.4 (25 mM), EDTA (5 mM), DTT (0.5 mM), NP-40 (0.5%), RNAase inhibitor (100 U/ml), and protease inhibitors. One tenth of the total volume of nuclear lysate was reserved for use as an input control. Chromatin was needle-sheared, antibody was added to the solution (anti-SRp55 antibody [Millipore] for RIP samples, anti-IgG antibody [Millipore] for negative control samples), and samples were incubated at 4°C with gentle rotation to enhance antibody-protein binding. One hour later, magnetic protein G beads (Dynabeads, Thermo Fisher Scientific) were added and the samples re-incubated for 1h at 4°C with gentle rotation. Unbound material was removed by two washes with RIP buffer and one in PBS. The de-crosslinking reaction was performed at 70°C for 10 minutes. RNA and protein were isolated with TRIzol reagent (Thermo Fisher Scientific). cDNA was generated with High Capacity cDNA Reverse Transcription kit (Thermo Fisher Scientific).
Real-time PCR was performed as described above in the paragraph “RNA analysis”.
Statistical significance was assessed with the Student’s t test, and results were considered significant when P values were <0.05.
|
10.1371/journal.pcbi.1006584 | Multi-scale computational study of the Warburg effect, reverse Warburg effect and glutamine addiction in solid tumors | Cancer metabolism has received renewed interest as a potential target for cancer therapy. In this study, we use a multi-scale modeling approach to interrogate the implications of three metabolic scenarios of potential clinical relevance: the Warburg effect, the reverse Warburg effect and glutamine addiction. At the intracellular level, we construct a network of central metabolism and perform flux balance analysis (FBA) to estimate metabolic fluxes; at the cellular level, we exploit this metabolic network to calculate parameters for a coarse-grained description of cellular growth kinetics; and at the multicellular level, we incorporate these kinetic schemes into the cellular automata of an agent-based model (ABM), iDynoMiCS. This ABM evaluates the reaction-diffusion of the metabolites, cellular division and motion over a simulation domain. Our multi-scale simulations suggest that the Warburg effect provides a growth advantage to the tumor cells under resource limitation. However, we identify a non-monotonic dependence of growth rate on the strength of glycolytic pathway. On the other hand, the reverse Warburg scenario provides an initial growth advantage in tumors that originate deeper in the tissue. The metabolic profile of stromal cells considered in this scenario allows more oxygen to reach the tumor cells in the deeper tissue and thus promotes tumor growth at earlier stages. Lastly, we suggest that glutamine addiction does not confer a selective advantage to tumor growth with glutamine acting as a carbon source in the tricarboxylic acid (TCA) cycle, any advantage of glutamine uptake must come through other pathways not included in our model (e.g., as a nitrogen donor). Our analysis illustrates the importance of accounting explicitly for spatial and temporal evolution of tumor microenvironment in the interpretation of metabolic scenarios and hence provides a basis for further studies, including evaluation of specific therapeutic strategies that target metabolism.
| Cancer metabolism is an emerging hallmark of cancer. In the past decade, a renewed focus on cancer metabolism has led to several distinct hypotheses describing the role of metabolism in cancer. To complement experimental efforts in this field, a scale-bridging computational framework is needed to allow rapid evaluation of emerging hypotheses in cancer metabolism. In this study, we present a multi-scale modeling platform and demonstrate the distinct outcomes in population-scale growth dynamics under different metabolic scenarios: the Warburg effect, the reverse Warburg effect and glutamine addiction. Within this modeling framework, we confirmed population-scale growth advantage enabled by the Warburg effect, provided insights into the symbiosis between stromal cells and tumor cells in the reverse Warburg effect and argued that the anaplerotic role of glutamine is not exploited by tumor cells to gain growth advantage under resource limitations. We point to the opportunity for this framework to help understand tissue-scale response to therapeutic strategies that target cancer metabolism while accounting for the tumor complexity at multiple scales.
| Cancer remains one of the leading causes of death worldwide. A central challenge in understanding and treating cancer comes from its multi-scale nature, with interacting defects at the molecular, cellular and tissue scales. Specifically, the molecular profile at the intracellular level, behavior at the single-cell level and the interactions between tumor cells and the surrounding tissues all influence tumor progression and complicate extrapolation from molecular and cellular properties to tumor behavior [1–3]. Understanding the multi-scale responses of cancer to microenvironmental stress could provide important new insights into tumor progression and aid the development of new therapeutic strategies [2]. Therefore, cancer must be studied and treated as a cellular ecology made up of individual cells and their microenvironment. This ecological view should account for the competition and cooperation of different molecular and cellular players, and for both the physical and biological characteristics of the environment in which tumor evolves. Such perspectives complement studies of the genetic drivers of tumor and potentially provide new bases for treating this disease [4].
Central to an ecological perspective of tumors is metabolism, the biochemical process by which cells derive energy and biomass from the nutrients available in their environment while excreting products of metabolism back to the environment. This exchange of metabolites impacts the distribution of resource in the environment and sets constraints on the availability of resources to individual cells [5]. Therefore, metabolism couples the behavior of individual cells to the characteristics–spatial-temporal organization and phenotypic make-up–of the full population. Recently, cancer metabolism has drawn renewed attention in the field of cancer biology [4,6]. Following the early observations of the unique tissue-scale metabolic profile of tumors made by Otto Warburg in the 1920s, discoveries of oncogenes and molecular cues in tumor-associated metabolic alterations have renewed the hope for therapeutic routes that target cancer metabolism [7].
In his seminal work, Warburg noted the distinct metabolic profile of tumor cells with high glycolytic rate and lactate production in the presence of oxygen. This so-called Warburg effect or aerobic glycolysis has been widely observed in different types of tumor cells (Fig 1A, ①) [8]. This original observation by Warburg led him to hypothesize that aerobic glycolysis is caused by impaired respiration; in turn, this defect results in cancer [9]. It is now well accepted that this hypothesis is incorrect as most tumor cells retain functional mitochondria [10,11]; we still lack a full understanding of the origin and consequences of the Warburg effect. More recently, other hypotheses have been proposed in the field of cancer metabolism such as the reverse Warburg effect (Fig 1A, ②) and glutamine addiction (Fig 1A, ③) [3,12–16]. Despite support of these three hypotheses from various experimental studies, significant uncertainty remains with respect to their definitions, their origin, and their impact on tumor progression and therapeutic interventions. Unraveling these fundamental questions could open a clearer path to targeting cancer metabolism as a therapeutic strategy.
In the past few years, studies of cancer metabolism have begun to elucidate how the metabolic alterations in tumor cells can influence tumor progression [9,17–21]. A definitive characteristic of tumor cells is uncontrolled proliferation. Compared to healthy cells that remain quiescent in most of their life cycle, tumor cells proliferate rapidly, accompanied by high rates of metabolic uptake. This metabolic profile of tumor cells leads to significant depletion of metabolites in the local microenvironment, resulting in resource limitations. Additionally, byproducts and waste products produced by the metabolism of tumor cells can potentially hinder the growth of neighboring cells or act as sources of alternative metabolic substrates [16,22,23]. Although studies have made efforts to capture these experimental observations mathematically [20,24–29], we are unaware of computational studies that test the implications of these hypotheses with respect to metabolic behaviors at the individual cell level, intercellular interactions mediated by shared metabolic environment, and the collective behavior that together define fitness and growth potential of the tumor. Recent computational work has made progress toward capturing the multi-scale complexity of cancer. These studies investigated the effect of tumor microenvironmental factors, specifically molecular cues and metabolites, on tumor population dynamics and provided insights into the cooperative behaviors of tumor subpopulations [30–34]. Similar intraspecies competition or cooperation are often observed in microbial organisms and heavily studied from a population ecology perspective [35–37]. Theories and modeling tools are better developed in the microbial field due to the relatively convenient validation from experiments [38–40].
In this study, we take a multi-scale modeling approach to describe the intracellular, cellular, and multicellular behaviors of cells within a tumor (Fig 1). With this framework, we investigate the following hypotheses: Warburg Effect/Aerobic glycolysis (①), Reverse Warburg (②), and Glutamine Addiction (③). We begin by translating hypotheses from experimental studies into constraints and objectives within the FBA (Fig 1A). We proceed to use FBA to obtain the yield coefficients (Y = maximum growth rate/flux of metabolite) for use in Monod-like kinetics of cellular growth at the individual cell level (Fig 1B). Finally, we simulate the growth dynamics of these cells at the multicellular scale to elucidate the implications of these metabolic scenarios (Fig 1C). We address the impact of the metabolic phenotypes implied by current hypotheses on the growth dynamics of tumor cells in the resource-limited microenvironments that emerge after tumor initiation. This modeling framework opens a route to explore tissue-scale tumor dynamics with explicit account taken for these metabolic scenarios in an efficient manner.
Fig 1 illustrates, schematically, the multi-scale approaches we use. At the intracellular scale, we use Flux Balance Analysis (FBA) to construct a network that captures the central metabolism of mammalian cells (Fig 1A). In Fig 1A, the arrows represent fluxes of species within a reduced representation of cell metabolism and cell growth; the detailed network used in FBA is presented in S1 Fig. Key steps associated with three hypotheses are labeled: Warburg effect (①) is distinguished by high glycolytic flux and lactate production; reverse Warburg (②) is distinguished by the uptake of lactate; and glutamine addiction (③) is distinguished by uptake of glutamine as a carbon source to feed TCA cycle. We build the biomass template reaction (S1 Fig) based on major precursors for biomass synthesis by reducing Shlomi and coworker’s genome scale biomass template [20]. We impose a cellular maintenance reaction with a baseline rate to define the required minimum metabolism of cells (see Methods). We modify constraints and objective functions within the FBA network to define the characteristics of the different hypotheses (labeled in Fig 1A). We estimate parameters based on literature (see S1 Table). We acknowledge that the altered metabolic phenotype of tumor cells may be due to prior genetic events that occurred in the cell, such as loss of tumor suppressors (e.g., p53) [41]. However, we only consider the metabolic phenotypes of the cells at fixed genetic profiles here since we focus on impact of metabolic profiles on tumor growth over time scales (days) that are short relative to those required for the emergence and accumulation of genetic alterations in the cells (months or years). At the cellular scale (Fig 1B), we use the imposed maximum growth rates (μm,n [hr-1]) and the metabolic uptake and production rates of the metabolites (qi/n [g/g-DW-hr]) obtained from FBA to determine yield coefficients ((Yi/n [g-DW/g]) for each metabolite (i) and corresponding metabolic phenotype (n):
Yi/n=−μm,nqi/n
(1)
These yield coefficients link our intracellular treatment of metabolism by FBA and our cellular and multicellular treatment of resource utilization and growth. We model biomass (Xm [g]) growth of each cell type as a Monod-like process parameterized by maximum growth rate for each metabolic phenotype, μm,n [hr-1] and a Monod function of metabolite concentrations, fn({Cj})Monod:
dXmdt=Xm⋅∑nμm,n⋅fn({Cj})Monod
(2)
We provide detailed discussions of the Monod functions in the next Section. We used the same value of maximum growth rate for each phenotype of each cell at both the FBA (Fig 1A) and cell-scale (Fig 1B). We report parameter values in S1 Table. Additionally, we use a threshold in cell diameter to define the doubling of the cell by linking biomass growth to the volume (Vm [L]) expansion of the cell at a fixed dry mass density (ρ [g-DW/L]):
dVmdt=1ρdXmdt
(3)
To bridge the treatment of metabolic processes at the cellular and multicellular scales, we solve steady state species balances for each explicit metabolite at each time step within iDynoMiCS [39]:
∂Ci∂t=Di∇2Ci+∑mρ∑nqi/n⋅fn({Cj})Monod
(4)
where Ci [g/L] is the concentration of ith metabolite, Di [m2/day] is the diffusion coefficient of ith metabolite. Here, the species balances can be safely treated as being at steady state because the time step in our simulation (1 hour) is selected to resolve cell growth and is long compared to typical transients in metabolism [39].
We integrate Eqs 1–4 into iDynoMiCS to track the growth of individual cells within a continuum matrix occupied by other cells in which metabolites diffuse (term 1 in Eq 4). The concentration of metabolites at the multicellular scale governs the cellular biomass growth (Eq 2) and the biomass kinetics in turn influences the concentration profile of metabolites (term 2 in Eq 4), and subsequently the growth kinetics of the surrounding cells. Cells are treated as hard spheres [38]. This spatial-temporal interaction between the cells and the microenvironment is a dynamic process that changes at each time step within iDynoMiCS.
We simulate growth in both radial and axial geometries in iDynoMiCS (Fig 1C): 1) Radial, two-dimensional growth (Fig 1C i))–tumor cells (red) grow radially out from an initial cluster of cells with metabolites supplied at the edge of the cell mass such that radial gradients of concentration emerge (color map). As the tumor grows, concentration gradients of metabolites become significant, making the tumor growth a diffusion-limited process that can result in different growth dynamics as well as distinct spatial distribution of cell subpopulations. 2) Axial, one-dimensional growth (Fig 1C ii))–layers of tumor cells (red) and stromal cells (blue) are initiated near a blood vessel that supplies metabolites (from the top), such as glucose and oxygen in the blood stream. Growth pushes cells deeper into the tissue, away from the vessel, such that strong gradients of metabolite can again occur.
The radial simulations (Fig 1C i)) provide a qualitative understanding of the growth dynamics in different metabolic scenarios; axial simulations (Fig 1C ii)) allow us to further quantify the observed dynamics. In both cases, we initiate tumor cell clones (same Monod parameters) surrounded by a varying number of layers of stromal cells (defined by distinct metabolic and growth parameters–see Fig 2 and Table 1). These arrangements capture tumor growth with initiation occurring at different distances from local vascular structure and thus at different levels of metabolic stress. We proceed to track growth as a function of depth of initiation and metabolic phenotype. We perform 11 replicates, with randomly seeded initial positions of tumor and stromal cells within their compartments; all other parameters in the simulation were kept the same across these replicates for each metabolic scenario. Additionally, we kept these random initial seeding positions of cells the same across simulations of the three metabolic scenarios to eliminate any effect that comes from the difference in the initial seeding when comparing the scenarios. As we are interested in initial stages of avascular growth, we do not account for later stage processes such as angiogenesis. Further, we do not account for cell death explicitly in our simulations; tumor cells in zones with severely depleted metabolites remain quiescent based on the Monod-like growth kinetics. When evaluating total tumor size, this assumption is equivalent to counting dead cell mass within the necrotic core as part of the tumor; this definition is consistent with that of previous studies [42–46].
With the aim of providing intuition on the outcomes of simulations and characteristic physical parameters, we also calculate the Krogh length, shown schematically in Fig 1C iii). Here, we define the Krogh length of a metabolite as the length at which the concentration of metabolite becomes zero given the uptake of the metabolite with zeroth order growth kinetics for the cell phenotype in the region (see Methods). While this is an extremely simple model that couples zeroth order kinetics with a continuum description of reaction and diffusion in the tissue, we will show that it provides insights into the characteristics by which reaction and diffusion govern the growth of tumors.
With this multi-scale computational framework, we study the tumor population dynamics in a spatial-temporal manner and investigate the consequences of different hypotheses in cancer metabolism from a population ecology perspective. This perspective examines the impacts of phenotypic composition, spatial structure and reaction-diffusion on tumor growth.
Before we further specify the hypotheses depicted in Fig 1 individually, we define the metabolic phenotypes of the cell types implicated in these hypotheses based on observations in the literature. We integrate our interpretations of these metabolic mechanisms into FBA to obtain the uptake and production rates of metabolites (see Table 1): In our approach, we assume each cell type (e.g., healthy stromal cell) can adopt more than one metabolic phenotype (e.g., aerobic under normoxic conditions and anaerobic under hypoxic conditions). These different metabolic phenotypes are implemented as objective functions and constraints in FBA and in turn, result in different flux distributions (Fig 2, coded by color). We then obtain yield coefficients (Yi/n) for the ith metabolite in the nth metabolic phenotype of cells by linking maximum growth rate (μm,n) of the mth cell type to the uptake and production rates (qi/n) (Eq 1); the Yi/n serve as measures of the efficiency with which the metabolites generate biomass: the bigger the value of Yi/n is, the more efficiently the nth metabolic phenotype utilizes the ith metabolite to grow.
Fig 2 summarizes predictions from FBA for the metabolic profiles of these cell types under distinct metabolic phenotypes. The metabolic switch from normoxia to hypoxia or to hypoglycemia leads to drastic changes in metabolic fluxes; the values in box represent fluxes of the specific metabolites when they display different metabolic profiles, coded by color (see caption). These flux distributions in turn lead to different uses of metabolites as reflected in yield coefficients (presented in Table 1). We present detailed description of each phenotype of the cells in the following subsections.
To gain a qualitative understanding of the impact of the various metabolic scenarios on tumor growth in a diffusion-limited microenvironment, we first ran simulations in an unconstrained 2-D domain, as shown in Fig 1C i); metabolites were delivered through a diffusive boundary layer of fixed thickness that surrounds the growing tissue (see Methods). Fig 3 presents the form of the tumors at initiation (t = 0) and after 100 days of growth for Warburg tumor cells (WN = 2) with healthy stromal cells (top row), Reverse Warburg cells with hijacked stromal cells (middle row), and glutamine-addicted tumor cells with healthy stromal cells (bottom row). The three columns are for initial seeding of tumor cells beneath 1, 3, and 5 layers of stromal cells, as indicated in the images of the initial configuration of the cells (t = 0). As the number of layers of stromal cells increases, the growth of tumor cells becomes compromised due to the reduced access to the metabolites. By hindering diffusion and consuming oxygen and glucose, the stromal cells decrease the accessibility of these metabolites to the tumor cells. In all cases, we note that the proliferation of the tumor cells led to their breaking through the layers of stromal cells; for the cases with significant growth, the stromal cells became engulfed within the tumor, as is frequently observed in actual tumors [61]. We also note the emergence of irregular front of the tumor in the scenario of Reverse Warburg effect. We suspect that this irregularity arises from growth instability due to the moderate availability of metabolites at the growth front [35]. We note qualitatively different effects of the addition of layers of stromal cells on tumor growth for the different scenarios: with 5 layers of stromal cells, the growths of both Warburg and Glutamine-addicted tumor cells were strongly delayed, whereas the impact on the growth of Reverse Warburg cells was modest. These observations motivate a deeper investigation of the mechanisms that control response to metabolic stress in these scenarios.
We proceeded to dissect the metabolic scenarios further with simulations in a confined geometry in which solute (i.e., metabolite) diffusion and tissue expansion were constrained along a single direction, as shown in Fig 1C ii). This axial scenario approximates the local environment adjacent to a blood vessel (upper boundary). Fig 4 presents an overview of the growth behavior in this geometry. For this overview, we simulated the Warburg scenario, with Warburg tumor cells (WN = 2) and healthy stromal cells (also see Fig 2).
Fig 4A shows the snapshots of tumor growth and the corresponding concentration fields of oxygen and glucose at various time-points for tumors initiated beneath 1, 3, and 5 layers of stromal cells. In the colormaps of the concentration fields, we see that when the tumor initiated closer to the source (top row with 1 layer of stromal cells), the Warburg tumor cells had access to ample oxygen and glucose to fuel their growth at early time (t = 0, empty circle); at late time (t = 25 days, filled circle), significant depletion of both oxygen and glucose occurred, but the uppermost layer of tumor cells still benefited from high metabolite concentrations to grow. However, when the tumor initiated farther away from the source (middle and bottom rows), the diffusion limitations and consumption by the stromal cells limited the metabolites available to the tumor cells, even at early times (empty diamond, empty square). This limitation persisted until the tumor cells broke through the stromal layer and gained access to higher concentrations of metabolites (filled diamond, filled square).
Fig 4B presents the trajectories of tumor growth from 11 simulation runs in each case shown in Fig 4A. We first note that for all initial conditions, the growth appears to proceed through two phases, starting with slower growth that then transitions to faster growth; these two regimes are most evident for 3 and 5 layers of stromal cells. By observing the cellular configurations in the simulations (see S1–S3 Movies), we identify that the transition occurs when the tumor cells break through the layers of stromal cells and gain access to high concentrations of metabolites. When the tumor cells started to grow, the reaction-diffusion in the intact layers of stromal cells limited the supply of metabolites to the tumor cells. Under such conditions, the growth of tumor cells was significantly compromised due to the lack of oxygen (note the more severe depletion of oxygen relative to glucose in Fig 4A, also see Eq 7 in Model); the microtumor was nearly quiescent. Once this slow growth led to the penetration of one or more tumor cells through the layers of stromal cells, those tumor cells transitioned toward their aerobic growth regime (term 1 in Eq 7) and quickly overwhelmed the stroma. Interestingly, the growth rates after breakthrough were constant (the growth curves are linear in time) and independent of initial conditions (all late-time slopes are the same in Fig 4B). This constant growth rate is distinct from the exponential growth that one would expect resulting from saturating Monod-like growth kinetics (Eq 5–10 in Model). This observation illustrates an important consequence of a diffusion limited microenvironment. We will comment further on the origin of this constant rate below.
In the case of 1 layer of stromal cells (black curves), the growth transitions rapidly (within the first days) to a high, constant rate. Furthermore, the trajectories of all the random initial seeding conditions are very similar. For 3 and 5 layers of stromal cells (blue and red curves), the first, slow phase lasts longer because the tumor cells experienced more severe limitations in their initial configurations. Additionally, in these cases, the trajectories of different initial conditions diverge strongly from one another due to the differences in the moment of transition from slow to fast growth. This observation reflects the fact that the time for tumor cells to break through the stroma is sensitive to small differences in the initial configuration of cells.
We now proceed to use axial simulations like those in Figs 1C ii) and 4 to investigate the growth dynamics in each of the three metabolic scenarios.
Within our scope of study of the Warburg effect through the multi-scale modeling approach (Figs 4 and 5), we confirmed a common hypothesis that Warburg effect impacts tumor cell fitness in metabolically limited microenvironments [64]. Interestingly, our predictions suggest that there may exist an optimal level of Warburg effect (reflected by the ratio of pyruvate fluxes to lactate and to the TCA cycle; the Warburg Number) for tumor cells to adopt depending on the details of the metabolic microenvironment in which the tumor cells initiate. This observation may help explain the experimentally observed phenotypic heterogeneity in cancer metabolism [65,66]. Such adaptation could occur via modification of the fluxes of pyruvate, for example with changes in enzymatic rates along either the TCA cycle or glycolytic pathways. From an ecological perspective, our predictions indicate that Warburg effect may provide a basis for adaptation of tumor cells to different environmental metabolic stresses [67].
For the reverse Warburg effect scenario (Fig 6), we provide the first mathematical description of the multi-cellular metabolic interactions proposed by Sotgia et al. [55]. We used our framework to explore the intracellular and multicellular consequences of reverse Warburg effect due to the interaction between glycolytic stromal cells (hijacked stromal cells) and lactate-consuming tumor cells (reverse Warburg tumor cells). We predict that the hijacked stromal cells have higher yields on oxygen than healthy stromal cells (Fig 2B i) and Table 1). This information confirmed the intuitive proposal of Sotgia et al. that the hijacked stromal cells assist in the growth of tumors that initiate deep within the stroma by allowing more oxygen to penetrate into the tumor compartment (Fig 6C–6E). We further note that, due to the adaptive character of reverse Warburg tumor cells, they are not sensitive to local lactate concentration in aerobic growth regimes (term 1 and 2 on the right-hand side of Eq 9 can be combined); this characteristic means their aerobic growth remains limited by oxygen and glucose only. Additionally, due to the utilization of lactate as carbon source in energy production in these tumor cells, their yields on oxygen is lower compared to tumor cells in the scenario of Warburg effect (requiring more oxygen for the same mole of carbon consumed). Therefore, the reverse Warburg effect leads to slower growth in favorable metabolic microenvironment (i.e., abundant source of metabolites available). However, when tumors initiate in microenvironments where resources are significantly reduced, the host-parasite relationship implied by the reverse Warburg effect (via cooperative utilization of oxygen between hijacked stromal cells and tumor cells) can provide growth advantage to tumors. Given that such growth advantage depends on the detailed structure of the metabolic microenvironment, we suggest that one must use a multi-scale framework like the one presented here to investigate the implications of these metabolic scenarios.
In the exploration of glutamine addiction (Fig 7), we defined the metabolic phenotype by hypothesizing that glutamine addiction coexists with Warburg effect. This hypothesis led us to propose a coupled contribution to biomass synthesis of tumor cells from glucose and glutamine as joint carbon sources. Specifically, we aimed to explore the role of glutamine in anaplerosis (as a carbon source to replenish the TCA cycle). We demonstrated with FBA that under our interpretation, glutamine addiction led to an increase uptake of oxygen (i.e., lower yield on oxygen) in glutamine-addicted tumor cells to maintain their redox balance and to meet the energy demand; this lower yield on oxygen represents a cost of using glutamine in the TCA cycle. We see the impact of this lower yield on oxygen in the reduced growth rate of glutamine-addicted tumor cells relative to Warburg tumor cells. We thereby conclude that glutamine addiction via the process of anaplerosis does not confer an advantage to the overall tumor growth primarily due to the strong dependence on oxygen. We argue that glutamine is not an effective alternative carbon source because tumor cells remain limited by glucose and oxygen.
Our study constrains future considerations of the roles of glutamine addiction in tumor growth by clearly demonstrating that the anaplerotic pathway cannot, alone, provide a growth advantage to tumors. With our focus on the anaplerotic role of glutamine using a simplified metabolic network, we did not account for other roles of glutamine in cellular demand explicitly [68,69]. For example, glutamine is known to be an important nitrogen source in nucleic acids and amino acids synthesis [57,70,71]. Additionally, glutamine contributes to the pool of metabolites that maintains NADPH/NADP+ balance [69,72] and to produce glutathione as an antioxidant to help the cell resist oxidative stress during rapid metabolism [70,72]. We conclude that a more detailed investigation that accounts for the multi-scale implications of these additional pathways is needed in the future.
With our approach, the growth curves captured in our spatially resolved model (a slow growth regime followed by a fast unidirectional linear growth) are compatible with the experimentally observed growth of avascular solid tumors [63,73]. Previous studies attributed the linear growth regime observed at late-time tumor growth to available space for growth and cell diffusion at the edge of the tumors [73,74]. Here, our simulations and analysis indicate that this effect can be entirely explained by diffusion limitations of metabolites.
In our exploration of Warburg and reverse Warburg effect, our approach provided a basis for exploring the heterogeneity in metabolic phenotypes that has been suggested by recent experiments [65,66]. For example, the crossover of growth rates that we observed from early to late times (Figs 5C and 6C) suggests that adaptation of metabolic phenotypes (e.g., from high to intermediate WN or from RW to Warburg) could improve overall growth potential of tumors.
In parallel with experimental approach, computational tools allow for high throughput investigation of hypotheses that are emerging rapidly in the field of cancer study [24,29,33,34,68,75,76]. Particularly, a multi-scale modeling framework such as the one presented here can provide a basis for predicting cell-level to tissue-scale response to therapeutic interventions. For example, the action of inhibitors of key regulators of cellular metabolism such as PI3K [77] can be accounted for in FBA as flux constraints (e.g., a reduced upper bound on glycolytic flux); the obtained uptake rates of metabolites could then be propagated through to the tissue-scale ABM in our framework in order to examine the effect on tumor growth at the population scale.
We finish by emphasizing that our interpretations of the three metabolic scenarios studied here are not unique either with respect to the choices of constraints and objectives imposed for FBA or the details of the cellular configurations within our simulations. Our modeling framework can accommodate a large diversity of hypotheses and should serve as a powerful tool with which to evaluate emerging ideas and experimental observations from the rapidly evolving field of cancer metabolism.
To capture the intracellular details of different metabolic phenotypes of cells, we adopt the well-established framework of FBA.
In our study, the central carbon metabolism of human was constructed with 140 reactions and 92 metabolites (S1 Fig). Of those 140 reactions, 34 consist of boundary exchange of metabolites such as uptake and secretion, 26 consist of mitochondrial exchange of metabolites with the cytosol, 1 is the biomass template reaction, 1 reaction for maintenance, and the 78 remaining reactions are transformations of metabolites that occur in the cytosol and mitochondrion. The biomass template reaction for growth in the human model was adapted from the Shlomi et al.’s genome-scale model [20]. Shlomi et al.’s biomass template reaction consists of 30 biomass compounds including amino acids (0.78 g/g-DW), nucleotides (0.06 g/g-DW), and lipids (0.16 g/g-DW). These biomass requirements were combined and reduced to their upstream precursors for simplification in our biomass template reaction. For example, stoichiometric equivalence of ribose-5-phosphate, the precursor of nucleic acids, was used in place of nucleotides in their final form. For the maintenance rate, we first sampled a range of values from 0 to 10 mmol ATP/g-DW-hr [78–80] and concluded that the overall qualitative trend of our FBA results was not affected by this choice. Therefore, for simplicity, a maintenance rate of 5 mmol/g-DW-hr is used consistently for all cell types. This maintenance rate represents 73% of the total energy expenditure, comparable to what was previously reported for mammalian cells, which is 65% [80]. Using our reduced biomass function, our glucose yields (YG/n) matched closely with that of Shlomi and coworkers [20]. For example, within the metabolic phenotype of WN = 0, at the same growth rate range and maintenance rate of 0, the yield coefficient (specific growth rate per glucose) of our reduced order model (0.0984 g-DW/mmol) was within 4% of that found with Shlomi et al.'s genome scale model (0.094 g-DW/mmol).
Under hypoxic conditions (CO << KO), we assumed a quiescent phenotype for all cell types. To capture the hypoxic condition, we minimized the oxygen uptake rate while maintaining a growth rate of 1 ×10−6 hr-1 to represent the quiescent state.
For tumor cells in the metabolic scenarios of reverse Warburg effect and glutamine addiction, we used a quiescent phenotype for tumor cells under hypoglycemic conditions (CG <<KG). We achieve this condition in FBA by minimizing glucose uptake while allowing uptake of lactate or glutamine respectively and constraining growth rate to be 1×10−6 hr-1.
iDynoMiCS is an individual-based modeling platform originally built for the study of microbial biofilms [39]. It allows computation of diffusion-reaction kinetics at individual cell level and has multiple built-in kinetic mechanisms, including Monod forms as in Eqs 5–10. Additionally, iDynoMiCS treats the cell movement through two mechanisms: displacements due to pressure-induced convection at the global scale based on Darcy’s law, and sterically induced displacements that avoid overlapping during the expansion and division of neighboring cells at a local scale. During a simulation, the pressure that is directly proportional to the rate of biomass generation or degradation is computed first to induce global convection, followed by the computation of “shoving” (random displacement) at local scale; these displacements are selected by a relaxation algorithm to avoid steric overlap. The shoving mechanism is propagated through all cells until the number of cells that are still moving is negligible, and leads to local random displacements of cells [39].
In our case, since we are explicitly interested in studying how diffusion-reaction kinetics impact the tumor growth under various hypotheses on cancer metabolism with no specific consideration of molecular guidance for cell movements, the random, local cell motion provided by iDynoMiCS serves as a reasonable approximation of cell dynamics within the tissue [81]. The 2-D simulation domain is discretized into a square grid on which the reaction-diffusion equation is solved by finite difference at each time step (Eqs 2 and 4). The domain is also divided into two compartments: the “tank” and the “biofilm”. The tank serves as the source of metabolites; we interpret this compartment to be the blood stream with which the tissue exchanges nutrients. The “biofilm” defines the tissue where the metabolites undergo diffusion and reaction; the local reaction rate for each metabolite is set by the density and metabolic character of the cells in the grid element. A boundary layer defines the resistant to diffusive mass transfer between the blood stream (“tank”) and the cells (“biofilm”). In our axial simulations, we allowed the exchange of metabolites only at the top of the domain by having zero-flux boundary condition at the bottom of the domain and periodic boundary conditions on the sides and in the 3rd dimension (S3 Fig). We set the concentrations of metabolites in the “tank” at their physiological concentrations in human blood stream (S1 Table). We selected a grid size for solving reaction-diffusion process to match individual mammalian cell size (~10 μm, [82]) and a boundary layer thickness, h, to represent the thickness of the vascular endothelium (S1 Table). The size of the cell was used to determine the density of the cell based on dry cell mass (S1 Table). With the density of the cell fixed, we calculated the spherical volume of the cell from biomass growth by conservation of mass. This volume was then used to calculate the diameter of cells at each time step. The calculated diameter at each time step was then used to compare to a threshold value to determine the division of the cell. Once the computational domain was defined, we then specified the reactions that govern the cell growth. In each reaction, we chose parameters such as half saturation constant (S1 Table). Together with parameters such as diffusion coefficients and physiological concentrations of metabolites obtained from the literature, we checked that the calculated value of the Krogh length (e.g., ~40μm for oxygen) was in the right range for mammalian tissue.
In the calculation of Krogh length, we treat the tissue as a continuum and represent consumption of oxygen and glucose as being zeroth order within the steady state reaction-diffusion equation. We calculated the Krogh lengths to determine the limiting metabolite in tumor cell growth in different metabolic scenarios (i.e., different WNs, Fig 5F). The Krogh lengths represent the typical depth of penetration of metabolites into the tumor compartment. We omitted the consumption contributed by anaerobic growth of the cells by assuming the metabolites get completely depleted before the cells switch to anaerobic growth regime. The calculation of Krogh lengths is illustrated in S2 Fig. The metabolite with shorter Krogh length will play a more significant role in determining the growth dynamics of tumor cells.
In Figs 5–7, we evaluated early-time growth rates as the initial slope of the growth curves by taking the difference of the averaged number of tumor cells for the first two outputs of simulation and dividing by the time interval. The time intervals are 10 days, 20 days and 50 days for the cases of 1, 3 and 5 layers of stromal cells for all three metabolic scenarios.
Late-time growth rates were obtained in a similar fashion but evaluated at different time intervals due to the difference in breakthrough times in different cases. A growth over 30 days between the time points 30 and 60 days was used in the case of 1 layer of stromal cells. A growth over 80 days between the time points of 120 and 200 days was used for calculation of late-time growth rates in the case of 3 layers of stromal cells. A growth over 200 days between the time points of 400 and 600 days was applied to the calculation of late-time growth rates in the case of 5 layers of stromal cells. These choices of time ranges were applied consistently in all three metabolic scenarios.
|
10.1371/journal.ppat.1004662 | Prion Infections and Anti-PrP Antibodies Trigger Converging Neurotoxic Pathways | Prions induce lethal neurodegeneration and consist of PrPSc, an aggregated conformer of the cellular prion protein PrPC. Antibody-derived ligands to the globular domain of PrPC (collectively termed GDL) are also neurotoxic. Here we show that GDL and prion infections activate the same pathways. Firstly, both GDL and prion infection of cerebellar organotypic cultured slices (COCS) induced the production of reactive oxygen species (ROS). Accordingly, ROS scavenging, which counteracts GDL toxicity in vitro and in vivo, prolonged the lifespan of prion-infected mice and protected prion-infected COCS from neurodegeneration. Instead, neither glutamate receptor antagonists nor inhibitors of endoplasmic reticulum calcium channels abolished neurotoxicity in either model. Secondly, antibodies against the flexible tail (FT) of PrPC reduced neurotoxicity in both GDL-exposed and prion-infected COCS, suggesting that the FT executes toxicity in both paradigms. Thirdly, the PERK pathway of the unfolded protein response was activated in both models. Finally, 80% of transcriptionally downregulated genes overlapped between prion-infected and GDL-treated COCS. We conclude that GDL mimic the interaction of PrPSc with PrPC, thereby triggering the downstream events characteristic of prion infection.
| Prion diseases are a group of infectious, invariably fatal neurodegenerative diseases. Progress in developing therapeutics is slow, partly because animal models of prion diseases require stringent biosafety and are very slow. We recently found that treatment of cerebellar slices with antibodies targeting the globular domain (GD ligands) of the prion protein (PrP) is neurotoxic. Here we compared this model to prion infection, and describe striking similarities. Both models involved the production of reactive oxygen species, and antioxidants could reverse the toxicity in cerebellar slices and even prolong the survival time of prion-infected mice. Antibodies targeting the flexible tail of PrP that prevent toxicity of GD ligands reduced the toxicity induced by prions. Endoplasmic reticulum stress, which is involved in prion toxicity, is also found in GD-ligand induced neurotoxicity. Finally, changes of gene expression were similar in both models. We conclude that prion infection and GD ligands use converging neurotoxic pathways. Because GD ligands induce toxicity within days rather than months and do not pose biosafety hazards, they may represent a powerful tool for furthering our understanding of prion pathogenesis and also for the discovery of antiprion drugs.
| Prion diseases are lethal infectious diseases that propagate through the conversion of the cellular prion protein (PrPC) into a pathological conformer, the scrapie-associated prion protein (PrPSc) [1]. Neuronal expression of PrPC is required to mediate the neurotoxicity of PrPSc [2] and possibly also of other protein aggregates [3], yet the pathways leading to neurotoxicity are largely unknown. While caspase activation, autophagy, and Ca2+ dysregulation have been shown to occur after prion infections [4,5], ablation of Bax and caspase-12, or overexpression of Bcl-2, does not delay incubation time of prion-infected animals [6,7]. Induction of autophagy, despite enhancing PrPSc clearance in vitro and in vivo, did not prolong survival time of prion-infected mice [8]. Furthermore, excessive unfolded protein responses (UPR) in the endoplasmic reticulum (ER) plays a significant role in the pathogenesis of prion and other neurodegenerative diseases [9,10], yet the biochemical events emanating from prion replication and leading to UPR induction are unknown, and it is unclear how extracellular aggregates can trigger pathology in a subcellular compartment to which they have no direct access.
Prion infection of cerebellar organotypic cultured slices (COCS) has proven to be an extraordinarily faithful and tractable model of prion disease. Prion-infected COCS replicate all salient biochemical, histological, and pathophysiological events which occur during prion infections in vivo, including PrPC-dependent prion replication [11,12], neuroinflammation with proliferation of microglia and astrogliosis, spongiosis, and neuronal cell loss. In prion-infected COCS, calpain inhibition confers neuroprotection without reducing prion replication, suggesting that calpains are involved in neurotoxicity [13].
We have reported that exposure to antibody-derived anti-PrP ligands (full-length antibodies, F(ab)1 fragments thereof, and recombinant single-chain miniantibodies) targeting the globular domain (GD) of PrPC [14] induces rapid cerebellar granular cell (CGC) degeneration in COCS and in live mice. Since this toxic effect was also attenuated by calpain inhibitors [15], we wondered whether the two triggers of PrP-dependent cell death, GDL and prions, might induce similar neurotoxic cascades.
Here we report that antibodies against the flexible tail (FT) of PrPC, which prevent GD ligand (GDL) toxicity in COCS [15], also counteracted neurotoxicity in prion-infected COCS, suggesting a role for the FT in both models. Furthermore, GDL treatment and prion infection triggered similar intracellular cascades including PERK activation [9] and reactive oxygen species (ROS) production. Also, a comparative analysis of transcription in prion-infected vs. GDL-exposed COCS showed extensive similarities between these two paradigms of PrP-related toxicity. We conclude that prions and GDL share downstream pathways of toxicity, and that in both instances the FT is the main molecular effector of prion-mediated toxicity.
Rapid neurotoxicity is elicited in COCS and in vivo by several monoclonal antibodies, single-chain variable fragments (scFv), and F(ab)1, and F(ab)2 fragments directed against the globular domain of PrPC [15]. We collectively termed these reagents “globular domain ligands” (GDL). In all of the experiments described below, we used the POM1 holoantibody (67nM) as a validated paradigm of GDL-associated toxicity. Also, we have previously reported that neurodegeneration and prion replication similarly occur in COCS exposed to the three prion strains, RML, 22L, and 139A [13]. Here, we used RML infection as an extensively characterized paradigm of prion infection. Prion infection of COCS from tga20 transgenic mice overexpressing PrPC [16] elicited toxicity more rapidly than in wild-type COCS [13], and was used for all experiments except when otherwise indicated. As controls, pooled mouse immunoglobulins (IgG) and non-infectious brain homogenate (NBH) were used.
First, we compared the progression of neurodegeneration in GDL-exposure vs. prion infection of COCS by measuring the area positive for neuronal-nuclear antigen (NeuN) within the CGC layer, and by counting cells stained by propidium iodide (PI). The NeuN+ area was used to estimate COCS viability, while the density of PI+ cells correlated with the intensity of ongoing damage. A previously published time-course experiment [15] was repeated including additional time points. PI+ cells peaked at 3 days post-exposure (dpe) (S1A Fig.) and decreased around 7–10 dpe in GDL-treated COCS. Also, significant loss of NeuN+ granule cells was detectable at 3 dpe (Fig. 1A). In prion-infected COCS, we observed a peak of PI+ cells at 38 days post infection (dpi) (S1B Fig.) and significant neuronal cell loss at 45 dpi (Fig. 1B).
ROS, particularly superoxide, are causally involved in GDL toxicity [15]. We therefore asked whether prion infection resulted in ROS production, and whether ROS scavenging might be beneficial. We measured ROS production in live GDL-treated and RML-infected COCS by fluorescent recording of dihydroethidium (DHE) oxidation products [17]. GDL-treated COCS were treated with DHE at various time points between 1 h and 10 days after POM1 exposure (Fig. 1C). Enhanced fluorescence from DHE oxidation products was observed at 4 h (67 nM). Exposure to a fivefold higher POM1 concentration (335nM) resulted in toxicity even after 1 h. Significantly increased fluorescence was observed in prion-infected COCS (Fig. 1D, RML “+”) starting at 25 dpi and reached a peak at 38 dpi, but not in COCS exposed to non-infectious brain homogenate (RML “-”).
Consistently with what we found in GDL-exposed tga20 COCS, we observed significant ROS production, measured by DHE incorporation, in GDL-exposed COCS from wild-type (Bl/6) mice at 7 and 14 dpe (S2A Fig.). This result provides further validation for our view that prion related pathologies show very similar characteristics in wild-type and tga20-derived tissues.
ROS generation was also measured by lucigenin conversion, which detects superoxide anion radicals [18]. COCS exposed to GDL displayed increased lucigenin conversion [15], which was quenched by diphenylene iodonium (DPI), an inhibitor of ROS-producing electron transporters including NADPH oxidases (NOX) [15]. Similarly, we observed elevated lucigenin conversion in prion-infected COCS at 42 dpi, indicating a strong increase in superoxide production. Furthermore, addition of DPI quenched ROS production in prion-infected COCS (Fig. 1E).
We also measured ROS production in vivo using DHE. Terminally sick RML-infected mice were injected intraperitoneally with DHE, and DHE oxidation products were detected in brain homogenates. Forebrains and cerebella of prion-infected mice showed higher levels of fluorescence than NBH-inoculated control mice (Fig. 1F).
If the superoxide burst in prion-infected COCS is a direct consequence of prion infection, interference with prion replication should reduce ROS production. We therefore subjected prion-infected COCS to a panel of compounds that had previously been found to antagonize prion replication, including pentosan polysulfate (PPS), congo red (CR), and amphotericin B (Amph). Prion-induced ROS production was reversed by treatment with PPS, CR, and Amph (Fig. 1G lower half). Hence, ROS production is a general feature of prion toxicity downstream of prion replication.
PPS, CR, and Amph may be effective because they intercalate with prions, or because they activate neuroprotective pathways independently of their interactions with PrPSc. We therefore tested the effects of PPS, CR, and Amph on GDL-treated COCS. We found that ROS production was not reduced (Fig. 1G, upper half) and neurodegeneration was not prevented (Fig. 1H), whereas PPS, CR, Amph counteracted neurotoxicity in prion-infected COCS [13], with PPS being protective for at least 55 dpi (Fig. 1I). We conclude that the prionostatic properties of these compounds, rather than any off-target effects, were indeed the proximal reason for ROS suppression.
Analogously to what we observed in GDL-exposed COCS, the ROS scavengers ascorbate and N-acetyl cysteine (NaC) completely prevented prion neurotoxicity in COCS (Fig. 2A), although neither compound affected prion titers (Fig. 2B). Furthermore, ascorbate did not affect PrPSc accumulation, total PrP levels, or processing of PrPC into the C1 fragment in prion-infected COCS. Only the C2 fragment was decreased (S3A–S3D Fig.).
As previously shown for GDL toxicity, the membrane-impermeable antioxidant isoascorbate and the superoxide dismutase mimetic MnTBAP conferred protection against prion toxicity, suggesting that the relevant ROS species are extracellular in both instances (Fig. 2A). In contrast, the nitric oxide synthase inhibitor 1400W was ineffective in both prion-infected (Fig. 2A) and GDL-exposed COCS [15]. These data indicate that the ROS moiety instrumental to prion-induced neurodegeneration is superoxide, rather than nitric oxide, in both models.
We then investigated whether ascorbate would be neuroprotective over protracted periods of time. RML-infected COCS were treated with ascorbate and harvested at various times between 45–53 dpi. Remarkably, ascorbate significantly reduced neurodegeneration of RML-infected COCS for ≥53 dpi (Fig. 2C). Finally, we asked whether antioxidants might be protective against prion-induced neurotoxicity in vivo. For this, we administered the enterically activated antioxidant, acetylated hydroxy tyrosol (AcHyT, 2g l-1 added to drinking water) [19], to tga20 transgenic mice. AcHyT was previously shown to block the toxicity of GDL in vivo [15]. Tga20 mice pretreated with AcHyT for 7 days were intracerebrally infected with 22L prions (30μl, diluted 10–2) and treatment with AcHyT was continued until mice reached the criteria for termination of the experiment. Treated animals showed a modest, but significant, life extension (Fig. 2D). Hence, AcHyT is protective in vivo against the toxicity of both prions and GDL.
We have previously shown that calpain inhibitors, but not caspase inhibitors, prevent cell death in GDL-exposed [15] and RML-infected COCS [13] (Fig. 2A). In order to test whether ROS signaling occurs upstream of calpain activation, we studied the effects of antioxidants on the catabolism of fodrin, which is specifically cleaved by calpains into fragments of 145 and 150 KDa. This cleavage was blocked by calpain inhibition [13] yet was unaffected by antioxidant treatment in both RML-infected and GDL-exposed COCS, indicating that ROS production is triggered by events that are dependent on (“downstream” of) calpain activation (Fig. 2E, F). This hierarchical sequence may not be unique to PrP-related toxicity, and other calpain activators may plausibly also induce ROS.
Excitotoxicity is a potent ROS inducer [17], and PrPC can modulate NMDA and voltage-gated calcium channels [20,21]. We therefore investigated if inhibitors of NMDA and AMPA/kainate ionotropic glutamate receptors, or of a mitochondrial membrane permeability transition pore, could protect COCS against GDL or prion neurotoxicity. However, none of the inhibitors were protective in either model (Fig. 2A). Also, inhibiting ryanodine receptor-mediated calcium release from the endoplasmic reticulum (ER) with Dantrolene, was not protective (Fig. 2A). None of the tested compounds were inherently toxic, as the viability of IgG-treated or NBH- exposed tga20 COCS were unaffected (S4 Fig.).
High-affinity ligands to the FT of PrPC such as the POM2 antibody [14] are not only innocuous, but counteract the toxicity of GDL. Moreover, interstitial FT deletions prevent GDL toxicity in vitro and in vivo, indicating that the FT is required to execute GDL toxicity [15]. To determine whether the FT also mediates toxicity in prion infection, we treated prion-infected COCS with the POM2 antibody, which recognizes the octapeptide repeats of the FT. POM2 prevented prion-mediated neurodegeneration in tga20 COCS, whereas equimolar amounts of IgG had no beneficial effect (Fig. 3A). We determined prion titers by the scrapie cell-assay in end-point format (SCEPA; [22,23], Fig. 3B), which measures the minimal concentration that still can infect the cells and is currently the most precise measurement of infectivity in vitro. Crucially, prion titers were not significantly affected. This finding disproves the possibility that neuroprotection was caused by reduced infectivity, and suggests that POM2 acted specifically on prion neurotoxicity by interfering with events triggered by the encounter of prions with their target cells.
Western blots of PK-digested samples showed that POM2 treatment led to the appearance of PrP-immunoreactive higher-molecular weight bands (Fig. 3C), possibly representing SDS-stable PrPSc oligomers concomitant with reduced immunoreactivity in the 27–30 kDa range. The total PK-resistant PrP immunoreactivity was determined by densitometric quantification of the entire lane, and was found to be similar to that of samples that had not been exposed to POM2. We conclude that POM2 induced a shift in the distribution of PrPSc moieties without affecting its overall quantity. This interpretation is congruent with the results of prion titer determinations (Fig. 3B).
Since ER stress has previously been shown to be involved in prion disease [9,10], we examined the levels of phosphorylated PERK (p-PERK), phosphorylated eIF2α (p-eIF2α), and ATF4 in both paradigms of PrPC-dependent neurotoxicity. We found a trend towards increased levels of p-PERK, as well as significantly increased p-eIF2α and ATF4 in RML-infected COCS at 42 dpi (Fig. 4A), confirming the activation of the unfolded protein response in prion infections. Surprisingly, we found that POM1-exposed COCS also showed increased p-PERK, p-eIF2α, and ATF4 at 3 dpe (Fig. 4B).
To further explore this phenomenon, COCS prepared from wild-type mice were exposed to POM1 for 7 and 14 days (S2B–S2C Fig.). At 7 dpe there was a trend towards increased p-PERK and p-eIF2α levels, whereas ATF4 was unchanged. At 14dpe we found significantly increased levels of p-eIF2α and ATF4, suggesting again an involvement of the PERK pathway as observed in the tga20 COCS. This suggests that signals emanating from GDLs can propagate to the ER and initiate a response similar to that seen in prion infections.
The transcriptional changes occurring in COCS infected with prions and exposed to GDL were studied by microarray hybridization. Genes were considered to be differentially expressed if they exhibited a fold change of ≥ 2 (p value < 0.01) between RML and NBH (prion infection paradigm) or between POM1 and IgG (GDL exposure paradigm). Upregulated and downregulated genes were compared at various time points (Fig. 5A; S1–S2 Tables). To account for the different velocity of neurodegeneration between the two models, we compared transcriptional profiles at the time at which NeuN staining loss became significant (3 dpe for GDL vs. 45 dpi for prion infection). The largest overlap of transcriptionally altered genes was found when GDL-treated COCS at 3 dpe were compared to prion-infected COCS at 45 dpi. At these time points, COCS shared 38% of all upregulated genes (Fig. 5B left; S3 Table) and 80% of all downregulated genes (Fig. 5B right). At the peak of ongoing cell death in both models (3 dpe for GDL vs. 38 dpi for prion infection; S1A–S1B Fig., measured by PI incorporation), we found that 38.4% (15/39) of upregulated genes were identical. Only one of these fifteen genes, Fosb, has been annotated as possibly involved in the signaling initiated by activated ROS [24]. The remaining 14 genes have been assigned to various cellular pathways, but we failed to identify an overrepresentation of any specific pathway.
The analysis of individual genes, followed by the compilation of a list of candidates using arbitrary cut-off criteria (typically fold-change and p values) may not reveal biologically important effects on pathways. For example, a modest yet concerted increase in the activity of several genes feeding into the same pathway may be more consequential that a strong increase of a single member gene. We therefore set out to evaluate microarray data as predefined gene sets that could be assigned to certain pathways. Gene set enrichment analysis (GSEA) is an important approach to the analysis of coordinate expression changes at a pathway level [25]. Specifically, we applied this method to our microarray data in order to specifically investigate whether the genes constituting the TNF-ROS-CASP3 pathway are significantly regulated in a coordinated manner in POM1 and RML-exposed COCS. Indeed, we found using GSEA, that the TNF-ROS-CASP3 pathway indeed was significantly regulated in both POM1-exposed COCS after 3d (p = 0.037) and RML-exposed COCS after 38d (p = 0.03) and after 45d (p = 0.026). This result shows that in both paradigms genes belonging to the same ROS-dependent pathway are activated upon exposure. Only 3 genes were downregulated at 38 dpi, one of which was downregulated at 3 dpe in GDL-exposed COCS.
The top 40 upregulated pathways at each time point were identified using GeneGo MetaCore software. When comparing POM1 at 3 dpe to prion infection at 45dpi, there was an overlap of 19 pathways (47.5%), while 9 out of the top 10 active pathways at 45dpi RML were present in the top 40 of POM1 7 dpe (S4 Table).
When genes in POM1-treated COCS at different time points were compared with 45 dpi prion-infected COCS, the correlation of genes increased with time and reached a plateau at 7 dpe POM1, with a correlation coefficient close to 0.8 (Fig. 5C). We then examined the involvement of pathways from the GeneGo MetaCore database that had been described to be activated upon prion infection, such as the ER stress response [9], ERK inhibition [26,27], autophagy [28], CCL2 signaling [29], and TNF-ROS-casp3 [29,30]. These pathways were activated in POM1-treated COCS in a pattern that strongly correlated with RML-infected COCS at 45dpi (correlation value between 0.8–0.9). Scatter plots and heat maps of the genes involved in the five signaling pathways (S5A–S5E Fig.; S6 Fig.) support this view.
To validate the regulation of genes in the microarray data, we performed nanostring analysis for 40 genes on the same RNA preparation that was used in the microarray analysis. Differential expression of the selected genes at various time points in prion-infected and GDL-exposed COCS (S5 Table) confirmed the results of the microarray analysis.
Using multiple paradigms in organotypic cultures and in vivo, we show that the toxic antibody POM1 induces largely overlapping pathogenetic cascades as bona fide prion infections. Not only were all strategies preventing GDL-induced neurodegeneration (such as calpain inhibition, ROS scavenging and FT binding) found to be neuroprotective against prions, but compounds neuroprotective against other kinds of insult (such as caspase inhibitors and glutamate antagonists) were ineffective against both GDL and prions. Moreover, the results of transcriptomic analyses are compatible with the contention of a large overlap in the downstream effectors of both pathways. Besides highlighting the commonalities between GDL and prion-related neurodegeneration (S7 Fig.), these observations set both conditions apart from other types of neurodegenerative conditions.
Treatment of prion-infected COCS with antioxidants did not interfere with the aggregation of PK-resistant material, as was previously shown for calpain inhibitors [13]. This adds to the evidence that ROS scavengers and calpain inhibitors mitigate toxicity by interfering with events triggered by prion replication. A plausible model of pathogenesis predicates that toxicity is triggered by binding of either GDL or PrPSc to the globular domain of PrPC. Since inhibitors of prion replication decreased ROS production, but did not protect from GDL toxicity and did not reduce ROS production (Fig. 1G), we conclude that ROS production is downstream of both prion replication and GDL binding.
The modest therapeutic effect of the antioxidative therapy with AcHyt in vivo is not unexpected, since their involvement occurs downstream of both prion replication and calpain activation, and suggests the existence of additional pathways of toxicity that remain operational even after scavenging ROS.
How can GDL execute such a faithful molecular mimicry of prion infection? We favor the hypothesis that GDL and PrPSc share the same docking site on cellular PrPC. Engagement of the latter site enacts a long-range allosteric transition of the FT, which in turn triggers the toxic cascade. The above scenario cannot be tested directly because of the technological barriers still hampering structural studies of PrPSc, yet it is at least compatible with the structure of the POM1:PrPC complex, as determined by X-ray crystallography [31,32].
Solforosi et al [33] claimed that anti-PrP antibodies induced toxicity by crosslinking PrPC, as F(ab)1 fragments were innocuous in their study. However the monovalent scFv and F(ab)1 fragments of antibody POM1 lead to toxicity in vitro and in vivo [15]. This refutes clustering of PrPC as a cause of toxicity in the present study.
Thus far, quests for anti-prion therapeutics have been rare and mostly unsuccessful. Two crucial reasons are the hazards associated with prion infectivity and the dearth of rapid, validated models of prion-induced toxicity. The validation of GDL as prion mimetics will help identify novel nodes in the pathogenetic cascades leading to neurodegeneration, and it is likely that some of these nodes may represent druggable targets.
We also identified differences between the pathogenesis of GDL exposure and that of prion infections. Firstly and most glaringly, the kinetics of GDL-induced neurodegeneration (days) is much faster than that of experimental prion infections (months). Secondly, although GDL function as a prion mimetic, they differ from bona fide prions by not inducing the classical misfolding and aggregation of PrP, by failing to induce deposition of protease-resistant PrP, and most crucially by being non-infectious. Thirdly, compounds that interfere with prion replication, such as pentosan polysulfate, do not alter the toxicity caused by GDL. All of these observations are compatible with the interpretation that GDL, while not leading to the generation of PrPSc, trigger the same signaling pathway as prion infections.
The different speed of disease development may be taken to suggest that GDL exposure and prion infections are fundamentally dissimilar. In our view, however, this does not contradict the hypothesis that these two models share pathogenetic pathways. Prion infection is initiated by trace amounts of prions within brain homogenates, with the PrPSc concentration only gradually increasing upon infection of a progressively larger numbers of host cells.
The prion isolate used for a standard slice culture infection contains 7.9*106 ID50 in 10 μl [12]. Antibodies in slice culture experiments were used at 67 nM in 1 ml of medium, which corresponds to 4.0*1013 molecules. We conclude that, in the case of the antibodies, the exposure to the bioactive principle exceeds that of prions by ca. 7 logs. This calculation is conservative since it disregards that prion inocula were removed after exposure, and that large prion aggregates are unlikely to efficiently penetrate tissues. Hence the bioactive principle in antibody preparations exceeds that in prion infections by several orders of magnitude, which yields—in our view—a highly plausible explanation for the difference in the kinetics of neurodegeneration.
Finally, recent evidence suggested the involvement of the unfolded protein response in prion-induced cell death in vivo [9]. Here, we confirm its involvement in RML and GDL-induced cell death in COCS. All of the above suggests that GDL-induced neurodegeneration represents a phenocopy of bona fide prion infections. If this conjecture is correct, targeted manipulations of the FT may be beneficial against neurotoxicity in prion infections. Indeed, we found that binding of the FT by antibodies was neuroprotective to prion-infected slice cultures, yet did not appreciably reduce prion titers—indicating selective suppression of the cytotoxic events downstream of prion replication. Therefore, binding of the FT could modify the course of the disease by uncoupling prion replication from prion toxicity. This hypothesis remains to be validated.
Since the GDL-induced toxicity model closely mimics multiple aspects of prion-induced neurotoxicity of prion-induced neurotoxicity, it seems reasonable to utilize GDLs for phenotypic screens aimed at identifying potential antiprion therapeutics. While confirmatory counterscreens will still require proof of efficacy against infectious prions, we posit that GDL toxicity may form the basis of convenient high-throughput and non-biohazardous assays of chemical and biological libraries.
All compounds were purchased from Sigma/Aldrich unless otherwise stated.
Prnpo/o;tga20+/+ (tga20) mice were on a mixed 129Sv/BL6 background [16,34]. C57BL/6 mice were used as a wild-type mouse strain.
10-week old tga20 mice were administered acetylated hydroxy tyrosol in drinking water ad libitum (2 g l-1 with an approximate intake of 8 ml daily). After 7 days of treatment, mice were anesthetized and intracerebrally inoculated with the 22L prion strain (30μl of 1% homogenate into the temporal cortex). Prion-inoculated animals were examined every second day and euthanized upon reaching pre-specified criteria for the terminal stage of disease. All prion-inoculated mice developed typical signs of scrapie and prion infection was confirmed in all cases by western blotting for protease-resistant PrPSc with the anti-PrP antibody POM1 (S8 Fig.) [14].
Cultured organotypic cerebellar slices were prepared as previously described [11]. Briefly, cerebella from 10–12 day old pups were cut into 350 μm sections and kept in Gey’s balanced salt solution (GBSS) (NaCl 8 g l–1, KCl 0.37 g l–1, Na2HPO4 0.12 g l–1, CaCl2 2H2O 0.22 g l–1, KH2PO4 0.09 g l–1, MgSO4 7H2O 0.07 g l–1, MgCl2 6H2O 0.210 g l–1, NaHCO3 0.227 g l–1) supplemented with the glutamate receptor antagonist kynurenic acid (1 mM) at 4°C. Six to ten slices were then plated per Millicell-CM Biopore PTFE membrane insert (Millipore) and residual buffer was removed before placing the inserts into a cell culture plate containing “slice-culture medium” (50% vol/vol MEM, 25% vol/vol basal medium Eagle and 25% vol/vol horse serum supplemented with 0.65% glucose (w/vol), penicillin/streptomycin and glutamax (Invitrogen)). Culture medium was exchanged thrice weekly and tissue cultures were kept in a humidified cell culture incubator set to 37°C with 5% CO2.
Antibody treatment and prion inoculations were performed as previously described [11,15,35]. Briefly, for antibody experiments, POM1 was spiked into the medium 10–14 days post-culturing, a time point at which COCS had recovered from acute phenomena associated with tissue dissection. Fresh POM1 was provided at every medium change. Cultures were harvested for biochemical analyses or fixed for immunocytochemical analyses at different time points.
For prion experiments, immediately after dissection, free-floating sections were incubated with infectious brain homogenates for 1 h at 4°C. Sections were then washed twice in 6 ml GBSSK, and 6–10 slices were transferred onto a 6-well Millicell-CM Biopore PTFE membrane insert (Millipore). Residual buffer was removed and inserts were placed into a 6-well culture plate and incubated in standard slice culture medium. POM2 treatment (335nM) was initiated after plating and re-supplied at every medium exchange.
For antibody experiments, drug treatment was initiated at the time of antibody addition (10–14 days post-culturing), whereas for prion experiments, drug treatment was started at 21 dpi when PrPSc is detectable in the cultures. Drugs were re-supplied at every medium change. As a control, the toxicity of each compound was tested in parallel in IgG-treated slices and NBH-inoculated slices. The drugs and concentrations used were (+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d] cyclohepten-5,10-imine maleate (MK-801, 20 μM), 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX, 20 μM), cyclosporine A (1 μM), ascorbate (1.5 mM), isoascorbate (1.5mM), N-(3-methyl-5-sulfamoyl-1,3,4-tiadiazol-2-ylidine)acetamide (methazolamide, 10 μM), MnTBAP (100 μM), benzyloxycarbonyl-Val-Ala-Asp (OMe) fluoromethylketone (zVAD-fmk, 40 μM), diphenyleneiodonium chloride (DPI, 5 μM), N-([3-(Aminomethyl)phenyl]methyl)- ethanimidamide dihydrochloride (1400W, 20 μM), N-benzyloxycarbonyl-L-leucylnorleucinal (calpeptin, 20 μM), N-acetylcystein (NaC, 1 mM), (2S,3S)-trans-epoxysuccinyl-L-leucylamido-3-methylbutane ethyl ester (E64d, 15 mM, Bachem), 1-[(5-(p-Nitrophenyl)furfurylidene)amino]-hydantoin sodium salt (Dantrolene, 10 μM). A summary table of the used tool compounds and their biological targets are reported in S6 Table. The drugs and concentrations used for anti-prion compounds were pentosan polysulphate (PPS, 300 ng ml-1, generously provided by Bene Pharmachem), Congo red (1 mg ml-1), and amphotericin B (4.5 mg ml-1).
Inserts containing the slices were transferred to new plates containing PBS for washing (twice) and tissue was then scraped off the membrane using 10 μl per slice of lysis buffer (0.5% sodium deoxycholate (DOC), 0.5% Nonidet P-40 (NP-40) supplemented with complete mini protease inhibitor cocktail (Roche) and PhosphoStop (Roche) in PBS). The harvested tissue was homogenized by trituration using a 30G syringe and protein concentrations were measured using the bicinchoninic acid assay (Pierce). Samples were mixed with loading buffer (NuPAGE, Invitrogen) and heated at 95°C for 5 min. Equal volumes were loaded (10 μg proteins per lane) and separated on a 12% Bis-Tris polyacrylamide gel or for higher molecular weight proteins, on a 4–12% gradient gel (NuPAGE, Invitrogen), and transferred onto a nitrocellulose membrane. These membranes were blocked with 5% w/vol Topblock (Fluka) in TBS-T (Tris-buffered saline supplemented with Tween20 (150 mM NaCl, 10 mM Tris HCl, 0.05% Tween 20 (vol/vol)) for 1 h and incubated with primary antibodies diluted in 1% Topblock in TBS-T at 4°C overnight. After 4 washes of 15 minutes each with TBS-T, membranes were incubated with secondary antibody diluted in 1% Top Block in TBS-T for 1 h at RT. Primary mouse monoclonal antibodies used were: POM1 mouse IgG1 antibody raised against PrPC (anti-PrPC; 200 ng ml–1), anti-α-fodrin (AA6, 100 ng ml-1, Millipore), anti-GAPDH (200 ng ml-1, Millipore), anti-actin (200 ng ml-1, Chemicon) and anti-calnexin (1:3000, Enzo Life Sciences). Secondary antibodies were horseradish peroxidase (HRP)-conjugated rabbit anti–mouse IgG1 (1:10,000, Zymed), and goat anti–rabbit IgG1 (1:10,000, Zymed). The following rabbit monoclonal antibodies were used: anti-phospho PERK (Cell Signal 3179S), anti-PERK (Cell Signal 3192S), anti-phospho eIf2α (Cell Signal 9721S), anti-eIf2α (Cell Signal 9722S), and anti-ATF4 (Cell Signal 11815S).
Blots were developed using SuperSignal West Pico chemiluminescent substrate (Pierce) and signals were detected using the VersaDoc system (model 3000, Bio-Rad) or Fuji. Quantification of band intensities was performed using Quantity One 4.5.2 software (Biorad). For specific detection of PrPSc, 20 μg of protein were digested with 25 μg ml-1 proteinase K in 20 μl final volume of digestion buffer (0.5% wt/vol sodium deoxycholate and 0.5% vol/vol Nonidet P-40 in PBS) for 30 min at 37°C [11]. Loading buffer was added and samples were boiled at 95°C for 5 min to inactivate PK. PNGase F treatment was performed using a commercially available kit, according to the manufacturer’s protocol (New England Biolabs). In brief, 10 μg of proteins was treated with 2 μl denaturation buffer in a 20 μl sample volume and incubated for 15 min at 95°C. A reaction mixture containing 2.6 μl G7, 2.6 μl NP-40 (10%) and 0.5 μl PNGase was added to the samples and incubated for 2h at 37°C. Samples were then boiled in presence of loading dye, and subjected to western blot analyses.
For immunofluorescence staining, organotypic slices were rinsed twice in PBS and fixed in 4% formalin overnight at 4°C. After washing, membrane inserts were incubated in blocking buffer (0.05% vol/vol Triton X-100 and 3% vol/vol goat serum in PBS) for 1 h and incubated with primary antibodies diluted in blocking buffer at 4°C for 3 days. The primary antibodies and concentrations used were mouse anti-Neuronal Nuclei (NeuN, 1 μg ml-1, Serotec), and directly conjugated mouse anti-NeuN-Alexa488 (0.5 μg ml-1, Millipore). The primary antibodies were detected using Alexa-conjugated secondary antibodies (3 μg ml–1, Molecular Probes). Membrane inserts were washed four times with PBS and the counterstaining agent 4,6-diamidino-2-phenylindole (DAPI) (1 μg ml–1) was added during the third washing step. Membranes were cut and mounted with fluorescent mounting medium (Dako) on a glass slide. Images were taken at identical exposure times with a fluorescence microscope (BX-61, Olympus) equipped with a cooled black/white CCD camera using a 4x objective. Morphometric analyses were performed to quantify the area of immunoreactivity using image analysis software analySIS vs5.0.
For PI incorporation, slices were rinsed with PBS and incubated for 30 min with PI (5 μg ml-1). Live images were recorded at 5x magnification using a fluorescent microscope (Axiovert 200) equipped with a cooled CCD camera using a 5x objective and processed using image analysis software analySIS vs5.0.
The lucigenin conversion assay was carried out at room temperature (RT). Inserts containing 5–10 slices each were washed in PBS and lysed with a 30G syringe in Krebs-Ringer solution supplemented with complete mini protease inhibitor cocktail (Roche). 50 μl of tissue lysate was transferred to a 96-well white microplate containing 175 μl assay solution and 0.25 μl lucigenin (10 mM) per well. Background measurements were performed using a chemiluminescence reader prior to the addition of 50 μl NADPH (1 mM) to each well. Subsequently, the NADPH-dependent signal was read and subtracted. Data are presented as relative light unit mg-1 total protein (each bar: average of 4 inserts ± s.d.).
For DHE conversion measurements, slices were inoculated, incubated for 40 dpi, and washed twice in GBSS. They were then incubated in GBSS containing DHE (10 μg ml-1). After 20 minutes of incubation at RT, 3 images/slice were recorded by live fluorescence microscopy using Axiovert 200 equipped with a cooled CCD camera and using a 10x objective. Three images were recorded per slice in three individual folia of the cerebellum. Fluorescence of DHE oxidation products was assessed by morphometry using constant thresholds.
In vivo assessment of ROS production followed the protocol described by Murakami et al [36]. Thirty minutes prior to euthanasia, mice were injected intraperitoneally with 200 μl DHE, and brain tissue was homogenized in 50 mM KH2PO4, 1 mM EGTA, and 150 mM sucrose. Fluorescence of DHE oxidation products was measured in 250 μl of 2% (w/v) homogenates using a fluorimeter with Ex/Em 485/585nm and a cutoff of 570. Relative fluorescence units were normalized to protein concentration.
Prion-susceptible neuroblastoma cells (subclone N2aPK1) [22,23] were exposed to 300 μl cerebellar slice homogenates, with 6 replicates per dilution, in 96-well plates for 3 days. Cells were subsequently split three times 1:10 every 3 days. After the cells reached confluence, 25’000 cells from each well were filtered onto the membrane of ELISPOT plates, treated with PK (0.5 μg ml–1 for 90 min at 37°C), denatured, and infected (PrPSc) cells were detected by immunocytochemistry using alkaline phosphatase-conjugated POM1, mouse anti-PrP, and an alkaline phosphatase-conjugated substrate kit (Bio-Rad). We performed serial ten-fold dilutions of experimental samples in cell culture medium containing healthy mouse brain homogenate. Scrapie-susceptible PK1 cells were then exposed to dilutions of experimental samples ranging from 10–4 to 10–7 (corresponding to homogenate with a protein concentration of 10 μg ml-1 to 0.01 μg ml-1), or to a 10-fold dilution of RML or healthy mouse brain homogenate. Samples were quantified in endpoint format by counting positive wells according to established methods [22,23].
One-way ANOVA with Tukey’s post-hoc test for multi-column comparison, or Dunnett’s post-hoc test for comparison of all columns to a control column, were used for statistical analysis of experiments involving the comparison of three or more samples. Paired Student’s t-test was used for comparing two samples. Results are displayed as the average of replicates ± s.d.
COCS were exposed to RML/NBH or POM1/IgG for various time points; for each time point and treatment, four cell culture inserts (n = 4) with 10 slices were used. RNA was extracted from 10 slices per insert using TRIZOL reagent (Invitrogen, USA) and purified with RNeasy columns (Qiagen, USA). Quality was assessed using BioAnalyzer (Agilent US). Labeled cDNA was fragmented and hybridized to GeneChip Mouse Genome 430 2.0 Array (Affymetrix, USA) which contains 45 000 probe sets. The data was analyzed with R/Bioconductor. Preprocessing and normalization was done using the RMA algorithm [37] and differential expression was assessed using the limma [26] package.
The nCounter Analysis system has been introduced previously [38]. Briefly, for each gene of interest, two sequence-specific probes are designed. The probes are complementary to a 100-base region of the target mRNA. The first probe is covalently linked to an oligonucleotide containing biotin (capture probe), and the second probe is covalently linked to a color-coded molecular tag that provided the signal (reporter probe). Forty-nine probe pairs for test genes and control genes were contained in the nCounter CodeSet.
All mouse experiments were carried out according to Swiss law and conducted under the approval of the Animal Experimentation Committee of the Canton of Zurich (permits 200/2007, 90/2013 and 130/2008). The animal care and protocol guidelines were obtained from http://www.blv.admin.ch/themen/tierschutz/index.html?lang=en and strictly adhered by the experimenters and animal facility at the institution where the experiments were performed.
|
10.1371/journal.pcbi.1006751 | Modelling the transport of fluid through heterogeneous, whole tumours in silico | Cancers exhibit spatially heterogeneous, unique vascular architectures across individual samples, cell-lines and patients. This inherently disorganised collection of leaky blood vessels contribute significantly to suboptimal treatment efficacy. Preclinical tools are urgently required which incorporate the inherent variability and heterogeneity of tumours to optimise and engineer anti-cancer therapies. In this study, we present a novel computational framework which incorporates whole, realistic tumours extracted ex vivo to efficiently simulate vascular blood flow and interstitial fluid transport in silico for validation against in vivo biomedical imaging. Our model couples Poiseuille and Darcy descriptions of vascular and interstitial flow, respectively, and incorporates spatially heterogeneous blood vessel lumen and interstitial permeabilities to generate accurate predictions of tumour fluid dynamics. Our platform enables highly-controlled experiments to be performed which provide insight into how tumour vascular heterogeneity contributes to tumour fluid transport. We detail the application of our framework to an orthotopic murine glioma (GL261) and a human colorectal carcinoma (LS147T), and perform sensitivity analysis to gain an understanding of the key biological mechanisms which determine tumour fluid transport. Finally we mimic vascular normalization by modifying parameters, such as vascular and interstitial permeabilities, and show that incorporating realistic vasculatures is key to modelling the contrasting fluid dynamic response between tumour samples. Contrary to literature, we show that reducing tumour interstitial fluid pressure is not essential to increase interstitial perfusion and that therapies should seek to develop an interstitial fluid pressure gradient. We also hypothesise that stabilising vessel diameters and permeabilities are not key responses following vascular normalization and that therapy may alter interstitial hydraulic conductivity. Consequently, we suggest that normalizing the interstitial microenvironment may provide a more effective means to increase interstitial perfusion within tumours.
| The structure of tumours varies widely, with dense and chaotically-formed networks of blood vessels that differ between each individual tumour and even between different regions of the same tumour. This atypical environment can inhibit the delivery of anti-cancer therapies. Computational tools are urgently required which facilitate a deeper understanding of the relationship between blood vessel architectures and therapeutic response. We have developed a computational framework which integrates the complex tumour vascular architecture to predict fluid transport across all lengths scales in whole tumours. We apply our model to two tumour cell-lines and show that differences in their inherent vascular structures influence flow through cancerous tissue. We also use our platform to predict the fluid dynamic response following vascular normalization therapy in realistic, static tumour networks and show that the response is dependent on tumour vascular architecture. We hypothesise that therapy may alter the permeability of interstitial tissue to fluid transport and show that lowering interstitial fluid pressure is not a necessary therapeutic outcome to increase tumour perfusion.
| Architectural heterogeneities in cancerous tissue limit the delivery of anti-cancer drugs by inhibiting their ability to circumnavigate the entire tumour to all cancerous cells [1]. In solid tumours, drug penetration to the tumour core is hindered by physiological barriers which can limit the delivery of targeted agents, with penetration exacerbated by the size of the agent [1–5]. Consequently, preclinical tools which provide a better understanding of therapy interactions within the tumour microenvironment are urgently required in order to increase treatment efficacy. In silico modelling is one such tool which can meet this need by testing novel therapeutic strategies at a much faster rate and cheaper cost than preclinical experimentation [6].
For a systemically-administered agent to effectively target diseased tissue, it must travel from the site of delivery to the site of disease, whilst minimally interacting with normal tissues and not degrading [7]. This is difficult to achieve in tumours since atypical endothelial proliferation of tumour vasculature leads to spatial variations in vascular density and branching patterns, distorted and enlarged vessels, and a highly convoluted network topology [8–10]. Further, vascular permeability is heightened and heterogeneous and so these immature blood vessels are generally leakier than those in normal tissue [3, 11].
The irregular microenivronment is typically characterised by hypoxia, acidosis and elevated interstitial fluid pressure (IFP) [12–14], which drive both tumour vascular proliferation and resistance to therapy [15]. Here, drug delivery may be hindered by the atypical nature of the tumour interstitium. The extracellular matrix (ECM) consists of a cross-linked dense network of collagen and elastin fibres, far denser than usually seen in normal tissue [16]. A denser matrix can inhibit oxygen and nutrient delivery, as well as providing significant resistance to the advection and diffusion of therapeutic particles [1], since key determinants of intratumoural fluid and mass delivery include tissue hydraulic conductivity and vascular compliance [17]. Several therapeutic interventions have sought to limit the effects of these physical barriers by manipulating the microenvironment to enhance the delivery of macromolecular agents [16, 18]. For example, normalising the tumour vasculature to reduce vessel permeability thereby increasing drug penetration [12]; and manipulating the connective tissue, and therefore interstitial hydraulic conductivity, using a platelet-derived growth factor (PDGF) antagonist to reduce tumour IFP [19].
Heterogeneities in the underlying morphology of tumours, such as vessel diameters and lengths, and inter-branch distance, exist across individual tumours and tumour cell-lines [20]. These variations in tumour architecture lead to spatial variability in drug efficacy, which complicate efforts to design effective treatment strategies [7]. Experimental efforts have been made to understand the effects of tumour heterogeneity on fluid interactions across tumours, for example, wick-in-needle has been used to measure IFP across tumours [21–23]. However, this method disturbs the local microenvironment and only provides an IFP measurement at individual locations. Non-invasive methods have also been developed to estimate tumour IFP [24, 25]. For example, convection-MRI, which measures low-velocity flow in tumours at a resolution of ∼ 250 μm in vivo [26]. However, these methods fail to capture full spatial maps of flow at the micron-scale which are crucial to understanding how the combined intra- and extravascular spatial flow heterogeneities occurring at the scale of blood vessels affects the macro-scale flow dynamics and consequent delivery of drugs within a solid tumour. Biomedical imaging complemented by in silico methods provides scope to provide such detail.
Mathematical models have been used to investigate the tumour microenvironment and have provided detailed insights which may otherwise be unavailable experimentally. Seminal models have indicated that a leaky tumour vasculature induces elevated IFP, reduced fluid penetration into the interstitium [14, 27], and a non-uniform distribution of drug delivered to solid tumours [2, 3, 11]. Further, they have defined conventional IFP profiles in tumours—a uniform pressure at the core, with a large decreasing gradient towards the periphery. However, these models generally average spatially over the tumour vasculature and so fail to capture the micron-scale flow dynamics; and they assign a fixed pressure boundary condition on the periphery of the tumour which may artificial induce these conventional IFP profiles.
Subsequent studies have sought to incorporate the spatially heterogeneous effects of tumour vasculature using computer-generated synthetic networks which retain key features of tumour vascular architecture [28–34], or by integrating spatial variations in vascular permeability parametrised against in vivo experimentation [35, 36]. More recently, a hybrid image-based framework has been developed which combines realistic tumour vascular architectures and in silico modelling to predict tumour vascular blood flow [37] and intravascular tumour oxygenation [38]. However, computational models are urgently required which incorporate these highly detailed data to enable predictions of fluid and mass transport to the surrounding tissue [6].
Recent advances in ex vivo optical imaging of cleared tissue specimens have enabled large samples (up to 2 cm3 with > 105 blood vessels) to be imaged in three-dimensions, at resolutions down to a few microns [39]. We have developed a platform called REANIMATE (REAlistic Numerical Image-based Modelling of biologicAL Tissue substratEs) which combines optical imaging of cleared tissue with mathematical modelling and in vivo imaging, within a unified framework, to generate quantitative, testable predictions regarding tumour transport [40]. The platform uses high-resolution imaging data from large, intact, optically-cleared tissue samples to make in silico predictions of blood flow, vascular exchange and interstitial transport. REANIMATE enables new hypotheses to be generated and tested in a manner that would be challenging or impossible in a conventional experimental setting. We have previously used REANIMATE to explore the impact of vascular network topology on fluid transport and vascular disrupting therapy (Oxi4503) to two colorectal cell-lines (LS147T and SW1222) [40].
We develop here a computational model to efficiently simulate both intra- and extravascular fluid transport across large, discrete microvascular networks [40]. Our model simulates Poiseuille flow through the vasculature using the optimisation scheme of Fry et al. [41], parametrised and validated against in vivo ASL-MRI data [40]. Following a similar Green’s function method for oxygen transport [42], the vascular component is coupled, via a discrete set of point sources of flux, to a Darcy model which simulates the effective fluid transport in the porous interstitium. A linear system is formed whereby only the source strengths need to be resolved, making it more computationally efficient compared to finite difference or element methods which require a spatial, numerically-discretised mesh [42].
In this study, we detail the generalised model which allows for spatially heterogeneous hydraulic conductances and conductivities. A comprehensive description is provided of its application to whole tumour vascular networks. We then apply our model to an orthotopic murine glioma (GL261) and a human colorectal carcinoma xenograft (LS147T) and reproduce physiological conditions observed in literature. We perform sensitivity analysis to the model parameters associated with transvascular fluid delivery, such as vascular hydraulic conductance and interstitial hydraulic conductivity, to explore the impact on the tumour IFP and interstitial fluid velocity (IFV) profiles.
Finally, we use our computational framework to explore the biomechanics underpinning vascular normalization. Vascular normalization is a method that applies anti-angiogenic therapy to restore tumour vascular structure and function to physiological levels [9, 43–45]. By modifying vascular architecture, therapy aims to normalize tumour perfusion and oxygenation, thereby increasing the efficacy of chemo, radio and immunotherapy [44, 46–48].
Preclinical and clinical evidence indicates that anti-VEGF (vascular endothelial growth factor) therapy creates a transient window of vessel normalization which improves tumour oxygenation and the delivery of therapeutic agents [15, 49]. However, the extent and window of normalization, and the duration and dosage of an anti-cancer drug varies with tumour type [18]. Further, the gold standard for detecting normalization of tumour blood vessels (such as perfusion, microvessel density, morphology and permeability) in the clinic is via histological staining [50].
Here, we recruit our in silico model in combination with realistic, static vasculatures to replicate fluid dynamics changes observed experimentally during vascular normalization [51–53] by modifying parameters such as blood vessel diameters, and vascular and interstitial permeabilities. In doing so we hypothesise which of these biomechanics are responsible for experimental observations following therapy, in addition to how tumour type, and the inherent differences in vascular networks structures, affects tumour IFP and perfusion following normalization therapy.
Orthotopic murine gliomas and human colorectal carcinoma xenograft from the GL261 and LS147T cell-lines (n = 6 for each), respectively, were grown subcutaneously in 8–10 week old, female mice. Following 10 to 14 days of growth, in vivo arterial spin-labelling MRI (ASL-MRI) was performed on a subset of GL261 and LS147T tumours, from which a mean tumour perfusion of 130 ± 50 and 19 ± 8 ml/min/100g was measured [40], respectively. Following perfuse-fixation, tumours were harvested, optically cleared and imaged using optical projection tomography (OPT, Bioptonics, MRC Technologies, Edinburgh). All experiments were performed in accordance with the UK Home Office Animals Scientific Procedures Act 1986 and UK National Cancer Research Institute (NCRI) guidelines [54]. Full details of the experimental protocol is provided in d’Esposito et al. [40].
Whole-tumour blood vessel networks were segmented from the OPT data for both tumour types. A combination of three-dimensional Gaussian and Frangi filters were applied to the data to enhance vessel-like structures allowing for the segmentation of the blood vessels from the background (see Fig 1a). Skeletonisation of these thresholded data was performed in Amira (Thermo Fisher Scientific, Hillsboro, OR), which also converted the data into graph format (interconnected network of nodes and segments with associated radii, see Fig 1b). To ensure that vessel structures were accurately represented, three-dimensional networks were visually inspected against two-dimensional imaging slices for an accurate representation of vessel location and thickness. Full details of the validation can be found in the Supplementary Material of d’Esposito et al. [40].
In this study a GL261 and a LS147T tumour network were chosen from the d’Esposito et al. [40] datasets for in silico development and testing. Vessel diameters ranged from 17.9 ± 9.3 and 8.9 ± 2.8 μm, with branching lengths of 68.7 ± 48.3 and 88.8 ± 49.4 μm, respectively (see Table 1). Vessel branching angles, inter-vessel distance, radii and tortuosity measures were consistent with data from previous studies that extracted vascular architectures using different methods [20, 40].
Our computational framework is compartmentalised into two models. The first predicts blood flow through the tumour vasculature and the second predicts interstitial fluid flow throughout the cancerous tissue through use of non-singular Green’s functions. Our method enables application to whole, large vascular networks (> 2 cm3 with > 105 blood vessels), thereby permitting predictions of whole tumour fluid dynamics which incorporate the inherent architectural heterogeneities occurring at the micron-scale.
The intravascular component incorporates the model of Pries et al. [55] to simulate vascular blood flow, where the structural properties of the segmented tumour networks and haemodynamic parameters are used as inputs. Flow or pressure boundary conditions at all terminal nodes in the vascular network are required to predict blood flow throughout the network. These boundary data are very challenging to measure in vivo, so we deploy the flow estimation algorithm of Fry et al. [41] to estimate boundary data based on the assumption that the microcirculation is regulated in response to haemodynamics stimuli relating to flow and shear stresses [56]. The scheme estimates unknown boundary conditions by minimising the squared deviation from specified target network wall shear stresses and pressures derived from independent information about typical network haemodynamic properties. In essence, the algorithm removes the need to define conditions at all boundary nodes, to one where simulation sensitivity is weighted towards the definition of these two target parameters. This enables physiologically realistic blood pressure and flow distributions to be estimated across an entire vascular network and has been applied to breast tumour [37], colorectal tumours [40], cortex [57], glioma [40] and skeletal muscle [58].
The second component to our computational model describes fluid transport through the porous interstitium using a Darcy model. Here, the vascular flow solution is coupled to Darcy flow via Starling’s law which describes fluid transport across the endothelium. The vasculature is represented by a discrete set of points sources of flux where the source strengths are defined by the vascular blood flow solution. A similar approach has been applied to simulate O2 transport across various tissues [42, 57, 59] and to predict capillary flow in the absence of these network structures [60]. Our generalised approach enables us to explore the affect of vascular architecture heterogeneity on fluid transport within the interstitium for large-scale vascular networks with spatially heterogeneous tissue and vessel wall permeabilities.
The following sections present the mathematics behind our model, followed by a description detailing its application to large tumour networks.
It remains practically infeasible to measure vascular flows and pressures in individual microvessels in vivo, which necessitates a pragmatic approach to boundary condition assignment. Under the assumption that vessels along the tumour surface are connected to peritumoural vessels [37, 70], we developed an optimisation procedure which assigns vascular pressures to tumour surface vessels, based on a target pressure drop, with iterative adjustments to match in vivo measurements of mean perfusion from ASL-MRI (see Fig 2a and 2b). These in vivo data are acquired for the same tumours that were subsequently subjected to OPT analysis. Using this approach, we are able to ensure good agreement between in silico predictions and measured perfusion data [40].
In this study a vascular pressure of 30 or 20 mmHg for the GL261 and 45 or 15 mmHg for the LS147T tumour was randomly assigned to 5% of surface boundary nodes, in order to meet the required tissue perfusion. To ensure randomness, the peritumoural nodes were represented by a list. The elements in the list were rearranged randomly using a uniform random number generator where the system clock was used to seed the random number engine. The nodes located in the top 5% of the list were then randomly assigned a low or high pressure using an equivalent randomised approach.
During preliminary simulations it was found that if high/low pressures were prescribed in close local proximity to each other, unphysiological flows were predicted due to the steep local vascular pressure gradient. In order to prevent this, a subroutine was designed so that values at opposing ends of the pressure spectrum could not be located within a defined vicinity of each other. This “exclusion region”, centred on a given boundary node, was defined as an ellipsoid volume with diameters, along its three principal axes, equal to 5% of the tissue dimensions.
A proportion of boundary nodes are contained within the tumour volume as a consequence of either angiogenesis which form blind-end vessels, or artefacts of OPT. Previous studies approximated the fraction of blind-end vessels in sample MCa-IV carcinomas to be 33% [71]. However, blind-end information is not available for either GL261 or LS147T tumours. Therefore, consistent with previous computational studies [37], blind-ends were randomly applied to 33% of remaining boundary nodes (using the previous randomised approach), with the remaining 62% of boundary nodes left as unknown in the flow optimisation scheme of Fry et al. [41].
Application of our boundary assignment method requires us to accurately compute mean perfusion across the tumours for a comparison against equivalent experimental data gathered in vivo using ASL-MRI. This requires an accurate definition of the tumour surface and volume to give an accurate approximation of a tumour’s mass and fluid flow into the tumour volume. For example, an overestimation of the tumour mass, assuming a cuboid tissue volume surrounding the tumour, can drastically underestimate tumour perfusion since, in this case, the tumour shapes are approximately ellipsoidal and perfusion is inversely proportional to the tumour mass. Similarly, an overestimation of flow into the tissue would overestimate tissue perfusion. Next we describe: 1) defining the surface of the tumour; 2) computing the IFV vectors across the tumour surface; and 3) approximating the total tissue perfusion.
1) The hull of a tumour is calculated using the Matlab (MathWorks Inc., Natick, MA) ‘boundary’ function applied to all nodes defined during vascular segmentation (see Fig 2c). The Matlab ‘fast loop mesh subdivision’ triangulation algorithm is then applied to further discretise the define tumour surface. 2) To approximate tumour perfusion requires us to define a set of normal vectors to the tumour surface to compute pressure gradients. We identify the centre of the tumour and duplicate the hull, which is then expanded to form a 10 μm gap between the two surfaces (see Fig 2d). IFP is then calculated across all nodal points on each surface, each paired by a vector normal to the opposing surface. A pressure gradient is then computed along each normal vector and the corresponding velocities are calculated. 3) A sphere packing algorithm is applied to the nodes on the original tumour hull, whereby no sphere overlaps neighbouring spheres (see Fig 2e). Any inflowing node (defined by the corresponding pressure gradient) is averaged across the great circle of its corresponding sphere and its contribution is summed together to calculate the interstitial component of tissue perfusion. Total tissue perfusion is calculated by summing over the peritumoural vascular inlets and interstitial perfusion values.
Finally, we prescribe baseline parameter values (see Table 2). We assume that the tumours were isolated in subcutaneous tissue in the absence of lymphatics, therefore the far-field pressure, p∞, was set to 0 mmHg. Due to a lack of experimental data, the vascular conductance and interstitial conductivity, Lp and κ, respectively, were given uniform values based on literature (2.8 × 10−7 cm mmHg−1 s−1 and 1.7 × 10−7 cm2 mmHg−1s−1, respectively [31, 75]). As the transport of blood plasma proteins is not modelled explicitly in our model, we assume a constant oncotic pressure gradient of 5 mmHg between the vasculature and interstitium, and set the oncotic reflection coefficient to a uniform value of 0.82 [72].
In the following section we apply our computational framework to a GL261 orthotopic murine glioma and a LS147T human colorectal carcinoma xenograft to form baseline flow solutions (see Figs 3 and 4). Next, we explore sensitivity to source parameters, which include source distribution, source radii and bilateral communication between the vascular and interstitial compartments. We then perform sensitivity analysis to the interstitial parameters (for example, vascular hydraulic conductance and interstitial hydraulic conductivity) on IFP and IFV profiles in the LS147T tumour. This is followed by in silico experiments to establish the tumour parameters, such as vascular and interstitial permeability, which change tumour IFP and perfusion following vascular normalization therapy, and show how vascular architectural heterogeneities affect treatment response.
Our vascular flow simulations are in good agreement with those in computational [37, 76] and experimental literature (see Table 3 and Fig 3). To test the variability induced by our stochastic boundary condition implementation, the optimisation procedure (detailed in Fig 2a and 2b) was repeated for a total of n = 12 for each tumour. Interstitial fluid flow was then simulated for each separate vascular flow solution, providing us with a baseline set of interstitial flow solutions. Blood flow across all simulations exhibited similar spatial distributions, with perfused vessels mainly restricted to the outer rim of the tumours [77]. The mean standard deviation of vascular blood pressures across all simulations was low with values of 0.82 and 1.53 mmHg (with maxima of 4.78 and 13.2 mmHg) for GL261 and LS147T, respectively (see Fig 5a and 5e). The elevated standard deviations in vascular pressure were located at the periphery of the tumours, which is to be expected as the high and low vascular pressures were stochastically assigned here. Furthermore, our mean blood velocity and vessel wall shear stresses agree with similar numerical modelling of vascular blood flow in the MDA-MB-231 breast cancer cell line [37].
Tissue perfusion (calculated as 83.7 ± 22.3 and 5.4 ± 0.3 ml/min/100g for the GL261 and LS147T tumours, respectively) was further validated by in vivo ASL-MRI measurements of 110 ± 70 and 19 ± 8 ml/min/100g for GL261 and LS147T, respectively. This implies that imposing physiologically realistic pressure boundary conditions generates physiologically realistic perfusion, and consequently accurate drug delivery solutions [40].
No literature IFP values were available for GL261 cell lines, however, our IFP was slightly elevated compared to that previously measured in LS147T in vivo (13.5 ± 11.3 mmHg [74]). Considering the range of IFP both here and in vivo [74] and the good accordance with in vivo perfusion, our results provided us with the confidence that our simulations can produce physiological IFP predictions (see Fig 4).
Examining the tumour radial IFP profiles, the LS147T network exhibited similar configurations as observed both experimentally in LS147T [17], in other cell lines [3, 24, 25], and in computational studies [31, 32, 36, 75], with elevated IFP at the tumour core (see Fig 4b and 4d). In addition, the LS147T network displayed a typical IFV profile radially, with an increasing IFV range towards the tumour surface due to the steeper pressure gradients at the periphery of the tumour. This indicated that bulk fluid filtration occurs at the high flowing vasculature located at the tumour extremity in this network (ρ = 0.41, p < 0.001, where ρ and p are the Pearson’s correlation coefficient and its corresponding p–value between tumour radius and vascular flow, respectively).
In comparison, the GL261 network also exhibited a traditional, yet steeper, IFP profile with a wider variance throughout the tumour (see Fig 4c). The IFV profile exhibited a typical profile increasing from the centre of the tumour to the periphery. However, its IFV peak was reached at ∼ 80% of the tumour radius, with a substantial decline in the latter 20% (see Fig 4c).
High IFP has been associated with low vascular density in A-07-GFP tumours [78]. Similarly here, we observed that GL261 and LS147T have distinct differences in their vascular architecture (see Fig 3 and Table 1) and that, in the case of LS147T, low variability in vascular density was associated with higher IFP (ρ = −0.584, p < 0.001—including in silico predictions from [40]). This may indicate that the inherent vascular architectural heterogeneity across tumour cell lines [20] directly impacts the IFP and IFV distributions, creating an unorthodox interstitial flow profile.
Here, we perform sensitivity analysis to understand the impact of parameter variance on predictions of tumour fluid dynamics. In doing so, not only do we understand the sensitivity to the model, but gain insight into the biological mechanisms at play.
For convenience we have split these parameters into two groups. The first we define as source parameters, which include the vascular blood pressure, pb,i and the source radius, r0,i for segment i, and the maximal spacing between each source along a given vessel, δ (see Fig 1c). The second group we call the interstitial parameters, which describe the biomechanical mechanisms which affect tissue transport. These include the oncotic reflection coefficient, σ, the hydraulic conductance of a vessel wall, Lp, the interstitial hydraulic conductivity, κ and the far-field IFP, p∞. In the following, if sensitivity analysis to an interstitial parameter was not being performed, it was set to the corresponding value in Table 2.
Vascular normalization is a therapeutic strategy used to restore tumour vasculature to a structural and functional state exhibited by healthy blood vessels [13, 15, 18, 49, 51, 79]. The impact of vascular normalization agents include a reduction in microvessel diameter, pruning of immature vessels, increased vascular maturity and pericyte coverage, and reduced vessel tortuosity [7, 51, 80]. As a result, normalization has been shown experimentally to lower tumour IFP [51, 52], enhance tumour vascular perfusion [52, 53] and improve drug delivery [48, 49, 81, 82]. However, the mechanistic links are missing which necessitates use of in silico modelling.
We next use our computational framework to investigate the effectiveness of vascular normalization in both GL261 and LS147T tumours where normalization therapy is administered intraperitoneally [48, 82]. We assume an ideal case in which vascular normalization has occurred in an axisymmetric manner whereby the effectiveness of the treatment increases from the tumour core to its periphery. To achieve this, the extent of stabilisation of vessel diameters and normalization of lumen permeability (through a combination of changes in Lp and σ), as a result of increase pericyte coverage, ranges linearly from typical tumour values in the core to physiological levels at the periphery. Here, vessel diameters were decreased by a factor of 1.99 [79], and set Lp and σ to 0.44 × 10−7 cm mmHg−1 s−1 and 0.91 [83], respectively.
Similarly to experimental [51, 52] and computational [81, 84] studies, vascular normalization reduced vascular perfusion (see Fig 7) and mean IFP (see S1 Fig) across both the GL261 and LS147T tumours. However, our predictions showed that IFP remained elevated at the core and that the overall reduction was induced by an increase in the pressure gradient across the tumours. This resulted in a small increase in IFV at the core of both tumours (see S1 Fig), yet changes in interstitial perfusion were minimal.
Our therapeutic predictions of vascular normalization are inconsistent with the magnitude of observed effects, such as reduction in IFP, compared to experimentation [52, 53]. This suggests that normalizing the permeability of the vessel lumen and stabilising vessel diameters are not significant therapeutic affects following vascular normalization. We advocate that vascular normalization reduces tumour IFP and increases perfusion via other factors, such as changes in interstitial permeability or a reduction in vascular network tortuosity.
Vascular normalization fortifies blood vessels through an increase coverage of mural cells and basement membrane [85]. Considering this and our sensitivity analysis for κ (see Fig 6d and 6h), we hypothesise that normalization locally alters the interstitial hydraulic conductivity of a treated blood vessel. We therefore sought to normalize interstitial hydraulic conductivity, in parallel and in the same manner as with vessel diameters, Lp and σ, to a value of 8.53 × 10−9 cm2 mmHg−1 s−1 [86] at the tumour periphery.
Normalizing interstitial hydraulic conductivity had substantial and contrasting consequences to fluid transport in the interstitium of both GL261 and LS147T tumours. In the case of GL261, IFP was elevated compared to baseline simulations throughout the tumour (see Figs 8a and 9a), with a mean increase of 70%. This, in conjunction with a small increase in the IFP gradient, led to elevated IFV in 80% of the tumour (see Figs 8a and 9a) and consistent patterns in tumour perfusion compared to baseline (see Figs 4a and 9b).
Treating LS147T significantly modified the IFP profile (see Fig 9d), elevating IFP by ∼ 9 mmHg at its core and developing a steeper gradient towards the periphery, in addition to a narrowing of IFP standard deviation (see Fig 8b). As a result, IFV increased throughout the tumour core (see Fig 9d), with an increase of 50% at its centre, and a predicted decreased compared to baseline in the outer 50% of the tumour (see Fig 8b).
To understand the response to treatment we analysed the source density, the ratio between sources and sinks of fluid flux, across each tumour type pre- and post-normalization. Fig 9c shows that GL261 is dominated by sources of flux pre-treatment, with the number of sources elevating at its core post-treatment. In comparison, a parity is observed between sources and sinks in the core of LS147T pre-treatment. Following normalization, the balance between sources and sinks drastically altered with no sinks existing between 0 to 0.3 of the LS147T radius, this is followed by a decline in the ratio to levels lower that baseline at the periphery, indicating an increase of sinks here.
Our results suggest that observed fluid dynamic changes following vascular normalization in literature [51, 52, 52, 53] may be as a result of local modification of the interstitial hydraulic conductivity (see S2 Fig). Further, that changes in perfusion may be a result of an steepening of the IFP gradient across a tumour and not necessarily a uniform reduction in IFP across the tumour. This suggests the importance of incorporating inherent spatial heterogeneities in vascular networks observed across tumour cell-lines [20] (see Table 1 and S3 Fig) into in silico studies of vascular normalization.
Elevated interstitial fluid pressure is frequently associated with solid tumours, where a conventional profile exhibits a uniformly high IFP in the core of a tumour decreases rapidly towards the levels of physiological tissue at the periphery [3, 14, 65]. This atypical characteristic forms a barrier to transvascular fluid and drug delivery, thereby diminishing therapeutic efficacy of anti-cancer treatments.
The passage of fluid through the interstitium is influenced by both hydrostatic and oncotic pressures in blood vessels and therefore by the heterogeneous architecture of tumour microvessels. However, the procurement of detailed fluid flow data in vivo across whole tumour networks is currently infeasible using conventional imaging and experimental techniques. However, recent advances in ex vivo high-resolution optical imaging techniques [39] allow whole three-dimensional tissue architectures to be extracted and reconstructed to act as inputs for detailed in silico modelling of fluid transport.
No study has explicitly modelled vascular and interstitial fluid flow using discrete, three-dimensional structural data from images of whole-tumour, real-world vasculature [6]. However, in a recent study, we presented our novel REANIMATE platform which extracts three-dimensional, whole tumour vasculature ex vivo from optically cleared tissue, which is then used to parameterise an in silico model of fluid transport guided by in vivo imaging data [40]. This has enabled us to perform quantitative, realistic predictions of fluid and drug delivery to tumours which has led to novel insights into a tumour’s inherent physical resistance to anti-cancer therapies [40].
We present a generalised framework to model vascular and interstitial fluid dynamics based on whole, explicit tumour vasculature, in a computationally tractable way. We detail its derivation and application to whole tumour vascular datasets and show how our model allows highly-detailed predictions of fluid flow within the tumour microenvironment by incorporating explicit tumour vasculature and spatially heterogeneous parameters, such as vascular and interstitial permeability. Our model allows flow heterogeneities to be quantified in a computationally efficient manner, when compared to finite-difference and element methods [42], and consequently can be applied to any vascular tissue.
We initially apply our framework to an orthotopic murine glioma and a human colorectal carcinoma xenograft from the GL261 and LS147T cell-lines, respectively, to present realistic, baseline simulations of the tumour microenvironment. We then perform sensitivity analysis to the underlying model parameters. The first are the source parameters, which are specific to our model, and include point source distribution and size. Secondly, the interstitial parameters, such as vascular conductance and interstitial conductivity, which are frequently represented in literature due to the prominent use of Starling’s law. In the second case, we perform analysis of how variation of these parameters modifies the IFP and IFV in an LS147T dataset. We finally mimic vascular normalization therapy, via varying model parameters, to generate hypotheses relating to fluid dynamic response observed in literature.
Our computational framework is based upon a Poiseuille model for vascular blood flow [41] which is coupled to a steady-state Green’s function solution to interstitial fluid flow. Here, tumour vasculature segmented ex vivo is represented by a discrete set of sources of fluid flux for bi-directional transport between the interstitium. Previous models either assume a homogeneous vascular network [3, 11, 14, 65], incorporate a computer-generated synthetic tumour network [28–34], or incorporate boundary conditions and spatial variations in tissue permeability to artificially represent vascular heterogeneity [35, 36]. However, vascular averaging methods do not fully encapsulate the intrinsic, local interactions between neighbouring blood vessels which contribute to global interstitial flow and synthetic networks are difficult to validate against real tumour architecture. Here we use vascular architecture from real, whole tumour networks, and through use of Green’s function methods, our model significantly reduces the computational size of the computational problem, allowing vessel-vessel interactions to be modelled at the micron-scale. Thus, we provide the means to perform in silico studies to hypothesis test the impact of vascular heterogeneity on the tumour microenvironment with relative ease.
Our simulations were performed on ex vivo structural imaging data from a GL261 and a LS147T tumour. As no in vivo flow or pressure data were available for the numerous boundary vessels, we developed a procedure whereby simulated data is optimised based on in vivo tissue perfusion data gathered using ASL-MRI [40]. This approach has produced solutions which are highly consistent with experimental measurements in the same tumours [40]. In this study, baseline vascular flow solutions across all tumour simulations are in good agreement with the perfusion data, alongside mean flows [76], velocities and vessel wall shear stresses [37], and fluid pressure in the interstitium [3, 17, 24, 25]. This provided validation that our model produces physiologically realistic results, providing a platform to investigate the tumour microenvironment.
We performed sensitivity analysis to the source parameters, such as updating the vascular flow solution, source distribution and source radii, to understand their influence on the flow communication between the vascular and interstitial domains. Our results exhibited a minimal sensitivity to IFP distributions by varying these parameters, with the exception of the assignment of source radii. Here, we hypothesise that greater care is required for spatially sparse tumours with a low vascular density, in order to ensure physiologically accurate simulations. Investigating the sensitivity to the interstitial parameters in the LS147T tumour, we found that raising the far-field interstitial pressure did not significantly alter the IFP distribution across the tumour, with increased IFP occurring at the tumour periphery, compared to baseline. Uniform variation in the oncotic pressure contribution, by modifying the oncotic reflection coefficient, only affected the magnitude of the interstitial pressure, with minor changes to the IFP and IFV profiles, similar to those reported in previous studies [2]. Increasing Lp and κ, raised the gradient of the interstitial pressure profile, thereby increasing fluid transport through the interstitium. However, our IFP distributions did not reach 0 mmHg at the periphery of the tumours as in previous studies [3, 14, 84]. This is due to not applying a fixed pressure at the tumour boundary in our model, which results in a smoother transition of IFP to the surrounding tissue, similar to previous computational modelling of tumour vascular heterogeneity [35, 36].
We next predicted the flow response following vascular normalization therapy mimicked by varying model parameters. Normalization of tumour vasculature was achieved by modifying vascular hydraulic conductances, oncotic reflection coefficients and stabilising blood vessel diameters in a linear radially varying fashion in both tumour networks. Our results indicate that whilst normalisation of lumen permeability and vessel diameters exhibited similar trends compared to experimentation such as a reduction in vascular perfusion post-treatment [52, 53], the magnitude in IFP reduction was minimal. Similarly, therapy which soley reduces vessel leakiness has been shown not to be effective in silico using synthetic tumour vasculatures [34].
As vascular normalization has been shown to stabilise the tumour vasculature by increasing mural cell and basement membrane coverage [85], we hypothesise that these structural changes alter the interstitial hydraulic conductivity. Subsequent simulations observed significant and unique changes in IFP, IFV and perfusion across each tumour cell-line. In GL261, we predicted an increase in IFP across the entire tumour, whereas in LS1247T, IFP exhibited a steep pressure gradient and elevated IFP in its core. Consequently, IFV was elevated in both tumour cores. We noted that this significantly altered the interstitial perfusion map in LS147T, with elevated perfusion at the tumour core, which dissipated towards its periphery.
We have shown that incorporating realistic tumour vasculature is key to accurately predicting the spatial flow heterogeneities induced by blood vessels. Our results suggest that reducing tumour IFP is not necessary to increase tumour interstitial perfusion and that developing a IFP gradient across the tumour is the key response. Further, we hypothesise that vascular normalization alters interstitial hydraulic conductivity within tumours. However, as our tumour vascular networks are static, we cannot discern the relative impact caused by a reduction in vascular tortuosity observed experimentally following therapy [51, 80]. Notwithstanding, we show that normalizing the tumour interstitial environment may provide an effective means to increase tumour perfusion.
Our results and proposed computational framework offer significant scope for future expansion. For example, recent in vivo methods provide a step forward in approximating conditions by measuring interstitial fluid velocity at the macro-scale [26]. These data could lead to greater accuracy when assigning boundary conditions specific to a tumour. Further, parameter values such as the vascular conductance and interstitial hydraulic conductivity were assigned using previous literature values since these tissue-specific measurements can be challenging to procure through experimentation. For example, interstitial conductivity values across healthy tissue have been reported to span four orders of magnitude [17]. New methods need to be developed to accurately quantify these values.
There are also opportunities to expand the computational model to incorporate more complex biological phenomena. For example, incorporating tumour compression of vessels due to increasing shear stresses within the tumour [29, 87], volumetric tissue growth and tumour angiogenesis which would allow us to develop vascular normalization treatment strategies which determine when to administer therapy during the normalization window [12].
We expect to find a wide utility for REANIMATE in a range of disease areas, particularly given the current interest in optical clearing methods and their widespread use in biomedical research. Our in silico framework is novel and timely and will find extensive use for hypothesis testing, to enable tumour biology and drug delivery to be better understood, which in turn may enable the next generation of cancer therapies.
|
10.1371/journal.pmed.1002844 | Kawasaki disease and 13-valent pneumococcal conjugate vaccination among young children: A self-controlled risk interval and cohort study with null results | Kawasaki disease is an acute vasculitis that primarily affects children younger than 5 years of age. Its etiology is unknown. The United States Vaccine Safety Datalink conducted postlicensure safety surveillance for 13-valent pneumococcal conjugate vaccine (PCV13), comparing the risk of Kawasaki disease within 28 days of PCV13 vaccination with the historical risk after 7-valent PCV (PCV7) vaccination and using chart-validation. A relative risk (RR) of 2.38 (95% CI 0.92–6.38) was found. Concurrently, the Food and Drug Administration (FDA) conducted a postlicensure safety review that identified cases of Kawasaki disease through adverse event reporting. The FDA decided to initiate a larger study of Kawasaki disease risk following PCV13 vaccination in the claims-based Sentinel/Postlicensure Rapid Immunization Safety Monitoring (PRISM) surveillance system. The objective of this study was to determine the existence and magnitude of any increased risk of Kawasaki disease in the 28 days following PCV13 vaccination.
The study population included mostly commercially insured children from birth to <24 months of age in 2010 to 2015 from across the US. Using claims data of participating Sentinel/PRISM data-providing organizations, PCV13 vaccinations were identified by means of current procedural terminology (CPT), Healthcare Common Procedure Coding System (HCPCS), and National Drug Code (NDC) codes. Potential cases of Kawasaki disease were identified by first-in-365-days International Classification of Diseases 9th revision (ICD-9) code 446.1 or International Classification of Diseases 10th revision (ICD-10) code M30.3 in the inpatient setting. Medical records were sought for potential cases and adjudicated by board-certified pediatricians. The primary analysis used chart-confirmed cases with adjudicated symptom onset in a self-controlled risk interval (SCRI) design, which controls for time-invariant potential confounders. The prespecified risk interval was Days 1–28 after vaccination; a 28-day-long control interval followed this risk interval. A secondary analytic approach used a cohort design, with alternative potential risk intervals of Days 1–28 and Days 1–42. The varying background risk of Kawasaki disease by age was adjusted for in both designs. In the primary analysis, there were 43 confirmed cases of Kawasaki disease in the risk interval and 44 in the control interval. The age-adjusted risk estimate was 1.07 (95% CI 0.70–1.63; p = 0.76). In the secondary, cohort analyses, which included roughly 700 potential cases and more than 3 million person-years, the risk estimates of potential Kawasaki disease in the risk interval versus in unexposed person-time were 0.84 (95% CI 0.65–1.08; p = 0.18) for the Days 1–28 risk interval and 0.97 (95% CI 0.79–1.19; p = 0.80) for the Days 1–42 risk interval. The main limitation of the study was that we lacked the resources to conduct medical record review for all the potential cases of Kawasaki disease. As a result, potential cases rather than chart-confirmed cases were used in the cohort analyses.
With more than 6 million doses of PCV13 administered, no evidence was found of an association between PCV13 vaccination and Kawasaki disease onset in the 4 weeks after vaccination nor of an elevated risk extending or concentrated somewhat beyond 4 weeks. These null results were consistent across alternative designs, age-adjustment methods, control intervals, and categories of Kawasaki disease case included.
| Kawasaki disease is an acute vasculitis that primarily affects children younger than 5 years of age. It is the leading cause of acquired heart disease in children in developed countries, and its etiology is unknown.
Vaccine Safety Datalink investigators found a nonstatistically significant elevated risk of Kawasaki disease after immunization with 13-valent pneumococcal conjugate vaccine (PCV13).
Spontaneous reports of Kawasaki disease after PCV13 have been received by the Vaccine Adverse Event Reporting System.
On the basis of these observations, the Food and Drug Administration (FDA) requested that Kawasaki disease after PCV13 vaccination in young children be studied in the claims-based Sentinel/Postlicensure Rapid Immunization Safety Monitoring (PRISM) system.
We used health insurance claims data of children 0 to <2 years of age from 6 Sentinel/PRISM Data Partners from 2010 to 2015.
Board-certified clinicians used medical records to adjudicate the Kawasaki disease cases for the primary study design. The adjudicators were blinded to the cases’ vaccination status.
The primary analysis used confirmed cases in a self-controlled design with a risk interval of Days 1–28 after vaccination, adjusting for the varying risk of Kawasaki disease by age.
A number of secondary analyses were conducted, including cohort analyses, analyses including cases of different levels of diagnostic certainty, analyses with alternative risk and comparison intervals, analyses using different populations for age adjustment, and a temporal scan statistical analysis.
More than 6 million doses of PCV13 were administered to the study population.
Eighty-seven cases of confirmed Kawasaki disease were included in the primary analysis, which did not find evidence of an increased risk of Kawasaki disease in vaccinees (relative risk [RR]: 1.07; 95% CI 0.70–1.63; p = 0.76).
Results of all secondary analyses were also null.
No evidence was found of an elevated risk of Kawasaki disease in the 28 days after PCV13 vaccination nor of an elevated risk extending or concentrated somewhat beyond 28 days.
| In 2000, the Food and Drug Administration (FDA) licensed the first pneumococcal conjugate vaccine (PCV), 7-valent PCV (PCV7; Prevnar; Wyeth), to protect young children against invasive disease caused by any of 7 serotypes of Streptococcus pneumoniae. Inclusion of PCV7 in the recommended child immunization program at 2, 4, 6, and 12–15 months of age resulted in decreased rates of invasive pneumococcal disease [1,2]. In early 2010, the FDA licensed a second vaccine, PCV13 (Prevnar 13; Wyeth), with the same vaccination schedule as PCV7, to replace PCV7 and protect against 6 additional serotypes that cause invasive pneumococcal disease in young children [3].
Although prelicensure trials of PCV7 and PCV13 found no increased risk of serious adverse events [4], postlicensure surveillance raised questions about a possible association between PCV13 and Kawasaki disease, an acute, self-limited vasculitis with a predilection for the coronary arteries and the leading cause of acquired heart disease in children in the United States. When Centers for Disease Control and Prevention (CDC)-sponsored Vaccine Safety Datalink investigators monitored the safety of the PCV13 vaccine during the first 2 years of life with respect to 8 health outcomes, a statistically significant result for Kawasaki disease after PCV13 was identified in the second of 12 group-sequential tests. The investigators conducted an end-of-surveillance analysis restricted to chart-confirmed cases and found a nonstatistically significant relative risk (RR) of 2.38 (95% CI 0.92–6.38) in the 0 to 28 days following vaccination with PCV13 compared with after PCV7 [5]. The FDA completed a safety review of the first 18 months of licensure of PCV13 under the FDA Amendment Act of 2007 Section 915, which included an analysis of the Vaccine Safety Datalink study results as well as an evaluation of the Vaccine Adverse Event Reporting System proportional reporting ratios for Kawasaki disease. The review resulted in an FDA Postmarket Safety Evaluation Summary posting that stated that there had been reports of Kawasaki disease following administration of PCV13 and that the FDA intended to initiate a larger study [6]. This report describes the now-completed study conducted in the Postlicensure Rapid Immunization Safety Monitoring (PRISM) program, a component of the FDA-sponsored Sentinel Initiative [7].
Six Sentinel/PRISM Data Partners, all in the US, contributed data: Aetna; Harvard Pilgrim Health Care; HealthCore (Anthem); Humana; OptumInsight LifeSciences; and Vanderbilt University School of Medicine, Department of Health Policy. Further details about the Data Partners are available on the Sentinel website [8]; most of the population represented by these Data Partners is commercially insured and not concentrated in any particular region of the country, although a small portion is covered by Medicaid in Tennessee. The study population included children less than 2 years of age who were members of any of the 6 Data Partners during 2010 to 2015 and who met 1 of 2 other enrollment criteria: (1) were exposed to at least one dose of any PCV vaccine and continuously enrolled at the Data Partner from birth through at least 84 days after their first dose of any PCV vaccine, or (2) were unexposed to any PCV vaccine and were continuously enrolled at the Data Partner from birth through at least 144 days of age (60 days, the time that many infants would receive dose 1, plus 84 days), with at least one documented healthcare visit between 14 and 150 days of age (i.e., up to 5 months of age). Gaps of up to 45 days between birth and the start of enrollment were allowed.
To model Kawasaki disease risk by age for the age adjustment implemented in the primary analysis, we used the Kids’ Inpatient Database for 2009 of the Healthcare Cost and Utilization Project (HCUP), Agency for Healthcare Research and Quality [9].
PCV13 was identified by the Current Procedural Terminology (CPT) code 90670 and the National Drug Codes 00005197101, 00005197102, 00005197104, and 00005197105 since January 1, 2010. Unspecified PCV was identified by the CPT code 90669 and Healthcare Common Procedure Coding System codes G0009 and S0195. Unspecified PCV vaccine since September 1, 2010, was assumed to be PCV13, based on the date of approval for PCV13 (February 24, 2010) and the fact that by July 2010 Pfizer reported that >90% of its private shipments of PCVs were for PCV13 [10]. Previous work in the Sentinel/PRISM system has found that most routinely administered vaccines are well captured by claims data [11].
Potential cases of Kawasaki disease were identified by the International Classification of Diseases 9th revision (ICD-9) code 446.1 and the International Classification of Diseases 10th revision (ICD-10) code M30.3 (acute febrile mucocutaneous lymph node syndrome) in the inpatient setting in any position (e.g., primary diagnosis, secondary diagnosis, etc.). (More than 95% of patients with an initial diagnosis of Kawasaki disease are hospitalized [12].) Only the first code in 365 days in the inpatient setting for patients at least 365 days of age or the first ever code in the inpatient setting for those under 365 days of age was considered, in order to exclude follow-up visits for Kawasaki disease.
Medical records of all potential Kawasaki disease cases (as ascertained by the algorithm) occurring during the 70 days after PCV13 vaccination were requested. In addition, records of potential cases without any known PCV vaccination were sought. We requested the inpatient hospitalization record associated with the healthcare claim with Kawasaki disease diagnosis code, as well as an outpatient follow-up visit for Kawasaki disease, including records of echocardiograms and angiograms.
Three board-certified pediatricians served as case adjudicators. Twenty cases were double-adjudicated, using prespecified classification rules. Once all discrepancies were resolved and classification rules refined, the chart review process continued with a single review of all remaining cases. The adjudicators were blinded to potential cases’ vaccination history.
Diagnosis of Kawasaki disease was based on the American Heart Association diagnosis guidelines and the CDC case definition [13,14]. Cases of confirmed (“Level 1”) Kawasaki disease were defined as those meeting one of the following criteria: (1) ≥4 principal features and a fever (≥38.0°C) persisting ≥5 days or until administration of intravenous immunoglobulin if given before the fifth day of fever, or (2) <4 principal features, fever (≥38.0°C) of any duration, and coronary artery disease (aneurysm or dilation) detected by either echocardiography or coronary angiography. The 5 principal clinical features are (1) changes in the extremities (erythema of palms or soles, edema of hands or feet, and/or periungual desquamation in the subacute phase), (2) polymorphous exanthem rash, (3) bilateral conjunctival injection without exudates, (4) changes in lips and oral cavity (inflamed lips or throat, strawberry tongue, or dry/cracking lips), and (5) cervical lymphadenopathy (at least one lymph node ≥ 1.5 cm in diameter) [13,15].
Cases of “possible” (“Level 2”) Kawasaki disease were defined as those having evidence of 2 or 3 principal features and ≥5 days of fever. The “possible” Kawasaki disease category was of interest, because some principal clinical features are frequently absent in young infants [13], and a large proportion of our study population was under 12 months of age. “Inconclusive” Kawasaki disease was defined as 1 principal feature and ≥5 days of fever.
During their record review, adjudicators determined the timing of symptom onset relative to the date of the coded diagnosis.
For cases of Kawasaki disease without a prior PCV vaccination history in the administrative claims data, the immunization record was sought from the child’s primary care provider for verification.
The primary risk interval (or risk “window”) used in the analyses was Days 1–28 after any dose of PCV13, where Day 0 was the day of vaccination. This risk interval was also used in the Vaccine Safety Datalink sequential analysis of PCV13 [5] and is indirectly supported by evidence that, among siblings, more than one-half of second Kawasaki disease cases in each family developed within 10 days of onset of symptoms in the first case, a finding consistent with a shared environmental trigger (or consecutive triggering infections with short incubation periods) and a relatively short latency period of days or weeks rather than months after exposure [16]. In a secondary analysis, we considered a postvaccination risk interval of Days 1–42. The 42-day interval allowed us to address any concerns that, if PCV13 were associated with an increased risk of Kawasaki disease, the true period of increased risk might go beyond the first 28 days.
The Sentinel Initiative is a public health surveillance activity [22], thus this study was not under the purview of institutional review boards.
The study protocol (S1 Protocol), STROBE checklist (S1 STROBE Checklist), and SCRI analysis datasets (S1 Data, S2 Data, S3 Data, and S4 Data) are provided as supporting information.
A total of 6,177,795 doses of PCV13 vaccine were administered to the study population.
There were 206 potential cases of Kawasaki disease, all ascertained by the presence of the ICD-9 code 446.1, meeting the criteria for chart review. Medical records were obtained for 184 (89%) of these. Of the 184 cases for whom charts were obtained, 125 (68%) were determined by clinical adjudication to be confirmed Kawasaki disease, 29 (16%) were determined to be possible Kawasaki disease, 4 (2%) were considered inconclusive, 18 (10%) lacked the necessary information for adjudicators to make a determination, and 8 (4%) were ruled out. The case-confirmation proportion was thus 68% for Level 1 Kawasaki disease and 84% for Level 1 plus Level 2.
In the SCRI logistic regression analyses that used the prespecified control windows, there were 43 confirmed cases in the risk window and 44 in the control window. No evidence of an elevation in risk was observed in either the (primary) HCUP–age-adjusted analysis or the unadjusted analysis—the RRs were 1.07 (95% CI 0.70–1.63; p = 0.76) and 0.98 (95% CI 0.64–1.49; p = 0.91), respectively. Adding in the possible Kawasaki disease cases did not qualitatively change these findings—there were 53 confirmed or possible cases in the risk window and 53 in the control window, for an RR of 1.09 (95% CI 0.75–1.60; p = 0.64) in the adjusted analysis and a RR of 1.00 (95% CI 0.68–1.46; p = 1.00) in the unadjusted analysis (Table 2).
The post hoc SCRI logistic regression analyses that used Days 29–56 following vaccination as the control window for all doses and restricted cases for analysis to those in which the difference between the hospital admission date and adjudicated symptom onset date was ≤14 days produced similarly null results. There were 41 cases in the risk window and 50 in the control window, and the adjusted and unadjusted risk estimates for the confirmed cases were 0.89 (95% CI 0.59–1.34; p = 0.57) and 0.82 (95% CI 0.54–1.24; p = 0.35), respectively. When the possible cases were included with the confirmed cases, there were 50 cases in the risk window and 61 in the control window, giving adjusted and unadjusted risk estimates of 0.89 (95% CI 0.61–1.29; p = 0.53) and 0.82 (95% CI 0.56–1.19; p = 0.30), respectively (Table 2).
The cohort for the analysis using the Days 1–28 risk window contained 80 potential Kawasaki disease cases (based on claims) in the risk window and approximately 474,000 exposed person-years. The cohort for the analysis using the Days 1–42 risk window contained 145 potential cases in that risk window and approximately 711,000 exposed person-years. Both data sets had 598 potential cases in unexposed time and 2.7 million person-years of unexposed person-time.
The risk estimates of potential Kawasaki disease in the risk window versus in unexposed time were 0.84 (95% CI 0.65–1.08; p = 0.18) for the Days 1–28 risk window and 0.97 (95% CI 0.79–1.19; p = 0.80) for the Days 1–42 risk window.
The nondifferential misclassification of the outcome entailed in not restricting the cohort analyses to chart-confirmed cases introduced noise, biasing the risk estimates toward the null. Given that the risk estimates were <1, unbiased estimates would have been somewhat lower than those observed.
Fig 2 shows the temporal distribution of Kawasaki disease symptom onsets for the 91 confirmed cases during Days 1–56 post–PCV13 vaccination for which the difference between the hospital admission date and adjudicated symptom onset date was ≤ 14 days. The temporal scan statistical test found no statistically significant clustering of cases. The lowest p-value of any grouping was 0.34.
In this large study investigating the relationship between PCV13 vaccination and Kawasaki disease during the 1–28 days after vaccination, we found no evidence of an association. The study included 87 confirmed cases in the primary SCRI analysis and approximately 700 potential cases and more than 3 million person-years in the secondary cohort analyses. The results were consistently null across alternative methods of analysis and age adjustment, alternative control intervals, and alternative levels of diagnostic certainty included.
Our null results contrast with the elevated (albeit not statistically significant) point estimate of Kawasaki disease risk during Days 0–28 after PCV13 found by Vaccine Safety Datalink investigators (RR = 2.38; 95% CI 0.92–6.38), who used historical rates of Kawasaki disease after PCV7 for comparison [5]. We consider our results to be quite robust because of the size of the study, with 6 million doses; the self-controlled nature of the primary analysis; and the qualitatively similar results obtained in all our analyses, both primary and secondary.
The possibility that any true period of increased risk might extend or be concentrated somewhat beyond 28 days post-vaccination was taken into consideration in several ways: (a) the primary SCRI analysis, which used a control interval of Days 43–70 for Doses 3 and 4, (b) a cohort analysis using a risk interval of Days 1–42, and (c) a temporal scan statistical analysis to detect clustering of onsets in any 1- to 28-day-long period during Days 1–56. No evidence for an increased risk after vaccination was found in any of these analyses.
The main limitation of the study was that we lacked the resources to conduct medical record review for all 685 potential cases of Kawasaki disease during 2010–2015, and in applying criteria to limit the cases for chart review, we unintentionally excluded potential cases with hospital admission during Days 71–84 after vaccination. As a result, some true cases with symptom onset within 70 days of vaccination could have been missed. However, this would have led to a bias toward finding an increased risk in the (primary) SCRI analysis, in which a Days 43–70 control interval was used for Doses 3 and 4. Yet no statistically significant elevated risk was found. In effect, the shorter follow-up period strengthens the null result. Moreover, the post hoc sensitivity analysis using a Days 29–56 control interval for all doses also produced null results. The study was not powered to assess the risk of Kawasaki disease by dose number.
In summary, we found no evidence of an elevated risk of Kawasaki disease in the 4 weeks after PCV13 vaccination nor any evidence of an elevated risk extending or concentrated beyond 4 weeks. The consistency of the results across alternative designs, age-adjustment methods, control intervals, and levels of case confirmation included suggests that the null findings are highly robust.
|
10.1371/journal.pmed.1002457 | Lansoprazole use and tuberculosis incidence in the United Kingdom Clinical Practice Research Datalink: A population based cohort | Recent in vitro and animal studies have found the proton pump inhibitor (PPI) lansoprazole to be highly active against Mycobacterium tuberculosis. Omeprazole and pantoprazole have no activity. There is no evidence that, in clinical practice, lansoprazole can treat or prevent incident tuberculosis (TB) disease.
We studied a cohort of new users of lansoprazole, omeprazole, or pantoprazole from the United Kingdom Clinical Practice Research Datalink to determine whether lansoprazole users have a lower incidence of TB disease than omeprazole or pantoprazole users. Negative control outcomes of myocardial infarction (MI) and herpes zoster were also studied. Multivariable Cox proportional hazards regression was used to adjust for potential confounding by a wide range of factors. We identified 527,364 lansoprazole initiators and 923,500 omeprazole or pantoprazole initiators. Lansoprazole users had a lower rate of TB disease (n = 86; 10.0 cases per 100,000 person years; 95% confidence interval 8.1–12.4) than omeprazole or pantoprazole users (n = 193; 15.3 cases per 100,000 person years; 95% confidence interval 13.3–17.7), with an adjusted hazard ratio (HR) of 0.68 (0.52–0.89). No association was found with MI (adjusted HR 1.04; 95% confidence interval 1.00–1.08) or herpes zoster (adjusted HR 1.03; 95% confidence interval 1.00–1.06). Limitations of this study are that we could not determine whether TB disease was due to reactivation of latent infection or a result of recent transmission, nor could we determine whether lansoprazole would have a beneficial effect if given to people presenting with TB disease.
In this study, use of the commonly prescribed and cheaply available PPI lansoprazole was associated with reduced incidence of TB disease. Given the serious problem of drug resistance and the adverse side effect profiles of many TB drugs, further investigation of lansoprazole as a potential antituberculosis agent is warranted.
| A recent report describes preclinical laboratory findings showing lansoprazole has strong activity against M. tuberculosis, including drug-resistant strains.
Other proton pump inhibitors, omeprazole and pantoprazole had no such activity.
No clinical investigations of this possible protective association with lansoprazole have yet been reported.
We studied a cohort of new users of lansoprazole, omeprazole, or pantoprazole from the United Kingdom Clinical Practice Research Datalink to determine whether lansoprazole users have a lower incidence of tuberculosis disease than omeprazole or pantoprazole users.
Comparing 527,364 lansoprazole initiators with 923,500 omeprazole or pantoprazole initiators, lansoprazole users had a lower rate of TB disease with an adjusted HR of 0.68 (0.52–0.89).
No association was found with negative control outcomes; myocardial infarction (adjusted HR 1.04; 95% confidence interval 1.00–1.08) or herpes zoster (adjusted HR 1.03; 95% confidence interval 1.00–1.06).
In vitro, animal, and, now, clinical epidemiological data, all suggest that lansoprazole has activity against M. tuberculosis.
Pharmacodynamic and early phase clinical trials are warranted to assess whether lansoprazole, or its metabolites, might have a role in the prevention or treatment of M. tuberculosis infection or tuberculosis disease.
| In 2015, there were an estimated 10.4 million incident cases of tuberculosis (TB) globally resulting in approximately 1.4 million deaths [1]. There is little commercial or public investment in TB research and there are only six novel compounds currently in the TB drug development pipeline [2]. In 2015, there were an estimated 480,000 cases of multidrug-resistant TB [1]. Treatment regimens for drug-resistant TB (DRTB) are long and unpleasant, with serious side effects and poor outcomes [1,3,4].
Using a high throughput fibroblast survival assay [5], Rybniker and colleagues found the proton pump inhibitor (PPI) lansoprazole had activity against M. tuberculosis (MTB), including drug resistant isolates [6]. This activity was confirmed in murine models and was found to result from inhibition of the mycobacterial cytochrome bc1 complex, disrupting the respiratory chain [6]. Omeprazole and pantoprazole, other PPIs, had no activity against MTB. This may be attributed to lansoprazole being the only PPI with no substitutions on the benzimidazole ring. Such substitutions are not known to influence treatment efficacy for any existing PPI indication.
Developed for the treatment and prophylaxis of diseases exacerbated by gastric acid production, PPIs are among the most widely used drugs globally. Their side effect profile is favourable compared with drugs used to treat TB [7]. We used the United Kingdom Clinical Practice Research Datalink (CPRD) to compare the incidence of TB disease among individuals taking lansoprazole with that among individuals taking omeprazole or pantoprazole.
The study was approved by the London School of Hygiene & Tropical Medicine ethics committee (ref: 11880) and the Independent Scientific Advisory Committee (ISAC) of the Medicines and Healthcare Products Regulatory Agency. The final study protocol was made available to journal reviewers and is attached as supplementary material (S1 Protocol).
The CPRD contains anonymised data from UK general practitioners and includes approximately 8% of the UK population [8]. Information includes comprehensive recording of consultations, diagnoses, prescribed medicines, and basic demographics. Practices and patients are broadly representative of the UK population [8], and data quality is subject to rigorous audit. The data have been used to conduct over 900 published studies, and data validity has been shown to be high for a variety of diagnoses; rates of recorded TB are similar to notification data from Public Health England, suggesting good case ascertainment [9]. Over 50% of CPRD patients have their general practice records linked to Hospital Episode Statistics (HES). HES include all inpatient National Health Service (NHS) hospitalisations (coded using International Classification of Diseases [ICD]-10) [10]. The January 2015 version of CPRD was used (data available 9th Sep 1987–5th Jan 2015).
All patients aged 16 years or over with at least 12 months research-quality follow up in the CPRD were eligible. From this group, we identified a) all new lansoprazole users and b) all new omeprazole or pantoprazole users. In both cohorts, patients had to have at least 12 months prior registration with no previous record of receiving any PPI. No limit on minimum duration of PPI exposure was made. We used prescribing records to determine intended treatment duration for each prescription, and imputed the population level median if this information was missing. In the UK, long term courses of medication are issued in small batches, each batch being covered by an individual prescription. A typical prescription provides enough medication to last 1 month, and so a 12-month course of treatment would usually involve 12 prescriptions. Although PPIs are available without prescription in the UK, this is unlikely to influence our estimates of exposure as it is doubtful many patients would be prescribed a PPI and simultaneously obtain a different one without prescription. Patients were excluded if they: had <12 months prior follow up; prior use of a different PPI; were aged <16 years at start of PPI treatment; or prior diagnosis of TB.
All clinical records indicating TB disease were extracted using Read Codes listed in Supplementary material (S1 Text). The earliest record was taken as the date of diagnosis. It is well recognised that there can be a considerable delay between infection with M. tuberculosis and diagnosis of TB disease. This means the aetiologically relevant exposure is likely to be earlier than the recorded diagnosis. In a Dutch study, amongst people diagnosed with TB, the median period between infection and diagnosis was 1.26 years, albeit with substantial variability [11]. For these reasons, all TB onset dates were moved earlier by 12 months in the primary analysis. Follow-up time for TB noncases was also censored 12 months earlier, as the final period of follow up would otherwise become ‘immortal’ time during which an outcome could not occur. Patients with <12 months follow up therefore contributed no follow up to the primary analysis. Supplementary material (S1 Fig) depicts how typical patient timelines were affected by this offset.
Approximately three-quarters of TB cases in England are among foreign-born individuals [12]. Given the higher force of infection in most countries of origin, it is likely that much of this TB results from reactivation of latent infection acquired abroad. The interval between arrival in England and TB diagnosis varies considerably by country of birth [12]. Among well-established migrant communities, TB cases may be a result of infection many years prior. The biology of latent MTB infection is poorly understood [13] and, to our knowledge, there have been no attempts to quantify the interval between ‘reactivation of latent infection and onset of symptoms or TB diagnosis. Biologically, the doubling time of MTB might be expected to be similar in both instances, and the mycobacterial burden required for symptoms to become apparent might also be expected to be similar. This logic would suggest that the same offset should be applied. However, given the considerable uncertainty surrounding this assumption, sensitivity analyses were undertaken (see below). Any misclassification of exposure status might be expected to bias our effect estimate towards the null.
PPI users are likely to be less healthy than nonusers [14]. The choice of omeprazole and pantoprazole users as the comparator group should mean we are comparing groups of people with similar health at baseline. These drugs are considered by most clinicians to be essentially interchangeable. However, if perceived health also influences the choice of individual PPI, this could be difficult to detect and account for. To guard against the possibility that any lansoprazole effect is driven by an unmeasured imbalance in baseline health, as a posthoc check we analysed the cohort for two additional ‘control’ outcomes; myocardial infarction (MI; associated with poor health) and herpes zoster (associated with impaired immunity). There is no reason to suspect either outcome is influenced by PPI choice. Each outcome was defined as the first record of this outcome in the CPRD clinical or referral record.
Covariates explored as potential confounders included age, sex, calendar year, smoking behavior, body mass index (BMI), alcohol use, ethnicity, drug abuse, any prior use of inhaled/oral corticosteroids, travel vaccines and antimalarial prescriptions (proxies for travel to TB endemic regions), diabetes, poorly controlled diabetes (one or more measure of HbA1c > 9% in the previous year), rheumatoid arthritis (RA), inflammatory bowel disease (IBD), chronic obstructive pulmonary disease, asthma, chronic kidney disease (CKD), depression, leukaemia, lymphoma, myeloma, and recorded HIV infection. We also adjusted for two variables measured at the practice level: the proportion of new PPI users given lansoprazole each year and the index of multiple deprivation (IMD) score.
Of note, ciclosporin and tacrolimus, both immunosuppressive drugs, are thought to interact with omeprazole but not other PPIs. It is possible that users of these drugs (who are probably at higher risk of TB) will be given PPIs other than omeprazole. For this reason, patients with a history of use of either drug were excluded from the study population.
Each participant’s follow up time began at the first prescription for lansoprazole or omeprazole/pantoprazole. All subsequent time was classified as follows:
PPI exposed time included a 60-day period after estimated treatment end date, allowing for stock piling and nonadherence. People starting treatment with omeprazole/pantoprazole but later receiving lansoprazole transitioned to the lansoprazole group at that time and vice versa. End of follow up was the earliest of first recorded TB disease, death, transfer to a different general practice, or last data collection date.
Cox regression was used to estimate hazard ratios (HRs) and 95% confidence intervals, comparing all lansoprazole-exposed time against all omeprazole- or pantoprazole-exposed time. A crude model was constructed with just the main exposure variable, followed by a model adjusting for all potential confounders with complete data. We then investigated potential confounding by variables with missing data (smoking, BMI, alcohol, ethnicity, and CKD). To do this, we constructed models excluding people with missing data on each variable. These variables were only retained in the final model if their inclusion altered the HR by >5%, thereby ensuring the maximum possible sample size was achieved. We ensured that the proportional hazards assumption was met by examining Schoenfeld residuals. An interaction term was fitted to look for effect modification between lansoprazole exposure and age. TB incidence was also measured after PPI treatment had been discontinued in an analysis comparing past lansoprazole users with past omeprazole or pantoprazole users. For the negative control outcome of MI, additional adjustments were made for the following risk factors if they had been recorded at any point prior to starting the PPI: coronary heart disease, cerebrovascular disease, peripheral vascular disease, other atheroma, hypertension, heart failure, and statin use.
The burden of TB disease may vary from practice to practice due to differences in patient populations (e.g., urban versus rural). To see whether this variation was related to the chance of being prescribed lansoprazole, we calculated practice-level TB prevalence (ever recorded diagnosis) and practice-level proportion of lansoprazole, omeprazole, and pantoprazole prescribing accounted for by lansoprazole over the study period. Linear regression was conducted to describe any association between them.
At the request of referees, we explored indications for PPI prescribing. Indication is not recorded or linked with the prescribing record by GPs, so we instead examined diagnoses recorded on the day the PPI was first prescribed.
The following sensitivity analyses were undertaken. 1) To address possible delays in diagnosis or in recording the TB diagnosis in the primary care record and the known variability in the incubation period, TB onset was redefined as A) date recorded in CPRD, B) 2 years earlier than recorded, and C) 5 years earlier than recorded. Longer incubation periods than those described in the Dutch study might be expected in first generation migrants infected in their country of origin [11]. 2) CPRD-linked HES inpatient data were searched for ICD10 codes indicating TB (A15–A19). An analysis restricted to CPRD HES-linked practices and time periods was conducted, taking the earliest of CPRD- or HES-identified TB disease. 3) The time following the end of estimated treatment duration at which we assumed therapy ceased was extended from 60 to 90 days. 4) Follow-up was censored after cessation or switch of PPI therapy (cessation defined as longer than 60 days not covered by a prescription). 5) As ethnicity is not recorded for all patients, and is unlikely to be perfectly captured for those with a record, we conducted an analysis restricted to patients recorded as having any white ethnicity. 6) Very short courses of lansoprazole therapy may not have an impact on TB incidence; we therefore conducted an analysis excluding people who received less than a 28-day supply of PPI over the study period.
There were 527,364 new users of lansoprazole and 923,500 new users of omeprazole or pantoprazole, after exclusions were applied (see Fig 1). The intended treatment duration was missing for 1% of individual PPI prescriptions, and the population median of 28 days per prescription was imputed for these records. Table 1 shows the background characteristics of the patients using data recorded on or before the first PPI prescription. Mean total exposure to PPI was 408 days for lansoprazole users and 386 days for omeprazole or pantoprazole users. Of note, the lansoprazole group had more current smokers (23% versus 16%) and lansoprazole was used less frequently than omeprazole and pantoprazole in more recent years (2006 onwards). Otherwise, the groups were largely similar. The 10 most frequently recorded clinical signs, symptoms, and diagnoses on the day a PPI was first prescribed were all synonyms for dyspepsia, gastroesophageal reflux disease, or abdominal pain. This did not differ between lansoprazole users and omeprazole or pantoprazole users. The mean time between first PPI prescription and first record of TB diagnosis was 1.2 years for lansoprazole users (standard deviation = 2.0) and 1.1 years for omeprazole/pantoprazole users (standard deviation = 1.8).
In the primary analysis, with recorded TB dates moved earlier by 12 months, the rate of TB was lower with 10.0 cases per 100,000 person-years (pyrs) (95% confidence interval; 8.1–12.4) in people receiving lansoprazole compared with 15.3 cases per 100,000 pyrs (13.3–17.7) in those receiving omeprazole/pantoprazole (Table 2). The crude HR was 0.65 (0.51–0.84), with a fully adjusted HR of 0.68 (0.52–0.89). Censoring follow-up at the first evidence of a treatment break resulted in a similar effect, with an adjusted HR of 0.59 (0.36–0.97). Considering postexposure periods of time when patients received no PPI, there was no detectable difference in TB incidence between patients who had received lansoprazole and those who had received omeprazole/pantoprazole. The HR was 0.94 (0.73–1.20), using the same outcome definition as in the primary analysis (with the outcome date, again, brought forward by 12 months, Table 2).
All sensitivity analyses produced results consistent with the primary analysis. For the analysis of HES/CPRD TB outcomes, confidence intervals were substantially wider and crossed unity, reflecting the reduced size of the dataset when restricting to patients with linked data (Table 2). In sensitivity analyses including variables recorded only in a subset of participants (IMD, ethnicity and CKD), inclusion of these variables, in a complete case analysis, had little impact on the HR (S1 Table). An analysis restricted to white patients also found similar results (HR = 0.72; 0.51–1.03). There was no evidence of an interaction between lansoprazole use and age, (p = 0.90, S2 Table), and the analysis allowing 90 days without a prescription before assuming therapy had ceased gave very similar results to the primary analysis (S3 Table).
Using linear regression, we found no evidence of any association between practice level TB prevalence and the practice level proportion of PPI prescribing accounted for by lansoprazole (p = 0.98; see S2 Fig for scatterplot).
For the negative control outcome of MI, patients receiving lansoprazole had very similar rates to those receiving omeprazole/pantoprazole both during periods of PPI exposure (adjusted HR = 1.04; 1.00–1.08, see Table 3), and post-PPI exposure (adjusted HR 0.98 (0.93–1.03). A similar pattern was seen for herpes zoster, with an on-treatment adjusted HR of 1.03 (1.00–1.06), and a post-treatment adjusted HR of 1.01 (0.98–1.04).
In a large, validated, and nationally representative dataset, we have demonstrated a protective association between lansoprazole use and newly diagnosed TB disease with an adjusted HR of 0.68 (0.52–0.89) when compared with omeprazole or pantoprazole use. This association was not seen in past users of these drugs and no association was seen between lansoprazole use and our negative control outcomes of MI and herpes zoster.
We selected a cohort of new adult users of lansoprazole, omeprazole, or pantoprazole, and the only clinical exclusion criteria was for patients with previous exposure to ciclosporin or tacrolimus, in order to avoid a potentially biased sample of lansoprazole patients at increased risk of TB disease. This exclusion affected approximately 0.1% of otherwise eligible patients, with all other exclusions based on age, length of time under observation in the CPRD, or prior history of TB. Our estimate of the TB notification rate in the study population is very similar to the 10.5 per 100,000 pyrs measured by Public Health England in 2015 [12]. People prescribed PPIs are not a random sample of the general population; they tend to be older and have more morbidity, especially gastrointestinal disease [14]. Nonetheless, if the association we report here is due to the pharmacological action of lansoprazole described by Rybniker [6], we can think of no biological reason why this effect would not be seen in the wider population.
We did not anticipate identifying many true confounders as, to explain our results, these would need to be associated with both the choice of PPI and TB disease. Whilst multiple risk factors for TB are known, what matters here is whether they are also associated with the choice of one PPI over another. Clinicians consider PPIs broadly equivalent and would generally not consciously select a specific PPI based on patient characteristics. In the UK, choice of specific drugs within a class may be mandated by guidelines for regional groups of general practitioners or influenced by cost, the efficacy of marketing activities, prescriber preference, or habit. One possible alternate explanation for our results is that regional variation in prescribed PPI was associated with local rates of TB infection. However, we found no association between practice level prevalence of TB infection and likelihood of prescribing lansoprazole. Similarly, whilst both choice of PPI and TB incidence varied over time, adjustment for calendar year did not affect our results. We were unable to adjust for some TB risk factors such as country of birth, homelessness, imprisonment, or prior use of biologic therapies. Nonetheless, we were able to adjust for proxy measures of some of these factors such as ethnicity, history of RA or cancer, and socioeconomic status. Indeed, the striking lack of change in the estimated HR when we adjusted for many potential risk factors suggests that there is little confounding. The protective association we report here is consistent with lansoprazole having clinical activity against MTB in humans.
If causal, the protective effect demonstrated may underestimate that which could be achieved in MTB infection or TB disease. For example, although the patients in our analysis would mostly have been prescribed their PPI once daily, perfect adherence to this regimen is unlikely. Adherence to treatment may be better in individuals with a life-threatening infection, taking treatment for a discrete period of time, and receiving appropriate support.
Little previous work has been done to investigate the association between PPI use and TB. However, Hsu et al. [15] found an association between acid suppressing medication and an increased risk of TB disease, which on the surface appears at odds with our findings for lansoprazole. However, the association declined to null with increasing duration of therapy with either a PPI or a histamine H2 receptor antagonist. This points towards reverse causality as a possible explanation, whereby people with unrecognised early symptoms of TB may be prescribed an acid suppressant.
It is not possible to determine from our results whether the association we demonstrated was against TB resulting from recent infection or from reactivation of latent infection. Most TB in England is thought to result from reactivation [12]. Were some individuals to have had early active disease on starting lansoprazole, use of a single drug might have been insufficient to prevent the emergence of resistance during the course of ‘treatment’. TB disease (though not latent infection) is usually treated with a combination of different drugs.
The absence of a persistent effect of lansoprazole after stopping treatment, in the context of a population within which incident TB is most likely due to reactivation of latent M. tuberculosis infection, suggests that lansoprazole, at these doses, does not sterilise. This is also consistent with the in vitro data, which suggest lansoprazole metabolites are bacteriostatic [6]. Several drugs used in regimens to treat TB disease are also bacteriostatic [16], e.g., cycloserine and para-amino salicylic acid. We note that isoniazid, a key drug used both to treat latent M. tuberculosis infection and TB disease, did offer long-term protection in early randomised controlled trials in individuals with latent infection [17]. However, recent mathematical modeling studies suggest that M. tuberculosis infection probably persists in the majority of individuals after a course of isoniazid. These analyses used data from largely HIV positive individuals enrolled in randomised controlled trials in settings with a high burden of TB disease [18–20]. The biology of latent M. tuberculosis infection is complex and poorly understood [13]. The balance between bactericidal and bacteriostatic activity for particular drugs can vary depending on the dose of drug given and the metabolic state of the mycobacteria [16]. The precise nature of any effect of lansoprazole cannot be ascertained from our data.
The original in vitro work found lansoprazole to be acting as a pro-drug [6]. The metabolite with activity against MTB, lansoprazole sulfide, is produced via intracellular metabolism. Whilst we could not study the direct effects of lansoprazole sulfide, it is a stable metabolite and importantly has no activity against the gastric H+ K+ ATPase, the PPI drug target [6]. Therefore, it might be used to treat MTB with fewer off-target effects. Lansoprazole and its metabolites have a number of attractive properties. PPIs have a very favourable side effect profile, as compared with drugs currently used to treat TB [7]. There is no evidence that lansoprazole interacts meaningfully with drugs commonly used to treat TB or HIV [21]. Lansoprazole is off-patent with inexpensive generic versions available. In addition, in vitro studies suggest the drug might have activity against both drug sensitive and drug-resistant strains [6].
We were unable to assess whether TB diagnosis was associated with lansoprazole dose, and it is not known how the currently licensed dose of lansoprazole compares with the optimal dose for activity against MTB. The initial in vitro work by Rybniker et al. tested the pro-drug lansoprazole itself. Strong activity against MTB was detected at a lansoprazole concentration of 10 μM and half maximal activity (IC50) was observed at 1.47 μM, or 2.2 μM in a second cell line [5,6]. When given orally at a typical daily dose of 30 mg, Cmax for lansoprazole in human plasma is 1 mg/ml [22], with little change after dosing for five days. This corresponds to a concentration of 2.71μM [23]—i.e., a concentration in plasma greater than the estimated IC50. Further work to determine whether effective concentrations of the sulfide metabolite can safely be achieved in relevant tissues in humans may be needed (e.g., in granulomas or pulmonary cavities). However, very high doses of lansoprazole sulfide (300 mg/kg) have been tolerated by mice [6], suggesting it may be possible to treat humans with higher doses than individuals in CPRD would have been receiving.
In this study, use of lansoprazole was associated with a reduced incidence of TB disease compared with omeprazole or pantoprazole. To our knowledge, these are the first observations to suggest that lansoprazole may have clinical activity against MTB in humans. They are consistent with evidence from both in vitro and animal studies [5,6]. Given the problems of antimicrobial resistance and the adverse side-effects seen with many antituberculous agents, these results are welcome. Our results do not directly address the question of whether lansoprazole or its metabolites would be effective as part of a treatment regimen for MTB infection or TB disease. However, since lansoprazole is safe and well-tolerated, there is a strong case for efficacy studies in humans.
|
10.1371/journal.pntd.0002879 | Mycobacterium ulcerans Ecological Dynamics and Its Association with Freshwater Ecosystems and Aquatic Communities: Results from a 12-Month Environmental Survey in Cameroon | Mycobacterium ulcerans (MU) is the agent responsible for Buruli Ulcer (BU), an emerging skin disease with dramatic socioeconomic and health outcomes, especially in rural settings. BU emergence and distribution is linked to aquatic ecosystems in tropical and subtropical countries, especially to swampy and flooded areas. Aquatic animal organisms are likely to play a role either as host reservoirs or vectors of the bacilli. However, information on MU ecological dynamics, both in space and time, is dramatically lacking. As a result, the ecology of the disease agent, and consequently its mode of transmission, remains largely unknown, which jeopardizes public health attempts for its control. The objective of this study was to gain insight on MU environmental distribution and colonization of aquatic organisms through time.
Longitudinal sampling of 32 communities of aquatic macro-invertebrates and vertebrates was conducted from different environments in two BU endemic regions in Cameroon during 12 months. As a result, 238,496 individuals were classified and MU presence was assessed by qPCR in 3,084 sample-pools containing these aquatic organisms. Our study showed a broad distribution of MU in all ecosystems and taxonomic groups, with important regional differences in its occurrence. Colonization dynamics fluctuated along the year, with the highest peaks in August and October. The large variations observed in the colonization dynamics of different taxonomic groups and aquatic ecosystems suggest that the trends shown here are the result of complex ecological processes that need further investigation.
This is the largest field study on MU ecology to date, providing the first detailed description of its spatio-temporal dynamics in different aquatic ecosystems within BU endemic regions. We argue that coupling this data with fine-scale epidemiological data through statistical and mathematical models will provide a major step forward in the understanding of MU ecology and mode of transmission.
| Buruli ulcer, caused by the pathogen Mycobacterium ulcerans (MU), is a severe skin disease occurring in tropical and subtropical countries. Strongly associated to freshwater ecosystems and especially swampy and flooded areas, transmission of this bacterium within ecosystems and across multiple aquatic organisms is still an enigma. Here, we studied in depth the temporal and spatial variations of MU presence in freshwater ecosystems and aquatic organisms in two areas of Cameroon where Buruli ulcer is endemic. We found MU widely present across ecosystems and taxonomic groups along the year and we described a general trend for MU persistence in the environment. Moreover, the colonization dynamics of aquatic ecosystems suggest that each kind of ecosystem may have distinct favourable times of the year for MU presence. In addition to setting the scene for a preventive approach for humans based on ecosystem characteristics, this study suggests that MU transmission is the result of complex ecological processes between biotic and environmental factors. Such results call for an integrative approach in order to disentangle the respective contributions of aquatic organisms and environmental conditions on MU presence and persistence in the environment.
| Mycobacterium ulcerans (MU) is the agent responsible of Buruli ulcer (BU), an emerging human skin disease affecting human populations in tropical and subtropical regions [1]. While effective treatment is available through a combination of rifampicin-streptomycin for small lesions, with additional surgery required in some cases, early access to treatment is often an issue, especially in poor rural areas where most of the disease burden accumulates [2]–[4]. Absence or delay of treatment may cause irreversible deformity, long-term disabilities, extensive skin lesions, and even severe secondary infections [5]. Public health efforts for disease control require early detection of cases, but MU ecology and the conditions triggering human infection are poorly understood, which undermines our capacity to detect areas at risk.
Buruli ulcer has been present in Cameroon since the first reported cases in 1969 from the Centre Province, in the districts of Akonolinga and Ayos [6]. The highest BU prevalence in this region dominated by tropical rainforest is distributed along the Nyong River basin, where swampy and flooded areas prevail [7]. A second endemic site appeared in 2004 in Bankim (Adamaoua Province), a region at the border with Nigeria in a transition zone between forest and savannah. Within this region, the construction of a dam in 1989 resulted in a large area of flooded land in the district, and BU cases are mostly concentrated between this dam and the Mbam River [8].
Distribution of human cases around the world seems to be closely related to freshwater ecosystems, especially to areas of slow flowing or stagnant waters [9]–[14]. Furthermore, emergence of cases in many parts of the world has been associated to the creation of swamps and flooded areas either naturally after heavy rains [13] or under the pressure of human action, i.e. construction of dams or irrigation [8], [15]. Micro-aerobic conditions may promote MU growth [16] and genomic analyses suggest that MU has adapted to a restricted environmental niche, possibly an arthropod [17]–[19]. Favorable conditions in these types of environment are likely to drive MU growth and persistence and may ultimately affect the transmission to human populations. A direct transmission could take place from the environment where MU is present through existing wounds or passive inoculation [20]–[23]. However, a direct link between the type of ecosystem and MU abundance in the environment has never been shown.
The role of aquatic communities of macro-invertebrates and vertebrates as a fundamental part of the aquatic ecosystem on MU ecology and transmission is also unclear. Following detection of MU in the environment from abiotic, i.e. water, soil [24], [25] and biotic samples, i.e. plants, fishes [26], tadpoles [27], insect larvae [28], snails [29], and water bugs [30], it has been suggested that bacteria present in the aquatic environment (water, plant biofilm, mud, and detritus) could be concentrated by filtering and grazing invertebrates and then be transmitted through predation up to higher levels of the aquatic trophic web [31]. In addition, some specific taxonomic groups could act as keystone species in the transmission of MU within the aquatic ecosystem [32]. Finally, water bugs of the families Belostomatidae and Naucoridae (Order Hemiptera), which are voracious predators of aquatic organisms may get colonized through this trophic web and transmit the bacteria to humans through biting [30], [33]–[35].
In order to better understand such a complex disease system, it is essential to address its changes over time and space. Freshwater ecosystems are highly dynamic with seasonal variations in abiotic and biotic parameters impacting on aquatic community assemblages and structures [36], [37]. However, comprehensive field studies performed in Africa to date have addressed temporal dynamics but in only one taxonomic order, Hemiptera water bugs [38], or have focused on aquatic communities but neglecting their temporal dimension [39], [40]. As a result, detailed information on temporal dynamics of MU persistence and spread in the whole aquatic community is dramatically lacking.
Here, we address this issue by performing a large-scale sampling of multiple aquatic communities over space and time in two BU endemic areas of Cameroon, Akonolinga and Bankim. This study aims to improve knowledge on MU environmental distribution and colonization of aquatic organisms throughout the year, with two specific objectives:
Between June 2012 and May 2013, periodic sampling of aquatic communities was performed in Akonolinga and Bankim, two regions in Cameroon where BU is endemic [7], [8]. In order to track colonization dynamics, monthly samples were collected in Akonolinga. In addition, sampling was performed every three months in Bankim, allowing a description of a wider range of environmental characteristics (savannah and tropical rainforest). Within each region, selection of survey sites was done in a two-step procedure. Initially, we classified the villages in each region based on (i) BU human prevalence and (ii) surrounding environmental conditions, according to national health data and land cover data respectively. We pre-selected a number of villages that represented a gradient in both of these parameters within each region. In order to evaluate the relevance of these sites for the study, this pre-selection was followed by on-site visits of all water bodies surrounding the villages and discussions with the local population and health authorities (accessibility, land-use change, human use, persistence throughout the year, etc.).
In all, 32 water sites were selected (16 in each region), including a large variety of streams, rivers, swamps and flooded areas. Streams were defined as bodies of water with a current and were clearly confined within a bed of up to 30 m wide. They included both rainforest streams in Akonolinga and rainforest and savannah streams in Bankim. Rivers were larger than streams, and their margin was highly variable depending on the season, being up to several hundred meters wide in periods of intensive rainfall. They included the Nyong and Mfoumou rivers in Akonolinga, but not the Mbam river in Bankim due to very strong currents that prevented appropriate sampling. We considered as swamps all permanent wetlands with stagnant or very slow flowing waters, many of which were created as the result of roads blocking the natural course of a stream. Finally, flooded areas were temporary bodies of stagnant water formed either naturally after heavy rains in flat areas of forest or savannah, or artificially as in the case of the Mapé Dam in Bankim.
Sampling in each region was performed between 8am and 4pm during 5 consecutive days. In order to ensure comparability of the results, identical methods were carried out by the same persons for all sites throughout the study. In each water body, 4 locations were chosen in areas of slow water flow and among the dominant aquatic vegetation. The sample was limited to those places accessible by a person with waders (depth max. 1.50 m). At each location, 5 sweeps were done with a metallic dip net (32×32 cm, 1 mm mesh size) within a surface of 1 m2 and at different depth levels (down to a depth of 1 m). All the material collected was placed into a bucket with water and passed through a 3-layer filter (32×32 cm grid; 20, 5 and 1 mm mesh sizes, respectively) with abundant water. The material in the first two layers was placed in white rectangular basins, and visible aquatic organisms were identified on site, classified and stored separately into tubes with 70% ethanol. The material contained in the last layer, a mixture of plant debris and small invertebrates, was put into 150 ml flasks with 95% ethanol and brought to the laboratory, where identification of all other individuals in the community (larger than 1 mm) was done with the use of a binocular microscope.
Aquatic macro-invertebrates were classified down to the family level whenever possible, using taxonomic keys provided in the Guide to the Freshwater Invertebrates of Southern Africa series [41]–[47] and other relevant literature [48]–[51]. In order to avoid cross-contamination between samples, all the equipment used in the classification (forceps, basins, gloves, Petri dishes, etc.) was discarded or decontaminated with NaOH 1 M at the end of each sample classification.
Individuals from the same sample were pooled for PCR analysis by groups of aquatic organisms belonging to the same taxonomic group. Two pooling strategies were used to fulfill the purposes of our study. First, for all sites, we tested a total of 6 sample-pools for each month and each site in order to better describe spatio-temporal dynamics of MU presence. For this, we chose the 5 most abundant taxonomic groups in all sites (to allow for comparability of results) plus a sixth group that was different in each site (to gain representation of all groups), and we pooled all individuals of the same group. Second, we chose 10 sites, 5 sites in each region, for which we applied a more in-depth molecular analysis every 3 months in order to have a better characterization of MU presence in taxonomic groups. Within these sites, all individuals of each taxonomic group were distributed in 4 sample-pools, and all taxonomic groups were tested. The same 10 sites were used along the year and this subgroup presented a similar geographical and environmental variability as the larger group of 32 sites. A maximum of 2 g of pool weight was established in order to avoid excessive inhibition during the qPCR analysis. For each sample-pool, composition, number of individuals and weight of the pool were recorded.
Pooled individuals were all ground together and homogenized in 50 mM NaOH solution using Tissue Lyser II (QIAGEN). Tissue homogenates were heated at 95°C for 20 min. DNA from homogenized insect tissues was purified using QIAquick 96 PCR Purification Kit (QIAGEN), according to manufacturer's recommendations. 10% negative controls were included for extraction and purification.
Oligonucleotide primer and TaqMan probe sequences were selected from the GenBank IS2404 sequence [52] and the ketoreductase B (KR) domain of the mycolactone polyketide synthase (mls) gene (Table 1) from the plasmid pMUM001 [52], [53]. QPCR mixtures contained 5 µl of template DNA, 0.3 µM concentration of each primer, 0.25 µM concentration of the probe, and Brilliant II QPCR master Mix Low Rox (Agilent Technologies) in a total volume of 25 µl. Amplification and detection were performed with Thermocycler (Chromo 4, Bio-Rad) using the following program: 1 cycle of 50°C for 2 min, 1 cycle of 95°C for 15 min, 40 cycles of 95°C for 15 s and 60°C for 1 min. DNA extracts were tested at least in duplicates and the 10% negative controls were included in each assay. Quantitative readout assays were set up, based on external standard curve with MU (strain 1G897) DNA serially diluted over 5 logs (from 106 to 102 U/ml). Samples were considered positive only if both the gene sequence encoding the ketoreductase B domain (KR) of the mycolactone polyketide synthase and IS2404 sequence were detected, with threshold cycle (Ct) values strictly <35 cycles.
All statistical analyses were conducted using R statistical software, version 2.14.0 [54]. Maps were created using ArcGIS 10.0 and information displayed in them was obtained from the USGS Shuttle Radar Topography Mission (elevation data) [55], IFORA project (hydrographic network) and Institut National de Cartographie du Cameroun (roads). Data on rainfall was obtained from the NASA Tropical Rainfall Measuring Mission [56]. Pearson Chi Square tests were used to compare proportions of positive sample-pools coming from different types of ecosystems and p-values were computed by Monte-Carlo simulation. One-sample proportions tests with continuity correction were used to calculate the confidence intervals of the proportions. Associations of MU colonization dynamics of taxonomic groups or MU colonization of different ecosystems with rainfall patterns were investigated by calculating the cross-correlation of the time series two by two.
Despite the great health and socio-economic burden borne by people affected with BU, little is known about the ecology and mode of transmission of this disease. MU is embedded in an environment that is inherently dynamic, but information on spatio-temporal dynamics of MU persistence and spread is dramatically lacking. The results shown here represent a step forward in the understanding of MU ecology. They provide the first account of MU spatio-temporal dynamics in aquatic communities from a variety of ecosystems within BU endemic regions. We show first that MU is ubiquitous within these regions and can be found in all types of freshwater ecosystems, but swampy areas seem more favorable to MU presence, as demonstrated in Bankim. Then, we confirm that MU is present in nearly all taxonomic groups of the aquatic community, but we show that groups common in streams are minimally colonized. Finally, we demonstrate that MU has distinctive temporal dynamics in each ecosystem and taxonomic group, suggesting that MU occurrence is probably driven by complex ecological interactions between environmental abiotic and biotic factors.
We found that MU presence in Bankim was more restricted to the south of this area, especially between the Mapé Dam and the Mbam River, where BU cases concentrate [57] and more swamps and flooded areas prevail. The construction of the dam has been previously associated to the emergence of cases in the area [8], [22] and proximity to the Mbam River was found to be a risk factor in a case-control study [22]. However, our results suggest that swamps created along the road, rather than the flooded areas created artificially by the dam or naturally near the Mbam after heavy rains, are more favorable to the presence of MU. Swamps are characterized by stagnant waters with low oxygen and high temperatures, which may create conditions favorable to MU growth and specific fauna in which to develop [16]–[19]. Furthermore, while water level and conditions in flooded areas are highly variable throughout the year, swamps are more stable environments, which could influence the differences observed in these two stagnant ecosystems [58].
In contrast, MU is present everywhere across the Akonolinga region and all environments presented very similar positivity, although the highest positivity concentrated near the basin of the Nyong river. While climate, land cover or human modifications of the environment could be behind these disparate regional distributions, it could also reflect a spread of the bacteria over time. Indeed, it is possible that MU initially persists in the most favorable environments (swamps), as in the case of Bankim where cases have been reported for less than 10 years [8], spreading over time to other environments where water conditions and aquatic communities are less favorable and/or intermittent along the year, as in the case of Akonolinga where MU is endemic for more than 40 years [6]. Flying insects could be responsible of this dissemination as previously suggested [38], [59]. Out of the two taxonomic orders that are both aquatic in adult stage and capable of flight (Coleoptera and Hemiptera), only Hemiptera was found positive in all types of ecosystem. Indeed, this group was found positive to MU in 65% of the sites, more than any other group of the aquatic community (table S4).
MU is present in nearly every group of the aquatic community and no taxonomic group stands among others as the major host carrier of MU. Aquatic vertebrates and invertebrates, as well as semi-aquatic groups, are positive for IS2404 and KR, with similar pool prevalence. This is in line with the idea of multi-host transmission dynamics and more particularly a transmission through ecological webs, where some species can highly contribute to MU transmission without experiencing a significantly larger positivity [32]. Nevertheless, some patterns arise for several specific taxonomic groups. Firstly, the most positive order in terms of pool prevalence are Lepidoptera larvae (caterpillars), an invertebrate with semi-aquatic families mostly living and feeding on riverine aquatic plants [42]. This finding suggests that some aquatic plants might play an important role on MU persistence and development in the aquatic ecosystem or in ecotone areas, and be a source of infection for herbivorous invertebrates. Indeed, some plants could harbor MU in endemic regions [24], [60] and they stimulate its growth under experimental conditions [61]. Secondly, groups of aquatic invertebrates that were found mainly in streams such as Trichoptera and Decapoda are among the groups with the lowest pool prevalence. These findings support the hypothesis that MU might not be well adapted to environmental conditions in this type of aquatic ecosystems.
Regarding the seasonal dynamics, MU is present in freshwater ecosystems and aquatic organisms throughout the year but there are fluctuations both between seasons and within each season, as previously demonstrated for MU colonization of water bugs [38]. The highest peak in positivity appears in August and October (i.e. over the high rainy season), and then decreases progressively throughout the high dry season (November to February). These findings could be consistent with the idea of a run-off of bacteria into the aquatic environment during periods of intensive rainfall, as previously suggested [24], [62]. However, the lack of correlation between rainfall patterns and the dynamics observed for the various ecosystems and taxonomic orders highlights that more complex interactions might take place within the aquatic community. Differences in feeding habits may explain the distinct colonization dynamics of different orders. For instance, while Hemiptera were found positive all year long (except in May), Coleoptera were repeatedly found negative for more than half a year (Figure 5). These two orders share many common features: they have both larval and adult aquatic stages, many are capable of flight, and their abundance dynamics along the year are very similar (Figure S3 and S4). However, while most families of Hemiptera are voracious predators of aquatic organisms (only Corixidae feed on aquatic plants), families of Coleoptera present a large spectrum of feeding habits that include predators, shredders, scrappers, filtering collectors and omnivorous organisms [41], [42]. Laboratory experiments support the idea of a trophic transmission of MU through predation [30], [34], [63], [64] and a mathematical model studying MU prevalence within 27 aquatic communities in Ghana suggested that a transmission through ecological webs is more likely than a purely environmental acquisition from contaminated water [32]. Our results support the hypothesis that biotic interactions may play a role in MU transmission and that MU dynamics could result from a complex interplay between environmental abiotic factors and variations in community assemblages.
We show that important fluctuations in MU positivity take place within each particular ecosystem. For most sites, we checked for the presence of MU in a given month and site by analysing 6 pools of aquatic organisms. This may be insufficient to demonstrate the absence of the bacteria in the ecosystem, since pool positivity overall was lower than 10%. We attempted to increase the chances to detect MU by pooling all individuals of the most abundant taxonomic orders in the aquatic ecosystem, which allowed us to pool and analyze over 60% of the 238,496 individuals sampled without losing comparability of the results. Furthermore, disparate sampling strategies for each region could be behind the differences found between the types of environment for the two study regions. Bankim was only sampled 4 months of the year as opposed to 12 months in Akonolinga. Therefore, we cannot rule out the possibility that sampling in Bankim may have taken place at appropriate times of the year for swamps but not for the other environments in this region. We tried to avoid this by sampling in Bankim at regular intervals (every three months), therefore capturing a maximum of variability along the year.
This study reinforces the idea that MU persists in a wide range of locations [24], [40] and taxonomic groups [28] and the pool positivity rates described here (nearly 10% overall) are consistent with previous studies [8], [28], [38]. This ubiquity of MU and its persistence in the environment throughout the year contrast with the focal distribution and low number of BU human cases in endemic regions. A possible explanation is that while we are likely to be detecting one (or several) of the MU ecovars present in the environment (previously referred to as mycolactone producing mycobacteria), this does not necessarily imply that we are detecting strains of MU with pathogenic potential to cause BU in humans [17], [65]–[67]. Future studies comparing the strain diversity of environmental and human samples with molecular techniques such as SNP typing [68], [69] could shed some light on this issue. Furthermore, we rely as previous studies on qPCR amplification of KR and IS2404 sequences as an indicator of the presence of MU, which gives no certainty of whether the DNA detected belongs to viable mycobacteria. The lack of an appropriate technique to culture MU from the environment remains a major limitation of fieldwork studies. Nevertheless, qPCR remains the gold standard for environmental studies on MU ecology [8], [27], [38], [40]. An alternative hypothesis is that while the presence of MU in the environment reflects a potential risk for infection, many environmental and socio-economic factors may need to come together to enable MU transmission to humans. Sero-epidemiological studies have shown that a large proportion of the population living in endemic regions have been exposed to MU, but only a small fraction develop the disease [70]. Therefore, MU might only trigger BU disease under certain environmental conditions (a bacterial concentration threshold and/or contact with a competent, infected vector) or in subpopulations in high contact with potential sources of infection and with increased susceptibility to infection (due to immunity, hygiene, etc.).
In conclusion, this study provides for the first time a detailed characterization through space and time of MU presence in two BU endemic regions with distinct environmental conditions. The understanding of MU ecology to date is still limited, especially regarding the conditions that allow this mycobacterium to persist in the environment and be transmitted to humans. Our study attempts to complete previous approaches by sampling multiple aquatic communities over time in order to better understand the influence of aquatic ecosystems on MU presence and its dynamics along the year. The global trend we describe for MU dynamics could be the result of complex ecological processes, with interactions between environmental abiotic and biotic factors that require deeper analysis, something that is beyond the scope of this paper. However, we believe that coupling data produced by such field studies with fine-scale epidemiological data and integrated through statistical and mathematical models could provide a major step forward in the understanding of MU ecology and BU mode of transmission.
|
10.1371/journal.pmed.1002224 | Haemolysis in G6PD Heterozygous Females Treated with Primaquine for Plasmodium vivax Malaria: A Nested Cohort in a Trial of Radical Curative Regimens | Radical cure of Plasmodium vivax malaria with 8-aminoquinolines (primaquine or tafenoquine) is complicated by haemolysis in individuals with glucose-6-phosphate dehydrogenase (G6PD) deficiency. G6PD heterozygous females, because of individual variation in the pattern of X-chromosome inactivation (Lyonisation) in erythroid cells, may have low G6PD activity in the majority of their erythrocytes, yet are usually reported as G6PD “normal” by current phenotypic screening tests. Their haemolytic risk when treated with 8-aminoquinolines has not been well characterized.
In a cohort study nested within a randomised clinical trial that compared different treatment regimens for P. vivax malaria, patients with a normal standard NADPH fluorescent spot test result (≳30%–40% of normal G6PD activity) were randomised to receive 3 d of chloroquine or dihydroartemisinin-piperaquine in combination with primaquine, either the standard high dose of 0.5 mg base/kg/day for 14 d or a higher dose of 1 mg base/kg/d for 7 d. Patterns of haemolysis were compared between G6PD wild-type and G6PD heterozygous female participants. Between 21 February 2012 and 04 July 2014, 241 female participants were enrolled, of whom 34 were heterozygous for the G6PD Mahidol variant. Haemolysis was substantially greater and a larger proportion of participants reached the threshold of clinically significant haemolysis (fractional haematocrit reduction >25%) in G6PD heterozygotes taking the higher (7 d) primaquine dose (9/17 [53%]) compared with G6PD heterozygotes taking the standard high (14 d) dose (2/16 [13%]; p = 0.022). In heterozygotes, the mean fractional haematocrit reductions were correspondingly greater with the higher primaquine dose (7-d regimen): −20.4% (95% CI −26.0% to −14.8%) (nadir on day 5) compared with the standard high (14 d) dose: −13.1% (95% CI −17.6% to −8.6%) (nadir day 6). Two heterozygotes taking the higher (7 d) primaquine dose required blood transfusion. In wild-type participants, mean haematocrit reductions were clinically insignificant and similar with both doses: −5.8 (95% CI −7.2% to −4.4%) (nadir day 3) compared with −5.5% (95% CI −7.4% to −3.7%) (nadir day 4), respectively. Limitations to this nested cohort study are that the primary objective of the trial was designed to measure efficacy and not haemolysis in relation to G6PD genotype and that the heterozygote groups were small.
Higher daily doses of primaquine have the potential to cause clinically significant haemolysis in G6PD heterozygous females who are reported as phenotypically normal with current point of care tests.
ClinicalTrials.gov NCT01640574.
| Primaquine is the only widely available treatment to eliminate latent liver stages of P. vivax malaria and thereby prevent relapse (radical cure).
Primaquine can cause potentially severe haemolysis in glucose-6-phosphate dehydrogenase (G6PD)-deficient patients. Hemizygous males and homozygous females are identified reliably as G6PD deficient by current phenotypic testing; however, heterozygous females may have a normal or deficient result on G6PD rapid diagnostic phenotypic tests. G6PD heterozygous females who have a normal G6PD phenotypic test are still susceptible to clinically important haemolysis caused by primaquine.
This study was designed to describe and quantify the haematocrit changes in G6PD Mahidol heterozygous females with a normal G6PD phenotype when taking primaquine for the radical cure of P. vivax malaria.
The researchers administered a standard high-dose primaquine regimen (0.5 mg base/kg/d for 14 d) or an alternative higher-dose primaquine regimen (1 mg base/kg/d for 7 d) to participants infected with P. vivax malaria who tested as G6PD normal by the G6PD fluorescent spot test. Haematocrit levels were followed for 2 wk and compared with respect to the G6PD genotype, quantitative G6PD enzymatic activity, and initial parasitaemia.
Higher daily doses of primaquine (1 mg base/kg/d) were associated with significant haemolysis in female participants who were heterozygous for Mahidol variant G6PD deficiency.
The standard high dose primaquine (0.5 mg base/kg/d for 14 d) was well tolerated in all female participants who had a normal G6PD phenotypic test.
In the absence of close monitoring, daily doses higher than standard high dose primaquine (greater than 0.5 mg base/kg/d) should not be used in females who are heterozygous for a G6PD mutation that is similar to or more severe than the Mahidol variant.
More information is needed on the haemolytic risk in G6PD heterozygous women and girls with different G6PD enzymatic activity levels.
New rapid diagnostic tests for G6PD deficiency should be quantitative and include assessment of haemoglobin levels.
Safer regimens of primaquine are needed for G6PD-deficient patients.
| Primaquine is the only widely available drug that is effective against P. vivax hypnozoites, the latent forms that emerge from the liver to produce relapses of P. vivax malaria. The recommended regimen to prevent relapse of P. vivax malaria (called radical treatment) is primaquine given for 14 d at a daily dose of 0.25 to 0.5 mg base/kg. However, although widely recommended for radical treatment of P. vivax and P. ovale infections, primaquine is often not given. This is because primaquine causes haemolysis in glucose-6-phosphate dehydrogenase (G6PD)-deficient individuals [1]. While G6PD deficiency is very common in malaria-endemic areas, G6PD testing is generally unavailable because the standard point-of-care test requires appropriate reagents, electricity, trained staff, and quality controls.
The point-of-care diagnosis of G6PD deficiency is usually made by a phenotypic screening test in which the G6PD-mediated reduction of NADP+ to NADPH is measured semiquantitatively. The standard test is the fluorescent spot test (FST), which assesses the fluorescence of NADPH in a blood spot under ultraviolet light [2]. This identifies blood samples with ≳30%–40% activity as abnormal. It is considered safe to give primaquine (and other oxidant drugs) to persons who screen as having “normal” G6PD activity determined by the FST or other comparable rapid tests. The G6PD gene is on the X-chromosome. Hemizygous males and homozygous females have markedly reduced G6PD enzymatic activity and will therefore be identified reliably by these tests. However, in G6PD heterozygous females, random X-chromosome inactivation (Lyonisation) [3] results in red cell mosaicism. Individual red blood cells therefore express either a normal or deficient phenotype, so the overall blood G6PD activity level is an average determined by the relative proportions of the two red cell populations. The average proportion of deficient red cells in the population of heterozygotes is approximately half, but because random X-inactivation occurs early in embryogenesis, some women may have normal G6PD activity in the majority of their red cells, whereas others may have a majority of cells that are deficient. As a consequence, heterozygous females with a “normal” G6PD phenotype by the FST may have up to 60%–70% of their red blood cells that are G6PD deficient and susceptible to haemolysis by oxidant stresses.
Along the Myanmar border with northwestern Thailand, the prevalence of G6PD deficiency in males is 9%–18%. The Mahidol variant is most common (88% of all variants); other variants include Chinese-4, Viangchan, and Mediterranean type [4]. As part of a dose optimisation trial to assess potentially simpler primaquine regimens for the radical treatment of P. vivax malaria in patients screened as “G6PD normal,” we report patterns of haemolysis in relation to the daily dose of primaquine and the G6PD genotype.
The trial was approved by the Faculty of Tropical Medicine, Mahidol University Ethics Committee (MUTM 2011–043, TMEC 11–008), and the Oxford Tropical Research Ethics Committee (OXTREC 17–11), and it was registered on the ClinicalTrials.gov website (NCT01640574).
This nested cohort study was performed within a randomized nonblinded clinical trial in patients with acute P. vivax malaria. It was conducted by the Shoklo Malaria Research Unit (SMRU), which operates clinics extending over 120 km along the Thailand–Myanmar border. The clinics serve migrants and displaced persons of Burman and Karen ethnicities originating from as far as 30 km inside Myanmar.
Patients more than 6 mo old with symptomatic P. vivax mono-infection confirmed by microscopy were diagnosed in outpatient clinics and enrolled in the study if they or the carers of children < 18 y old gave fully informed written consent. Patients were excluded if they were G6PD deficient by FST or were pregnant or breastfeeding an infant ≤ 6 mo old, had severe malaria, an allergy to trial drugs, haematocrit ≤ 25%, or received a blood transfusion within 3 mo, or could not comply with the trial requirements. Enrolment investigations included malaria smear, haematocrit, G6PD FST, and full blood count.
Randomisation to each of four treatment arms was computer generated in blocks of 20. Assignment of the participant to a treatment arm was centralized; a phone call from the clinic-based study staff was made to an independent staff member not involved with the trial. Trial codes were given sequentially, and written confirmation was provided.
Participants were randomized to one of the following treatment regimens:
Treatment with all drugs began on the day of enrolment. Doses were calculated using charts with predefined weight bands (S1 Appendix). All primaquine doses were given after food and were supervised. Participants were seen weekly after treatment was completed.
Investigations included malaria smear, haematocrit, full blood count, G6PD phenotypic tests (described below), and G6PD genotyping for the prevalent Mahidol variant (487G>A). Capillary haematocrit was measured daily until day 7 and again on day 14. Haematocrit was measured on 30 μL of blood obtained via fingerstick using a heparinized capillary tube centrifuged at 10,000 rotations per minute (rpm) for 3 min and then read manually with a Hawksley Micro-Haematocrit reader. Anaemia requiring haematinic treatment was defined as an absolute haematocrit <30%. Haemoglobin variants were analysed using Capillary Electrophoresis (Capillarys2, SEBIA, France) in some patients with anaemia. The G6PD FST (R&D Diagnostic, Greece) was performed using 5 μL of blood mixed with 100 μL of reagents (containing haemolysing agents and NADP+ /G6P substrates). After 10 min of incubation at room temperature, a 15 μL aliquot was spotted on filter paper and allowed to dry in air. The spots were then examined under long-wavelength UV light (ca. 340 nm) to visualize the naturally fluorescent NAPDH; spots that fluoresced were classified as normal, and those that did not were classified as deficient. G6PD genotyping for Mahidol variant was performed on all female participants using an established PCR–RFLP protocol [5]. In order to explain differences in haemoglobin reductions observed in heterozygous female participants with the same genotype, quantitative assessment of G6PD activity was performed at steady state 6 to 8 wk later and repeated twice per patient on two additional follow-up samples. Spectrophotometric quantitative assessment of G6PD enzymatic activity [6] was performed using the Trinity kit assay (Trinity Biotech, Ireland) in duplicate 10 μL whole blood samples. The absorbance at 340 nm at 30°C was measured over 10 min by a UV spectrophotometer (UV-1800, Shimadzu, Japan) with an electronically controlled temperature compartment. Estimated G6PD activity was calculated as the arithmetic mean of three repeated assessments and expressed per gram of haemoglobin (IU/gHb), as is commonly reported, and also per red blood cell (IU/RBC), which is considered more accurate in anaemic patients [7]. Population median values and ranges for both assessments have been established previously in this population [8].
The primary outcome of this analysis was the fractional haematocrit reduction up to day 14 after enrolment. Secondary outcomes included a description of the temporal patterns of haemolysis and characterisation of the effects of the initial parasitaemia and G6PD activity on haematocrit reduction. A symptom sheet including common complaints was reviewed daily with each participant during study drug administration. Adverse events occurring after treatment completion were then collected passively and reported until day 42.
Laboratory data were processed by laboratory staff blinded to the clinical status of the participants. Subjective data (i.e., symptoms and adverse events) were collected by clinical study staff not involved with the study analysis. Because the main trial was not blinded, there may have been potential bias in extracting complaints of common adverse effects caused by primaquine; however, the G6PD genotype of all participants was unknown at the time of drug treatment.
In this population, the prevalence of G6PD deficiency in males has been determined previously to be around 14% (varying from 9% to 18% depending on the area and/or ethnic group) [4]. Using this proportion as the allelic frequency and based on the Hardy-Weinberg principle, we expected 24% of females to be G6PD heterozygotes (varying from 15% to 30%) in the main trial. Because of the small sample size amongst heterozygote participants, we pooled the DP and CQ treatment regimens. Therefore, we analysed female participants in four groups:—by G6PD status—heterozygote and wild type, and—by primaquine dose—PMQ-0.5 and PMQ-1.
The hypothesis for this study was proposed before data collection, and the analysis was planned after data collection began. First, descriptive statistics were used to compare baseline participant characteristics between the four groups (as defined above). Then, a complete case analysis was performed; missing data were excluded. Haematocrit changes over time were normalized:
Fractional haematocrit change=(hctday x−hctday 0)/hctday 0
Unadjusted results were compared between groups and expressed as mean changes. A conservative estimate of a significant fractional haematocrit change was considered a priori to be +/− 15%—to take into account a technical variation of 10% and an actual 5% change [9]. However, fractional haematocrit changes of +/− 25% have been more commonly considered as clinically significant [10], so both were assessed. Multivariable linear regression adjusted for initial parasitaemia was used to assess the following:
Pearson’s correlation was used to determine the relationship between the mean of the maximum fractional haematocrit reduction and G6PD activity for the heterozygote groups. Multivariable logistic regression was used to assess differences between groups in the proportions of participants with fractional haematocrit reductions >15% and >25% unadjusted for initial parasitaemia (including a test for interaction between G6PD genotype and primaquine dose) and the presence of absolute haematocrit reductions below 30% (the threshold for anaemia treatment). The likelihood ratio test was used to determine whether the interactions were significant. Univariable logistic regression was used to assess symptoms during primaquine administration and adverse events between groups. Of the participants analysed, none were lost to follow up by day 14. Statistical analyses were performed using Stata version 13 (StataCorp, College Station, Texas, United States).
Between March 2012 and July 2014, 680 patients with acute P. vivax malaria and a normal G6PD FST were enrolled in the main trial. Details of all participants and the therapeutic outcomes will be reported elsewhere. Of the 241 female participants recruited, 34 (14%) were subsequently found to be G6PD Mahidol heterozygotes. One withdrew from the trial, leaving 33 with evaluable data (Fig 1). Symptoms, vital signs, and laboratory indices were similar in all treatment groups. G6PD heterozygous female participants taking PMQ-0.5 were younger (median 12 y [interquartile range (IQR) 7 to 27.5; range 1.5 to 45]) than heterozygous female participants taking PMQ-1 (19 y [IQR 14 to 27; range 6 to 38]) or wild-type female participants taking PMQ-1 (17 y [IQR 10.5 to 35; range 2 to 63]) or PMQ-0.5 (21 years [IQR 11 to 38; range 3 to 60]) (Table 1). Children less than 5 y old were under-represented in the heterozygous female PMQ-1 group.
The effect of primaquine dose on haemolysis was different in G6PD wild-type and G6PD heterozygous female participants (p = 0.0104 for interaction). There was no evidence of primaquine dose-dependent haemolysis in G6PD wild-type female participants when comparing PMQ-1 and PMQ-0.5, and there were no differences when comparing the CQ and DP groups (S1 Table). In contrast, primaquine was associated with clear dose-dependent haemolysis in G6PD heterozygous female participants. In the pooled analysis (CQ+DP) of G6PD heterozygous female participants, unadjusted for parasitaemia, 14/17 (82%) in the PMQ-1 group had fractional haematocrit reductions >15% compared with 6/16 (38%) G6PD heterozygous female participants in the PMQ-0.5 group (odds ratio [OR] 7.8 [95% CI 1.6 to 38.8]; p = 0.012). When comparing the proportions of G6PD heterozygous female participants with a fractional haematocrit reduction of >25%, the results were similar: 9/17 (53%) in the PMQ-1 group had fractional haematocrit reductions >25% compared with 2/16 (13%) G6PD heterozygous female participants in the PMQ-0.5 group (OR 7.9 [95% CI 1.4 to 45.8]; p = 0.022).
Among G6PD heterozygous female participants who took PMQ-1 with DP, the haematocrit reached its nadir on day 5 with a mean fractional reduction of −23.2% (SD 11.0) compared to −16.5% (SD 7.4) on day 6 in those taking PMQ-0.5. In the CQ group, the haematocrit reached its nadir on day 4 with a mean fractional reduction of −16.4% (SD 7.9) with PMQ-1 compared to −11.9% on day 5 (SD 11.1) with PMQ-0.5 (S1 Table). Irrespective of primaquine dose, by day 7 the mean fractional haematocrit rose again in all groups (except for three G6PD wild-type participants) despite continued primaquine administration (Fig 2 and S2 Table). In the pooled analysis, unadjusted for parasitaemia and including a test for interaction between G6PD genotype and primaquine dose, the effect of primaquine dose on haemolysis remained different in G6PD wild-type and G6PD heterozygous female participants (p = 0.0018 for interaction). The mean maximum individual fractional haematocrit reduction (mean of the individual nadirs) was similar between G6PD wild-type female participants taking PMQ-1 and PMQ-0.5, but in G6PD heterozygous female participants, it was significantly greater in the PMQ-1 group (−23.6% [95% CI −27.0% to −20.2%]) compared with the PMQ-0.5 group (−15.5% [95% CI −19.0% to −12.0%]), with a mean difference of −8.1% (95% CI −13.0% to −3.2%; p = 0.001) (Fig 3 and Table 2). When the same analysis was performed adjusting for parasitaemia, the effect of genotype and dose was reduced significantly, and the interaction of primaquine dose on haemolysis remained different in G6PD wild-type and G6PD heterozygous female participants (p = 0.0014 for interaction) (Tables 2 and 3). The estimated change in the mean maximum fractional haematocrit per 10-fold increase in initial parasitaemia in G6PD wild-type female participants (without testing for interaction between G6PD genotype and primaquine dose) was −2.9% (95% CI −5.1% to −0.60%; p = 0.012) in the PMQ-1 group and −2.0% (95% CI −4.1% to 0.06%; p = 0.056) in the PMQ-0.5 group. Corresponding values for heterozygous female participants were 1.5% (95% CI −3.0% to 6.0%; p = 0.503) in the PMQ-1 group and 2.4% (95% CI −2.3% to 7.0%; p = 0.315) in the PMQ-0.5 group.
The median interval to the nadir of haematocrit reduction was 3 d (IQR 2 to 5) in wild-type female participants in the PMQ-1 group and 3.5 d (IQR 2 to 6) in the PMQ-0.5 group. In heterozygous female participants, the corresponding values were 5 d (IQR 4 to 6) in both groups.
G6PD activity was measured in 21 heterozygous females and 21 control wild-type females at steady state, i.e., at least 6 wk after recruitment when the participants had recovered fully. After controlling for initial parasitaemia and including tests for interaction between G6PD activity and genotype (the interaction term between G6PD genotype and primaquine dose was not significant, so it was not included in the model), there was no correlation between the mean individual fractional haematocrit reduction and G6PD activity expressed as IU/gHb within any group. However, when expressed in IU/RBC, there was stronger evidence of a relationship. In the PMQ-1 group, the reduction in haematocrit for heterozygous females was 3.1% less (95% CI 0.66% to 5.44%; p = 0.014) for a 10% higher level of G6PD activity and 3.2% less (95% CI 0.83% to 5.54%; p = 0.009) in the PMQ-0.5 group (Fig 4).
The incidence of anaemia requiring haematinic treatment within the first 42 d from enrolment, defined as an absolute haematocrit <30%, was not different between heterozygote females who received PMQ-1 (11/17 [65%]) or PMQ-0.5 (6/16 [38%]) (OR 2.1 [95% CI 0.47 to 9.3]; p = 0.328). When comparing G6PD heterozygous and wild-type females in the PMQ-1 group, the heterozygotes were significantly more likely to require haematinic treatment (OR 8.9 [95% CI 2.6 to 30.6]; p = 0.001). A similar result was found when comparing G6PD heterozygous females in the PMQ-1 group with the wild-type females in the PMQ-0.5 group (OR 5.3 [95% CI 1.68 to 16.4]; p = 0.004). The only participants requiring blood transfusion were two heterozygous females who received PMQ-1; recovery occurred without further complications (Table 4). Active and passive reporting of symptoms potentially associated with haemolysis (headache, backache, abdominal pain, muscle pain, dizziness, palpitations, and fatigue) during study drug administration was similar in the different groups. Abdominal pain was not different between participants in the PMQ-1 compared to the PMQ-0.5 groups (OR 1.9 [95% CI 0.71 to 5.0], p = 0.201).
These results show that the standard high dose primaquine regimen given over 14 d (0.5 mg base/kg/d) is reasonably well tolerated by G6PD Mahidol heterozygous female participants, whereas significant haematocrit reductions (fractional haematocrit reduction >25%) were observed in those taking the higher daily dose (same total dose) given over 7 d (1 mg base/kg/d). Two heterozygous participants receiving the higher primaquine dose given over 7 d (1 mg base/kg/d) required transfusion. G6PD Mahidol wild-type female participants did not have an increased risk of primaquine-induced haemolysis when taking the same dose. Irrespective of primaquine regimen and despite continued dosing, haematocrits increased by day 7. The lack of subjective complaints and change in vital signs suggests that haematocrit declines were tolerated and these symptoms and signs did not identify moderate or even severe degrees of haemolysis [11].
Relapse in vivax malaria is a major cause of morbidity in tropical countries and in areas of high transmission contributes to mortality in young children from severe anaemia [12,13]. Prevention of relapse (radical cure) requires use of 8-aminoquinoline drugs, with the attendant risk of haemolysis in patients who are G6PD deficient. Since primaquine was introduced 65 y ago, it has been widely recommended and widely not given. This is because screening tests for G6PD deficiency have not been available generally, and the potential risks of giving primaquine unwittingly to a deficient patient were considered to outweigh the overall benefits [14]. Those risks depend on the degree of deficiency, determined by the G6PD genotype, and the exposure to the oxidant metabolites of primaquine, determined largely by the dose and the ability to metabolise the parent compound (reflecting in part cytochrome 2D6 activity and thus the 2D6 genotype) [15,16]. However, increased deployment of radical treatment is necessary if the enormous burden of vivax malaria is to be reduced, and this is being encouraged strongly [17]. To this end, simple point-of-care tests for G6PD deficiency have been developed to guide treatment. These tests have similar performance characteristics to the conventional NADPH FST and so will reliably identify G6PD homozygous females and hemizygous males, but, as this study illustrates clearly, tests with a detection threshold of ≳30%–40% of normal activity do not identify a large proportion of female heterozygotes. There is a significant dose-dependent haemolytic risk in some of these patients [18,19].
Although there are now more than 200 different G6PD deficiency genotypes described, usually resulting in an unstable enzyme, there is substantial phenotypic variability within genotypes, so severe haemolysis can still occur with so called “mild variants.” Amongst the more common G6PD variants globally, the African A-genotype is at the mild end of the spectrum, and the Mediterranean variant is at the severe end. The Mahidol variant studied in this trial is associated with low G6PD activities in hemizygous males, suggesting it is towards the severe end of the spectrum. Oxidant haemolysis in G6PD deficiency results in loss of the most deficient cells. In most deficiencies, these are the older and mature erythrocytes in which the unstable enzyme activity has declined such that oxidant defences can no longer be maintained. As these cells are haemolysed, they are replaced by young red cells from the bone marrow with much higher levels of enzyme activity. These young erythrocytes are therefore relatively resistant to oxidant stresses and are less likely to haemolyse [20]. As a result, in all but the most severe deficiencies, there is an initial fall in haemoglobin followed by a rise despite continued drug administration [21,22]. P. vivax malaria is also associated with haemolytic anaemia. In this study, primaquine had a negligible additional haemolytic effect over that caused by P. vivax malaria in women who were wild type for the G6PD Mahidol variant. However, heterozygous females with P. vivax malaria who had screened as “normal” (which suggests at least ≳30%–40% of their red blood cell population had normal levels of G6PD activity) experienced significant additional haemolysis as a result of taking a higher dose of primaquine over 7 d (1 mg base/kg/d). Although the potential for severe haemolysis in female G6PD heterozygotes has been well recognised [23], this has been little studied. In this study, haemolysis was dose dependent, as reported previously in hemizygous males [11]. Heterozygous females who received a higher primaquine dose of 1 mg base/kg/d haemolysed significantly more and required more treatment of their anaemia than those receiving the standard high 0.5 mg base/kg/d dose. Despite this, the haematocrit started to rise again, usually after the fifth day, as the new young red cells with higher levels of G6PD enzyme activity entered the circulation. This suggests that strategies to attenuate primaquine haemolytic toxicity by dose spacing should reduce risks in G6PD heterozygotes with the Mahidol variant but that starting radical treatment with higher daily individual doses of 8-aminoquinolines is potentially dangerous if blood transfusion is not immediately available.
Unfortunately, this relatively common, potentially high-risk group is not identified by current screening tests. Even the “gold standard” spectrophotometric assay of average erythrocyte G6PD activity (expressed as IU/gHb), which certainly cannot be performed widely in the rural tropics, showed a poor correlation with haemolytic toxicity. When spectrophotometric results were expressed as IU/RBC, a correlation was evident, which suggests that quantitative methods used to measure G6PD activity should be adjusted in anaemic populations. Handheld point-of-care quantitative testing methods and other qualitative tests of G6PD deficiency are being developed with detection thresholds of ~70% activity. The performance of these new methods and the number of truly normal (i.e., wild-type genotypes) individuals these tests may misidentify as G6PD deficient remain to be assessed.
There are a number of limitations in this analysis. Firstly, the primary trial was not designed specifically to address primaquine-induced haemolysis in G6PD heterozygous females. The number of heterozygous female participants was small, and pooling the CQ and DP groups by primaquine regimens was necessary. As with chloroquine, dihydroartemisinin-piperaquine has been shown to increase primaquine plasma concentrations [24], but it is not known whether bioactive metabolites are higher or lower than chloroquine. Because of the small number of participants analysed, conclusions could not be made on the relationship of haemoglobin type with G6PD deficiency, haemolysis, and anaemia after treatment with primaquine. There was a paucity of data in heterozygous females with enzymatic activity in the range of 40% to 60% of normal who took PMQ-0.5, and the fractional haematocrit reductions during primaquine administration in this group require further confirmation.
For the present, the practical therapeutic implications of these findings are that, for areas where G6PD deficiency genotypes of similar or greater severity to G6PD Mahidol are prevalent, daily primaquine doses greater than 0.5 mg base/kg/d should not be given to female patients irrespective of the G6PD screening result unless they can be closely monitored. More information is needed on haemolytic patterns and haemolytic risk in patients with G6PD Mahidol and other G6PD deficiency genotypes and with point-of-care quantitative testing when it becomes more widely available. Safer methods of giving primaquine are needed.
|
10.1371/journal.pgen.1006728 | Single-trait and multi-trait genome-wide association analyses identify novel loci for blood pressure in African-ancestry populations | Hypertension is a leading cause of global disease, mortality, and disability. While individuals of African descent suffer a disproportionate burden of hypertension and its complications, they have been underrepresented in genetic studies. To identify novel susceptibility loci for blood pressure and hypertension in people of African ancestry, we performed both single and multiple-trait genome-wide association analyses. We analyzed 21 genome-wide association studies comprised of 31,968 individuals of African ancestry, and validated our results with additional 54,395 individuals from multi-ethnic studies. These analyses identified nine loci with eleven independent variants which reached genome-wide significance (P < 1.25×10−8) for either systolic and diastolic blood pressure, hypertension, or for combined traits. Single-trait analyses identified two loci (TARID/TCF21 and LLPH/TMBIM4) and multiple-trait analyses identified one novel locus (FRMD3) for blood pressure. At these three loci, as well as at GRP20/CDH17, associated variants had alleles common only in African-ancestry populations. Functional annotation showed enrichment for genes expressed in immune and kidney cells, as well as in heart and vascular cells/tissues. Experiments driven by these findings and using angiotensin-II induced hypertension in mice showed altered kidney mRNA expression of six genes, suggesting their potential role in hypertension. Our study provides new evidence for genes related to hypertension susceptibility, and the need to study African-ancestry populations in order to identify biologic factors contributing to hypertension.
| Hypertension is a global health problem which affects disproportionally people of African descent. We conducted a genome-wide association study of blood pressure in 31,968 Africans and African Americans to identify genes conferring susceptibility to increased blood pressure. This research identified three novel genomic regions associated with blood pressure which have not been previously reported in studies of other race/ethnicity. Using experimental models, we also showed an altered expression of these genes in kidney tissue in hypertension. These findings provide new evidence for genes influencing hypertension risk and supports the need to study diverse ancestry populations in order to identify biologic factors contributing to hypertension.
| Genetic studies hold the promise of providing tools to better understand and treat clinical conditions. To achieve the clinical and public health goals of reducing hypertension and its sequelae, and to understand ethnic disparities in the risk for hypertension, there is a need to study susceptible populations for genetic determinants of blood pressure (BP). BP traits are highly heritable across world populations (30 to 55%).[1–4] Over 200 genetic loci have been identified in genome-wide association studies [5–13] and admixture mapping studies.[14–17] These variants explain approximately 3.5% of inter-individual variation in BP.[5, 7] However, there is still a paucity of studies focused on individuals of African descent. Most of the loci identified in the literature have not been replicated in individuals of African ancestry.[18, 19]
African Americans have higher mean BP, an earlier onset of hypertension, and a greater likelihood to have treatment-resistant hypertension than other ethnic groups.[20–23] Emerging research on Africans shows increasing prevalence of hypertension in urban African communities [24, 25] which are more Westernized than rural African communities and, so, more closely resemble communities in which African Americans live in the U.S. Hypertension contributes to a greater risk of coronary heart disease, stroke, and chronic kidney disease.[26–30] African Americans experience increased risk of these hypertension-related outcomes [31–34] but the underlying mechanisms, whether environmental exposures or increased genetic susceptibility, are unknown.
We hypothesized that additional variants associated with BP can be identified in people of African ancestry; some variants may be African-specific, as has been observed for multiple traits, including kidney disease [35] and metabolic syndrome.[36, 37] Other variants may be identified in novel loci based on a higher frequency of risk alleles in this population. We used high density imputed genotypes from the 1000 Genomes Project (1000G) to expand the genome coverage of genetic variants so that we could examine the evidence for association with BP traits.
Here, we report three novel loci associated with BP which are driven by variants that are common in or unique to African-ancestry populations. Through bioinformatics and experimental evidence of kidney gene expression in mice submitted to angiotensin-II (Ang II) induced hypertension, we provide evidence for a key role of these genes in the pathogenesis of hypertension. In addition, our study extends the discovery of BP loci to genes related to kidney and the immune systems, and provides biological relevance for these loci to BP regulation.
The study design and analysis process are shown in Fig 1. Study characteristics, genotyping, and quality control (QC) for discovery and replication samples are shown in S1 and S2 Tables. The discovery samples included 31,968 individuals of African ancestry from 19 studies. The replication samples included 4,184 individuals of African ancestry from three studies, 23,914 individuals of European ancestry from five studies, 14,016 individuals of Korean ancestry from three studies, and 12,278 individuals of Hispanic/Latino ancestry from one study.
Study-specific genomic-control inflation ranged from 0.98–1.06 (S3 Table, S1 Fig) and the linkage disequilibrium (LD) score regression intercepts of the single-trait BP meta-analyses calculated by the LD score regression approach ranged from 1.02–1.04. [38] These results suggest well-controlled population stratification.
The single-trait BP meta-analyses identified several genome-wide significant single nucleotide polymorphisms (SNP) at eight loci (P < 5.0×10−8, systolic BP (SBP): three loci, four SNPs; diastolic BP (DBP): three loci, three SNPs; pulse pressure (PP): three loci, four SNPs; and hypertension (HTN): one locus, one SNP), with the EVX1/HOXA locus identified for SBP, DBP and HTN (S2A–S2D Fig). When combining summary statistics for SBP, DBP, and HTN using the multi-trait approach CPASSOC,[39] we identified one locus by the multi-trait statistic SHom (EVX1/HOXA) and six loci by SHet (ULK4, TCF21, EVX1/HOXA, IGFBP3, CDH17, ZNF746) at P < 5×10−8 (S2E and S2F Fig). Note some loci overlap between single-trait and multi-trait findings.
We observed 264 variants with P < 1×10−6 for either single- or multi- trait GWAS and these variants were further analyzed by conditional association on the most associated SNPs at each locus (S4 Table). These analyses resulted in 72 independent associations, which included 58 SNPs with minor allele frequency (MAF) ≥ 0.05 and 14 with low frequency variants (0.01< MAF < 0.05) (S5 Table).
Among these 72 variants carried forward for trans-ethnic replication, nine variants, all low frequency variants (MAF<0.02), were not available in replication cohorts because they were either monomorphic in the replication population or had a low imputation quality, reducing our replication effort to 63 variants (S6 Table). Eleven independent variants at nine loci were significantly associated with BP traits at P < 1.25×10−8 in the combined discovery and replication analyses and are reported in Table 1. This significance level was determined by adjusting for two independent traits for SBP, DBP, PP and HTN, and two tests of multiple trait analysis. This includes six variants that reached significance level at discovery stage (P <5 x10-8). Two loci were identified only through multi-trait analyses (FRMD3, IGFBP3). Three of these nine loci are novel: TARID /TCF21, FRMD3, and LLPH/TMBIM4 (Fig 2A–2C). Four loci (ULK4, PLEKHG1, EVX1/HOXA cluster, and GPR20) have been reported in our previous BP GWAS of African ancestry (S3 Fig),[7, 18] and two loci (IGFBP3, CDH17) have been reported in multiple-trait analyses of African-ancestry studies (Fig 2D–2F).[39] A composite genetic-risk score using the eleven variants identified accounted for 1.89%, 2.92%, 1.03% and 1.08% of the variance for SBP, DBP, PP and HTN respectively.
Five of the eleven replicated variants are common in individuals of African ancestry but rare or monomorphic in individuals of non-African ancestry (rs76987554, rs115795127, rs113866309, rs7006531, and rs78192203)(Table 1). These five variants were 1) either low frequency or common variants in COGENT-BP African-ancestry samples; 2) low frequency in 1000G Phase I Integrated Release Ad Mixed-American ancestry (AMR); and 3) monomorphic in 1000G Asian ancestry (ASN) or European ancestry (EUR). One common variant was present in only 1000G samples of African ancestry (rs115795127 at FRMD3, Table 1). These variants were located at the three novel loci (TARID/TCF21, FRMD3, and LLPH/TMBIM4). Given the differences in allele frequency across continental-ancestry populations, we examined the evidence for selection at each of these loci using iHS, which measures the amount of extended haplotype homozygosity at a given SNP along the ancestral allele relative to the derived allele.[40] The iHS score for rs115795127 was 2.7 in African American samples from the Candidate-gene Association Resource (CARe) consortium (see Methods), suggesting selection at the FRMD3 locus (S7 Table).
We observed two independent genome-wide significant variants at the EVX1/HOXA locus (P < 1.25×10−8). The two variants, rs11563582 and rs6969780, are in weak LD (r2 = 0.21) (S3A–S3C Fig), and the LD pattern suggests that these SNPs are located in two blocks (S4 Fig). SNP rs11563582 is in strong LD with the previously reported SNP in the region (rs17428741).[18] SNP rs6969780 remained significant when conditioning on rs11563582 (S4 Table), thus demonstrating the presence of allelic heterogeneity at this locus. Two independent variants at ULK4 reached the significance threshold: rs7651190 and rs7372217 (LD r2 = 0.15) (S4E Fig). SNP rs7372217 is in strong LD with the previous reported SNP rs1717027.[18] The association evidence of rs1717027 can be explained by rs7372217 but not by rs7651190 in conditional analysis (S4 Table). Thus, rs7651190 is an independent association at this locus. At the GPR20 locus, our most significant SNP, rs78192203, is 8kb away and it is not in LD with the published SNP, rs34591516 (r2 = 0.008, D’ = 0.68 in African American CARe participants).
To gain insight into biologic mechanisms underlying genes associated with BP traits, we performed pathway analysis using publicly available databases. [41] The most relevant pathways identified were GSK3, Th1/Th2 differentiation, and Sonic Hedgehog (SHH) pathways (BIOCARTA): pyrimidine metabolism, apoptosis signaling pathway, and B cell activation (Panther); JAK Stat signaling, T cell receptor signaling, and B cell receptor signaling (Ingenuity); cytokine-cytokine receptor interaction and vascular smooth muscle contraction (KEGG); and neuronal activity, T cell mediated immunity, and tumor suppressor (Panther Biological Process) (Gene Set Enrichment Analysis [GSEA] P-value < 0.01, S8 Table). These analyses suggest enrichment of immune pathways for BP traits.
We performed functional annotation and cell type group enrichment analysis using the stratified LD score regression approach which uses data from ENCODE and the Roadmap Epigenetic Project, as well as GWAS results while accounting for the correlation among markers. [42] We estimated functional categories of enrichment using an enrichment score, which is the proportion of SNP-heritability in the category divided by the proportion of SNPs. We identified super enhancer (PEnrich = 5.4×10−5, Enrichment = 5.6 for DBP), enhancer (PEnrich = 4.8 ×10−4, Enrichment = 4.3 for HTN), and H3K27ac (PEnrich = 3.2×10−4, Enrichment = 3.6 for HTN) significant enrichment (Fig 3). These results support a role of identified noncoding regulatory regions in BP regulation. In addition, the following cell types showed significant enrichment (P ≤ 2.5 × 10−3): the immune (PEnrich = 1.4×10−9, Enrichment = 8.4 for DBP), kidney (PEnrich = 5.4×10−5, Enrichment = 4.8 for DBP), and cardiovascular (PEnrich = 8.9×10−5, Enrichment = 4.2 for SBP) systems (Fig 3).
We next determined the enrichment of variants at the eleven genome-wide significant loci for DNase l hypersensitive (DHS) sites in 34 tissue categories from ENCODE. At each locus, we identified variants in r2>0.1 with the index variant and calculated causal evidence (Bayes Factors) for each variant. We then tested for enrichment in the causal evidence of variants in DHS sites using fGWAS.[43] We found enrichment of blood/immune DHS (Enrichment = 3.1) and cardiovascular DHS (blood vessel Enrichment = 28.7, heart Enrichment = 2.0), in addition to DHS in several fetal tissues (S5 Fig). Candidate causal variants at several loci overlapped enriched DHS sites. For example, at the LLPH/TMBIM4 locus, the most likely causal variant, rs12426813, overlaps a DHS site active in immune (CD14+, CD4+, CD34+), blood vessel (HMVEC), and heart (HCF) cells (S5 Fig).
To examine whether the eleven significant SNPs are eQTL, we searched the genotype-tissue expression (GTEx) pilot database, which includes non-disease human tissue.[42] Among the eleven SNPs, three SNPs have been identified as eQTL: rs6969780 (HOXA2), rs7651190 (ULK4), and rs62434120 (PLEKHG1) (S9 Table). SNP rs6969780 is an eQTL for expression of HOXA2, HOXA7, HOTAIRM1, and HOXA5 in multiple tissues, including esophagus, artery, lung, skin, nerve, adipose, skeletal muscle, and stomach tissues. SNP rs7651190 is an eQTL for ULK4 and RPL36P20 in artery, whole blood, thyroid, nerve, esophagus, skeletal muscle, skin, brain, and stomach cells/tissues. SNP rs62434120 is an eQTL for PLEKHG1 in testis tissue.
To determine if identified genes are functionally involved in BP regulation in the kidney during hypertension,[44] we quantified gene expression in mice kidneys at baseline and during the hypertensive state induced by Ang II. This hypertensive model was chosen for two reasons: 1) to mimic the low plasma renin state, albeit more exaggerated than the level observed, in African-ancestry individuals that has been suggested to reflect the elevated renin-angiotensin system activity at the tissue level in the kidney [45], and 2) maintenance of hypertension in the Ang II model requires activation of the immune system that is implicated in several identified loci.[46, 47] Kidney gene expressions of the identified genes were compared to age-matched untreated mice after two weeks of Ang II infusion, which increases SBP. For the HOXA locus, we examined the expression of genes that are known to be expressed in the mouse kidney: Hoxa1 (2 isoforms), 5, 7, 9, 10 (2 isoforms), and 11. Among all the genes examined, Tmbim4 was the most abundantly expressed gene in the kidney at baseline. Six genes—Hoxa5, Hoxa10-1 isoform, Hoxa11, Tmbim4, Igfbp3, and Plekhg1—were significantly differentially expressed in the kidney after Ang II treatment compared to baseline (Fig 4). Except for Hoxa5, which showed a significant decrease (Fig 4A), the expression of all these genes increased after the intervention. The expression of six genes—Hoxa1-1 isoform, Hoxa7, Hoxa9, Hoxa10-2 isoform, Llph, and Ulk4—were unchanged after Ang II infusion (Fig 4B). The following genes were not expressed in the adult mouse kidney at baseline or after Ang II intervention: Frmd3-1 isoform, Frmd3-2 isoform, Grp20, Tcf21, Cdh17, and Hoxa1-2 isoform.
To date, this is the largest genome-wide analysis of African-ancestry populations to study genetic variants underlying BP traits using dense-coverage imputed genotypes. Our main findings are eleven independent variants at nine loci, significantly associated with BP traits, including three newly identified loci (TARID/TCF21, FRMD3, LLPH/TMBIM4). We also found evidence for additional independent SNP associations in fine-mapping of three previously described loci, ULK4, EVX1/HOXA, and GRP20.[18, 39]
The most significant variants at TARID/TCF21, FRMD3, GPR20, and CDH17 are common variants in COGENT-BP African-ancestry participants, but monomorphic or low frequency in non-African-ancestry populations. For example, rs115795127 at FRMD3 is rare in European populations (MAF = 0.0007) and absent in East Asian and Hispanic/Latino populations. Therefore, they could not be identified in GWAS of non-African-ancestry populations even when increasing sample sizes. We also show evidence for selection for the variant at FRMD3, although additional studies should confirm these findings. The African-specific variants were not well tagged by HAPMAP2 data and therefore were not detected in our previous African-ancestry GWAS.[18] Overall, our results suggest additional gain in discovery when using dense imputed genotypes and support a role of population-specific alleles in African and African-admixed populations contributing to BP regulation and hypertension. Furthermore, they support the rationale and the need to study diverse populations in order to more effectively characterize the genetic architecture of BP in populations and the ethnic disparities in hypertension.
Functional annotation of our lead variants showed co-localization with annotated elements, including super enhancer, enhancer, and H3K27ac chromatic mapping in immune cells and kidney tissues, which has not been previously reported, in addition to cardiovascular tissues. There was also evidence for regulatory function in these relevant tissues through gene expression regulation (eQTL) and through overlaps with DHS in relevant tissues/cells. This evidence was additionally supported by experimental findings of differential expression of six genes (Hoxa5, Hoxa10-1 isoform, Hoxa11, Tmbim4, Igfbp3, and Plekhg1) in the mouse kidney after HTN induced by Ang II treatment. Overall, our results suggest the functional importance of identified genes in regulating BP in both normal and hypertension states.
At the newly identified loci, SNP rs76987554 is an intronic variant in TARID (TCF21 antisense RNA inducing promoter demethylation) which has not been previously reported to be associated with BP traits. A nearby gene, TCF21 (transcription factor 21), is a transcription factor of the basic helix-loop-helix family, which is mainly expressed in the liver, kidney, and heart. TCF21 is involved in epithelial differentiation and branching morphogenesis in kidney development,[48] and was associated with hypertension in a study of individuals of Japanese ancestry.[49] At the chromosome 7, rs115795127 is an intronic variant to FRMD3 (FERM domain containing 3) which encodes a protein involved in maintaining cell shape and integrity. FRMD3 has been associated with type 1 and type 2 diabetic kidney diseases in different ethnic populations, including those of European, African, and Asian ancestries.[50] The diabetes variant, rs10868025, is not in LD with rs115795127 in our African American samples or in 1000G EUR samples (r2 = 0.00028 and 0.0018, respectively), thus representing an independent association at this locus.
At chromosome 9, the functions of LLPH and TMBIM4 genes in BP regulation are currently unknown. LLPH belongs to the learning-associated protein family and is highly expressed in the immune system and the adrenal gland. TMBIM4 encodes the transmembrane BAX inhibitor motif-containing protein 4 and is highly expressed in whole blood, the immune system, and the adrenal gland.[51] The most significant variant at this locus, rs113866309, overlaps a DHS in immune, blood vessel, and heart cells. In our experimental model in mice, Tmbim4 gene expression was significantly increased after Ang II-induced HTN. This gene has been shown to inhibit apoptosis[52] and to decrease the efficacy of inositol 1,4,5-triphosphate (IP3)-dependent release of intracellular Ca2+. [53] This raises the possibility that the TMBIM4 protein may serve to dampen the effect of Ang II, which activates IP3 in vascular smooth muscle cells through the stimulation of the angiotensin type 1 receptor.[51, 53, 54] Therefore, it is possible that in conditions of activated renin-angiotensin system, genetic variants that lower the expression of TMBIM4 may augment BP, whereas genetic variants that increase its expression may attenuate BP.
Other genes, such as Hoxa5, Hoxa10-1, Hoxa11, Igfbp3, and Plekhg1, were significantly differentially expressed after Ang II-induced HTN in our mice experimental models. The HOXA-cluster has been identified in our previous GWAS of BP in African ancestry and in a recent GWAS of BP in European ancestry[5] though the underlying mechanisms related to BP control are unknown. We identified two independent variants at this locus; further studies are needed to delineate which of the HOXA genes are most likely involved in the association. In our experimental mice model, the Hoxa10-1 isoform had a greater than 20-fold increase in kidney expression during Ang II-induced HTN compared to baseline levels. However, it remains to be determined whether it is an effect of Ang II in hypertension, or a compensatory response to hypertension. Future studies using genetic manipulation in rodents are required to determine whether these changes are specific response related to BP and Ang II or simply a generic response to stress.
We identified several additional pathways involved in BP traits, including the GSK3 pathway, which has been reported to influence Wnt-mediated central BP regulation.[55] The Th1/Th2 pathway is involved in the regulation of immune responses[56] and has been linked to hypertension and atherosclerosis.[57, 58] The role of the immune system in the development of hypertension has been suggested in clinical studies and experimental animal models.[59–64] This includes reports of overlap of genetic variant associations between BP traits and immune-disorders [65] and evidence of enrichment of immune pathways from GWAS of BP.[66] Mutations of SH2B3, a gene identified in a GWAS of hypertension, have been recently shown to attenuate Dahl salt-sensitivity hypertension through inflammatory modulation.[67] In addition, the actions of Ang II in the pathophysiology and maintenance of hypertension are in part mediated through the activation of the immune system.[46]
Our assessment of the clinical implications of identified variants is limited by available data on African-ancestry populations. For example, there are currently no large publicly available GWAS of coronary heart disease or stroke outcomes in African-ancestry populations. It should also be noted that most of our replication cohorts were from populations other than those of African ancestry. Therefore, the power of replication analysis could still be low, which explains why only 11 of 63 variants were successfully replicated.
In summary, we report 11 independent variants at nine loci that are potential regulators of BP in our African-ancestry population study. Three loci are new. Identified BP variants are enriched in immune, kidney, heart, and vascular system pathways. Our experimental findings suggest that several of these genes may be involved in the renin-angiotensin pathways in the kidney during hypertension. Further population studies and experimental models are required for a comprehensive assessment of the identified genes across the immune, kidney, and cardiovascular systems. Our study demonstrates the need to further study individuals of African ancestry in order to identify loci and new biological pathways for BP.
Each study followed protocols for phenotype harmonization. For individuals taking anti-hypertensive medications, we added 15 and 10 mm Hg to measured SBP and DBP, respectively, a standard method used in other BP GWAS.[6, 68] PP was calculated as the difference between SBP and DBP after addition of the constant values. HTN was defined by a SBP ≥ 140 mm Hg, a DBP ≥ 90 mm Hg, or use of antihypertensive drugs.[69]
Each cohort was genotyped on either Affymetrix or Illumina genotyping platforms. Pre-imputation quality criteria were applied as described in S2 Table, and included exclusion of individuals with discordant self-reported gender and genetic gender. Imputation was performed using the software MACH-ADMIX, MACH-minimac or IMPUTE2 [70–72] using the Phase 1 integrated (March 2012 release) multi-ethnic reference panel from the 1000G Consortium (http://www.internationalgenome.org/).[73]
Autosomal chromosome SNP associations for SBP, DBP, and PP were assessed by linear regression for unrelated data or by the generalized linear mixed-effects model for family data, under the assumption of an additive genetic model. All models were adjusted for age, age2, sex, and body mass index. Up to ten principal components were included, as needed as covariates in the regression models, to control population stratification.[74, 75] We used standardized pre-meta-analysis QC criteria for all 21 discovery studies.[76] At the SNP level, we excluded variants with 1) imputation quality r2 < 0.3 in MACH or <0.4 in IMPUTE2; 2) the number of informative individuals (2×MAF×N×r2) ≤ 30; 3) an effect allele frequency (EAF) difference larger than 0.3 in comparison with the mixture of 80% YRI and 20% CEU of 1000G; and 4) the absolute regression coefficient ≥ 10. SNPs that passed the QC were carried forward for inverse variance weighted meta-analyses, implemented in METAL.[77]
We applied the CPASSOC software to combine association evidence of SBP, DBP, and HTN. CPASSOC provides two statistics, SHom and SHet, as previously described.[39] SHom is similar to the fixed effect meta-analysis method[77] but accounts for the correlation of summary statistics of the multi-traits and for overlapping or related samples among the cohorts. SHom uses the trait sample size as the weight, so that it is possible to combine traits with different measurement scales. SHet is an extension of SHom, and it can increase the statistical power over SHom when a variant affects only a subset of traits. The distribution of SHet under the null hypothesis was obtained through an estimated beta distribution. To calculate the statistics, SHom and SHet, and to account for the correlation among the traits, a correlation matrix is required. In this study, we used the correlation matrix calculated from the residuals of the three BP traits after adjustments for covariates and principal components.
All independent SNPs identified with P < 10−6 (threshold chosen for suggestive association) in the discovery stage were carried forward for replication in African-ancestry individuals and in multi-ethnic samples of European Americans, East Asians, or Hispanics/Latinos (Fig 1). For single-trait analyses, we conducted fixed effect meta-analyses in the replication sets for each of four BP traits (SBP, DBP, PP and HTN), followed by a combined trans-ethnic meta-analysis of each trait. This was followed by a mega-meta-analyses, combining the results of discovery and replication for single traits using fixed-effects meta-analysis. We also performed a multi-trait CPASSOC analysis of SBP, DBP, and HTN in each replication study. Because CPASSOC only generated test statistics SHom/SHet and corresponding P values without effect sizes, we combined the association P values from all four replication populations using Fisher’s method (http://hal.case.edu/zhu-web/). Finally, we combined the CPASSOC meta-analysis results from the discovery and replication stages using Fisher’s method.
For a single trait GWAS discovery analysis, we used genome-wide significant level P = 5.0×10−8. We performed six different analyses, four single trait (SBP, DBP, PP and HTN) analyses and two CPASSOC (SHom and SHet) analyses for each SNP. For the four single correlated traits (SBP, DBP, PP and HTN), we calculated the number of independent traits using the eigenvalues of the correlation matrix, [78] which resulted two independent traits. Therefore, we counted four independent analyses, which were two independent single traits and two statistics of CPASSOC analyses, and applied an experimental significance level P = 1.25×10−8 for claiming a genome-wide significance when combining discovery and replication samples. We should point out that the two CPASSOC test statistics and a single trait statistic are not independent. Thus, the significance level P = 1.25×10−8 is conservative.
Since a locus may consist of multiple independent signals, we applied approximate conditional analysis implemented in GCTA-COJO[79, 80] using the summary statistics of SNPs with P < 1.0×10−6 from both of the individual trait meta-analyses (http://cnsgenomics.com/software/gcta/cojo.html). The LD among variants was estimated from the five African American cohorts from the CARe consortium.[79]
Pathway analysis was performed using the Meta-Analysis Gene-set Enrichment of variant Associations (MAGENTA) program (http://www.broadinstitute.org/mpg/magenta/).[41] Using the summary statistics from the four BP traits and two statistics from CPASSOC, from the discovery stage, we tested whether sets of functionally-related genes are enriched for associations. This method first converts the P values of SNPs into gene scores with correcting for confounders, such as gene site, number of variants in a gene, and their LD patterns, and then calculated a gene set enrichment P value for each biological pathway or gene set of interest using a non-parametric statistical test. The nominal GSEA P value refers to the nominal gene set enrichment P value for a gene set. The database of pathway/gene-sets to be tested include Ingenuity (June 2008), KEGG (2010), GO, and the Panther, signaling pathways downloaded from MSigDB and PANTHER (http://www.broad.mit.edu/gsea/msigdb/collections.jsp; http://www.pantherdb.org/).[81] We applied the parameters suggested by the authors, which includes the 75th percentile cut off of gene scores, the nominal GSEA P-value < 0.01 and the false discovery rate (FDR) < 0.3.
The enrichment of heritability of genomic regions to different functional categories, including cell type-specific elements, was evaluated using the method of LD score regression (https://github.com/bulik/ldsc).[42, 82] This method partitioned the heritability from the discovery GWAS summary statistics of four BP traits (SBP, DBP, PP, and HTN) while accounting for LD among markers.[42] We calculated enrichment, in functional regions and in expanded regions (+500bp) around each functional class, based on functional annotation, using a “full baseline model” previously created from 24 publicly available main annotations that are not specific to any cell type.[42] Enrichment was calculated based on the ratio of explained heritability and the proportion of SNPs in each annotation category. The standard error of enrichment was estimated with a block jackknife to calculate z scores and P values.[42] The multiple testing threshold was determined using the Bonferroni correction while accounting for two independent-trait analyses based on Ji and Li’s method[78] (P of 0.05/[25 classes × 2 traits]). We also performed cell-type-specific group enrichment analysis using cell-type-specific annotations from four histone marks (H3K4me1, H3K4me3, H3K9ac, and H3K27ac), which corresponded to 220 cell types. We divided the 220 cell-type-specific annotations into 10 groups: adrenal/pancreas, central nervous system (CNS), cardiovascular, connective/bone, gastrointestinal, immune/hematopoietic, kidney, liver, skeletal muscle and other. The analysis characterized cell-type-specific annotations within each group and calculated the enrichment of heritability for each group.[42]
We selected sets of variants in LD r2 > 0.1 from the eleven replicated variants, and calculated Bayes Factors and posterior causal probabilities for each variant from the effect sizes and standard errors, as previously described.[83] Each distinct variant associated with multiple traits was included in the analysis only once. The genomic annotations of DHS sites for 348 cell types from the ENCODE project were obtained and grouped into cell types associated with 34 tissues (http://genome.ucsc.edu/ENCODE/cellTypes.html). Four gene-based annotations—coding exon, 5-UTR, 3-UTR, and 1kb upstream of transcription start site (TSS)—from GENCODE transcripts were also obtained. Variants overlapping each of these annotations were then identified. Using the variant annotations and fGWAS (https://github.com/joepickrell/fgwas), we tested for enrichment of variants across all signals in 38 DHS categories, including in the four gene-based annotations in each model.[43]
We used the GTEx pilot database [82] (http://www.gtexportal.org/home/) to identify eQTLs in the successfully replicated SNPs.
To evaluate population differentiation and natural selection, using Haplotter,[40] we calculated the integrated haplotype score (iHS) in five cohorts of CARe so that we could measure the amount of extended haplotype homozygosity (http://coruscant.itmat.upenn.edu/whamm/ihs.html). Hence, we tested the evidence of recent positive selection at five significant SNPs with differences in allele frequency across continental-ancestry populations. The measures were standardized (mean 0, variance 1) empirically to the distribution of observed iHS scores over a range of SNPs with similar derived allele frequencies. This method assesses the evidence for selection by comparing the extended homozygosity for haplotypes on a high frequency derived allele relative to the ancestry background.[40]
Experiments were carried out in accordance with local and the National Institutes of Health guidelines. The animal protocol was approved by the University of Virginia Institutional Animal Care and Use Committee. Wild-type male mice on the 129S6 background at ~ 3 months of age were used for gene expression analyses. All mice were maintained on a 12-hour light-dark cycle with free access to standard chow and water in the animal facility of the University of Virginia.
The hypertension experimental model was induced using Ang II (Sigma-Aldrich, St. Luis, MO) delivered at 600 ng/kg/min for 2 weeks via Alzet mini-osmotic pumps (Durect Corporation, Cupertino, CA, model 2004), as previously described.[84] For gene expression analyses, RNA from kidney tissue was isolated by RNeasy Mini kit (Qiagen) and transcribed to cDNA by iScript TM cDNA synthesis kit (Bio-Rad). Real time PCR analyses were performed on iQTM5 Multicolor real time PCR Bio-Rad instruments using iQTM SYBER® Green Supermix. Hprt was used as a reference gene for normalization. Sequences of forward and reversed primers (FP and RP) for the gene expression studies are shown in S10 Table.
|
10.1371/journal.ppat.0040014 | Campylobacter jejuni Survives within Epithelial Cells by Avoiding Delivery to Lysosomes | Campylobacter jejuni is one of the major causes of infectious diarrhea world-wide, although relatively little is know about its mechanisms of pathogenicity. This bacterium can gain entry into intestinal epithelial cells, which is thought to be important for its ability to persistently infect and cause disease. We found that C. jejuni is able to survive within intestinal epithelial cells. However, recovery of intracellular bacteria required pre-culturing under oxygen-limiting conditions, suggesting that C. jejuni undergoes significant physiological changes within the intracellular environment. We also found that in epithelial cells the C. jejuni–containing vacuole deviates from the canonical endocytic pathway immediately after a unique caveolae-dependent entry pathway, thus avoiding delivery into lysosomes. In contrast, in macrophages, C. jejuni is delivered to lysosomes and consequently is rapidly killed. Taken together, these studies indicate that C. jejuni has evolved specific adaptations to survive within host cells.
| Campylobacter jejuni is one of the most common causes of food-borne illness in the United States and a major cause of diarrheal disease throughout the world. After infection through the oral route, this bacterium invades the cells of the intestinal epithelium, a property that is important for its ability to cause disease. Usually, bacteria and other material entering the cell move to compartments called lysosomes, where an acidic mix of enzymes breaks it down. This study shows that C. jejuni can survive within intestinal epithelial cells by avoiding delivery to lysosomes. In contrast, in macrophages, which are specialized cells with the capacity to engulf and kill bacteria, C. jejuni cannot avoid delivery into lysosomes and consequently is rapidly killed. These studies help explain an important virulence attribute of C. jejuni.
| Campylobacter jejuni is the leading cause of bacterial food-borne illness in the United States and a major cause of diarrheal disease throughout the world [1]. C. jejuni infection is also an important pre-condition for Guillain-Barré paralysis [2]. Despite its public health importance, relatively little is known about its pathogenesis. Examination of intestinal biopsies of humans [3], in vivo studies in infected primates [4] and other animal models [5–7], together with in vitro experiments using cultured human intestinal epithelial cells [8–10], have demonstrated that C. jejuni can invade non-phagocytic intestinal epithelial cells. However, to date, little is known about the molecular details of the mechanisms by which C. jejuni enters intestinal epithelial cells. Bacterial factors such as motility, glycosylation, and capsular synthesis have been implicated in C. jejuni internalization [11–14]. Strains with mutations in these pathways have defects in their ability to adhere to and invade host cells, as well as to colonize animals [12–19]. Bacterial invasion has also been correlated with C. jejuni's ability to stimulate the activation of MAP kinases leading to the production of the pro-inflammatory cytokine, IL-8 [20,21]. Taken together, these data suggest that bacterial internalization into intestinal epithelial cells is important in C. jejuni pathogenesis.
Although most host factors that are required for C. jejuni internalization into non-phagocytic cells remain unknown, this entry process appears to have unique cytoskeletal requirements. Most other bacterial pathogens such as Listeria monocytogenes, Shigella flexneri, and Salmonella typhimurium utilize the host-cell actin cytoskeleton to gain intracellular access [22]. However, C. jejuni is internalized into intestinal epithelial cells in a microtubule-dependent, actin-independent fashion [10], suggesting that this bacterium employs an entry mechanism unlike those reported for other bacterial pathogens.
The intracellular fate of C. jejuni remains unknown, although it is likely that this bacterium, similar to other intracellular pathogens, may have evolved specific adaptations to survive within host cells. Intracellular pathogens utilize a variety of strategies to survive and replicate within host cells. For example, some pathogens such as Trypanasoma cruzi [23], Listeria monocytogenes [24], and Shigella flexneri [22,25] break out of the phagocytic vacuoles after internalization and can replicate within the cytosol of the infected cell. Other pathogens, such as Leishmania, have evolved an array of adaptations to survive in the hostile environment of the phagolysosome, which is characterized by low oxygen tension, poor nutrient content, low pH, and the presence of a variety of antibacterial products such as antibacterial peptides and lysosomal enzymes [26]. Yet another group of intracellular pathogens survive within a vesicular compartment that does not fuse with lysosomes. For example, Salmonella typhimurium [27] and Mycobacterium tuberculosis [28] alter the biogenesis and dynamics of their vacuolar compartment preventing fusion to lysosomes.
All evidence to date indicates that after internalization into intestinal epithelial cells, C. jejuni resides within a membrane bound compartment [29–31]. We report here that C. jejuni survives within intestinal epithelial cells by deviating from the canonical endocytic pathway thus residing in a unique intracellular compartment that does not fuse with lysosomes.
Although C. jejuni internalization into host cells is believed to play a role in pathogenesis, little is known about its intracellular fate. We therefore examined the ability of C. jejuni to survive within intestinal epithelial cells. Human intestinal epithelial T84 cells were infected with C. jejuni, and total viable intracellular bacteria were determined at different times by counting colony forming units (CFU). Significant numbers of CFU (> 3 × 105/well) were recovered at early time points, however, over time, the number of CFU recovered decreased considerably (Figure 1A). By 24 h there was a ∼500-fold decrease in the number of CFU recovered from infected cells compared to 4 h after infection (Figure 1A). These results suggest that intracellular C. jejuni rapidly loses viability during the course of its intracellular stage. This was surprising as it suggested that the ability of C. jejuni to enter non-phagocytic cells might not confer a significant advantage to this bacterium. We therefore examined the possibility that internalized C. jejuni may alter its physiology in such a way that, although viable, it may not be culturable under the conditions used in this assay. Indeed, C. jejuni has been reported to enter a viable but non-culturable state when subjected to a variety of stimuli or environments [32–34]. To address this issue, we stained C. jejuni recovered from cultured intestinal T84 cells with reagents that distinguish viable from non-viable bacteria (see Materials and Methods). Using these reagents, we observed no decrease in viability of intracellular C. jejuni over time (Figure 1B and 1C). In fact, FACS analysis also revealed that the ratio of viable to non-viable bacteria did not change over the course of infection (Figure 1C and 1D). These results indicate that C. jejuni remains viable for at least 24 h after infection and suggest that it acquires a physiological state that does not allow the recovery of CFU under our standard culture conditions. We hypothesized that once internalized by intestinal epithelial cells, C. jejuni might adapt to the low oxygen environment within the cell by changing its mode of respiration. We therefore tested whether the intracellular bacterial population could be cultured if allowed to “recover” under conditions in which oxygen is very limiting. Human intestinal epithelial T84 cells were infected with C. jejuni and the number of CFU was determined after culturing under oxygen-limiting incubation or under 10% CO2 conditions. The number of CFU decreased drastically (∼500 fold) when bacteria were directly grown under an atmosphere of 10% CO2 (Figure 2A) or in GasPak jars (BBL Microbiology Systems, Cockeysville, MD) with BBL CampyPacks (BBL Microbiology Systems, Cockeysville, MD) (data not shown). However, there was no significant decrease in the number of CFU recovered over time when the plates were incubated under oxygen-limiting conditions for 48 h and then switched to an atmosphere of 10% CO2 (Figure 2A). The number of CFU recovered when plates where pre-incubated under low-oxygen conditions closely correlated with the number of viable bacteria quantified via FACS analysis (Figure 2B). These data demonstrate that C. jejuni remains viable within intestinal epithelial cells for at least 24 h, although it undergoes physiological changes such that requires exposure to oxygen-limiting conditions for its efficient recovery.
There are contradictory reports regarding the ability of C. jejuni to survive within macrophages [29,35–38]. Given our findings in intestinal epithelial cells suggesting physiological changes in intracellular C. jejuni, which affects its ability to be cultured, we re-examined its ability to survive within macrophages. Mouse primary bone marrow–derived macrophages (BMDM) were infected with C. jejuni and the number of CFU recovered over time was determined by incubating culture plates directly under either 10% CO2 or subjecting them to a 48 h pre-incubation under oxygen-limiting conditions prior to incubation under 10% CO2. We found a significant decrease in the number of CFU recovered over time regardless of the culture condition. By 12 h there was a severe reduction in the number of CFU recovered under both 10% CO2 or with pre-incubation under oxygen-limiting conditions (>2,000 and >100 fold, respectively, compared to the CFU recovered 1 h after infection) (Figure 2C), and no CFU were recovered at 24 h of infection regardless of the culture conditions. Taken together, these data demonstrate that intracellular C. jejuni can survive within intestinal epithelial cells but are killed by professional phagocytes.
Immediately after internalization into host cells, C. jejuni resides within a membrane bound compartment [29–31]. It is therefore possible that in order to survive within cells, this bacterium has evolved specific adaptations to survive within lysosomes or to modulate host cellular trafficking events to avoid fusion with lysosomes. If C. jejuni survives within lysosomes, the C. jejuni–containing vacuole (CCV) should be accessible to endocytic tracers. To test this hypothesis, Cos-1 cells were first infected with C. jejuni and subsequently exposed to the fluorescent endocytic tracer dextran, which was chased into lysosomes. As a control, cells were infected with a strain of S. typhimurium carrying a mutation in invA and a plasmid expressing the Yersinia pseudotuberculosis protein invasin. InvA is an essential component of the invasion-associated type III secretion system [39] and therefore this strain enters cells through the invasin-mediated pathway [40]. Subsequent to its uptake, the invasin-expressing bacteria is delivered to lysosomes (Watson and Galán, unpublished data). Immunofluorescense microscopy analysis revealed that 90% of the vacuoles containing S. typhimurium invA (invasin) colocalized with the endocytic tracer dextran (Figure 3A and Figure S1). In contrast, only ∼15% of the C. jejuni–containing vacuoles acquired detectible amounts of dextran (Figure 3A and Figure S1). These data suggest that the CCV is functionally separated from the canonical endocytic pathway. To confirm these results, we used another endocytic tracer, gold-labeled bovine serum albumin (BSA-gold) [41]. Cos-1 cells were first infected with C. jejuni or S. typhimurium invA (invasin) and subsequently exposed to BSA-gold, which was chased into lysosomes (see Materials and Methods). BSA-gold was then imaged using electron microscopy to determine if the CCV was accessible to this fluid phase endocytic tracer. Although BSA-gold colocalized with ∼75% of the vacuoles containing S. typhimurium invA (invasin) (Figure 3B, 3C and 3I), only 15% of the CCVs were accessible to the endocytic marker (Figure 3D, 3E and 3I). Furthermore, the CCVs that colocalized with BSA-gold appeared to be morphologically different from the CCVs that did not. The CCVs accessible to the endocytic tracer were spacious and contained additional electron dense materials, closely resembling lysosomes (Figure 3F). In contrast, the CCVs that did not colocalize with BSA gold had tight membranes around the bacteria and the compartments did not resemble lysosomes (Figure 3D and 3E). These data confirm the results obtained using fluorescence microscopy and provide additional evidence that the CCV is segregated from the canonical endocytic pathway.
Since our experiments established that C. jejuni quickly loses viability within macrophages, we hypothesized that in these cells this bacterium may be delivered to lysosomes. To test this hypothesis, BMDM were infected with C. jejuni and then incubated with media containing BSA-gold and examined by electron microscopy as described in Materials and Methods. In contrast to what we observed in epithelial cells, in macrophages over 90% of the CCVs were readily accessible to the endocytic tracer (Figure 3G, 3H and 3I). Similar results were obtained using fluorescent dextran as an endocytic tracer (data not shown). Taken together, these data indicate that in macrophages, C. jejuni is delivered to lysosomes where it cannot survive, while in intestinal epithelial cells C. jejuni is segregated from an endocytic pathway leading to lysosomes and consequently survives in a vacuolar compartment that is distinct from lysosomes.
We next examined whether avoidance of lysosomal delivery was essential for C. jejuni survival within epithelial cells. To this end, we carried out an experiment in which C. jejuni was internalized via the Fc receptor, a pathway known to lead to lysosomes [42]. Cos-1 cells expressing the murine Fc receptor were infected with either opsonized or non-opsonized C. jejuni and at different times after infection the CFU were determined by plating under the permissive oxygen-limiting conditions and further incubation in 10% CO2 environment. As shown in Figure 4A, the relative number of CFU recovered from cells infected with opsonized C. jejuni 24 h after infection was significantly (> 20 fold) lower than the number of CFU recovered from cells infected with non-opsonized bacteria. These results indicate that internalization via the Fc receptor results in a significant loss of intracellular viability, presumably because these bacteria are ultimately delivered to lysosomes. To confirm this hypothesis, we examined whether C. jejuni internalized via the Fc receptor was accessible to an endocytic tracer. Cos-1 cells expressing the mouse Fc receptor were infected with opsonized C. jejuni and subsequently exposed to fluorescent dextran. Consistent with the hypothesis that the loss of viability of C. jejuni internalized via the Fc receptor was due to its delivery to lysosomes, >80% of the opsonized bacteria colocalized with fluorescent dextran, compared to ∼20% of non-opsonized control (Figure 4B). Taken together, these data show that C. jejuni is unable to survive within lysosomes and further indicate that this bacterium has evolved a mechanism to avoid delivery to this compartment in order to survive within intestinal epithelial cells. Furthermore, these results also indicate that the mechanism of bacterial entry into host cells has a major impact in the ability of C. jejuni to survive intracellularly.
To investigate the biogenesis and trafficking of the CCV, we examined the dynamics of acquisition of both early and late endosomal markers. Cos-1 cells were infected as described in Materials and Methods and at different times after infection the presence of different endocytic markers was probed by immunofluorescence microscopy using specific antibodies. Fifteen minutes after infection the majority (∼65%) of the intracellular bacteria co-localized with the early endosomal marker EEA-1 (Figure 5A and Figure S2). However, by 60 min, only ∼20% of the CCV co-localized with EEA-1 and more than 80% were stained by an antibody directed to the late endosomal marker lamp-1 (Figure 5B and Figure S2). Two hours after infection, almost all CCVs stained with lamp-1. The acquisition of lamp-1 may therefore occur via an alternative pathway since, as shown above, the CCV does not fuse with lysosomes.
In an effort to better understand the nature of the C. jejuni compartment, we tested the CCV for the presence of other markers of the early and late endocytic pathway. We found that early in infection, 65–70% of the CCVs co-localized with the early endosomal markers Rab4, Rab5, and with a probe for phosphoinisitide 3 phosphate (green fluorescent protein fused to the PX domain of the 40 kD subunit of the nicotinamide adenine nucleotide phosphate oxidase) (Figure 5B-5D, Figure S3 and Video S1). Two hours after infection, less than 10% of the CCVs colocalized with any of these markers indicating that the CCV interacts transiently with these compartments (Figure 5B-5D). Furthermore, C. jejuni transiently acquires the late endosomal marker Rab7 (Figure 5E, 5F and Figure S4). At 45 min after infection, ∼65% of CCVs acquired Rab7-GFP, while at 2 h, only ∼20% of CCVs colocalized with Rab7-GFP. However, consistent with the observation that the mature CCV does not co-localize with endocytic tracers, the lysosomal marker cathepsin B was seen in only a very small proportion of the CCVs, even at 2 h after infection (Figure 5H and Figure S4), although it was present in 95% of vacuoles containing S. tyhimurium invA (invasin) (Figure 5G and Figure S4). These data further indicate that C. jejuni survives within a unique intracellular compartment that despite harboring the lamp-1 protein, is functionally distinct from lysosomes.
Our results indicate that at some point after internalization, the CCV deviates from the canonical endocytic pathway. Therefore, we set out to determine at what stage of the endocytic pathway this segregation might occur. The GTPases Rab5 and Rab7 are involved in the biogenesis of early and late endosomes, respectively [43]. Overexpression of dominant negative forms of these GTPases disrupt temporal and spatial delivery of internalized cargo to lysosomes [43]. To determine if acquisition of lamp-1 by the CCV required Rab5 or Rab7, Cos-1 cells were transfected with wild type or dominant negative forms of Rab5 (Rab5S34N) or Rab7 (Rab7N125I). The transfected cells were infected with C. jejuni and the acquisiton of lamp-1 by the CCV was assessed by immunofluorescence microscopy. As a control, a similar experiment was conducted using S. typhimurium invA (invasin), which traffics to lysosomes. Overexpression of Rab5S34N and Rab7N125I did not affect lamp-1 acquisition by the CCV, although it effectively prevented acquisition of this marker by the vacuoles containing S. typhimurium invA (invasin) (Figure 6 and Figure S5). These results demonstrate that the CCV acquires lamp-1 by an alternative pathway apparently segregating from the canonical endocytic pathway early after C. jejuni internalization.
Collectively, our data suggest that the mechanism by which C. jejuni enters epithelial cells may ultimately determine its intracellular fate. Although internalization through Fc receptors delivers C. jejuni to lysosomes, when entering via its own specific adaptations C. jejuni segregates from the endocytic pathway and avoids delivery to lysosomes. The mechanisms of C. jejuni internalization are unusual in that they do not require the actin cytoskeleton and are dependent on microtubules [10]. In fact, disruption of the actin cytoskeleton increases the efficiency of bacterial uptake (Figure S6). Previous studies have shown that addition of filipin, an agent that sequesters cholesterol, decreased the ability of C. jejuni to enter into cultured epithelial cells [44,45], suggesting that lipid rafts or caveolae may be required for efficient entry into cells. Consistent with this observation, we found that addition of the cholesterol-sequestering agent methyl-beta cyclodextrin (MβCD) blocked C. jejuni internalization into T84 in a dose-dependent manner (Figure S6). Similar results were obtained with Cos-1 cells (data not shown). To further investigate the potential involvement of lipid rafts or caveolae in C. jejuni internalization, we examined the CCV for the acquisition of caveolin-1 and flotillin-1, two markers associated with these membrane domains [46–49]. Cos-1 cells were transfected with plasmids encoding GFP-tagged forms of caveolin-1 or flotillin-1, and the association of these markers with the CCV was examined by time lapse and fluorescence microscopy as described in Materials and Methods. C. jejuni acquired caveolin-1-GFP and flotillin-1-GFP immediately after internalization (Figure 7A and 7B). Quantification of this association determined that at early time points during infection, ∼60% of the CCVs colocalized with both caveolin-1-GFP (Figure 7C) and flotillin-1-GFP (Figure 7D and Video S2). The association, however, was transient since at later points after infection <10% of the CCVs were seen in association with these markers (Figure 7C and 7D). Vesicles devoid of bacteria but labeled by flotillin-1-GFP were also observed immediately after C. jejuni internalization, and some of them eventually fused with the nascent CCV (Video S3 and S4).
Caveolin-1 and flotillin-1 have been shown to be involved in various endocytic events, including the internalization of microbial pathogens [50]. We therefore further examined the potential involvement of caveolin-1 or flotillin-1 in C. jejuni internalization into non-phagocytic cells. Depletion of flotillin-1 by siRNA did not result in a measurable decrease in the abiliy of C. jejuni to enter cells (p = 0.13) (Figure S7). However, expression of a dominant interfering mutant of caveolin-1 (caveolin-1Y14F) significantly decreased C. jejuni internalization (p = 0.02) (Figure 7E). Taken together these data indicate that caveolin-stabilized lipid membrane domains (i.e., caveolae) are important for C. jejuni efficient entry into non-phagocytic cells.
The GTPase dynamin is involved in pinching off of the nascent endosome in both clathrin- and caveolae-mediated endocytosis [51–54]. We therefore tested the potential involvement of dynamin II in C. jejuni internalization. Cos-1 cells were transfected with a plasmid expressing a dominant negative form of dynamin II (dynIIK44A), which has been shown to inhibit both clathrin and caveolae-dependent endocytosis [52–54], and the ability of C. jejuni to enter those cells was examined by fluorescence microscopy as described in Materials and Methods. Although expression of dynIIK44A-GFP effectively blocked the uptake of transferrin (data not shown), it did not inhibit the uptake of C. jejuni (Figure 7F). In fact, there was a modest enhancement of C. jejuni internalization in the presence of wild-type dynamin II. These results indicate that the role of caveolin-1 C. jejuni entry into cells may not be related to its role in caveolae-mediated endoyctosis. Rather, caveolin-1 or caveolae may play a role in the signaling events leading to bacterial uptake, perhaps by facilitating the spatial organization of critical signaling molecules. In fact, tyrosine kinases have been reported to be essential for C. jejuni entry into host cells [44,45,55], a result that we have confirmed (Figure S6). Since efficient signaling through receptor tyrosine kinases is known to require lipid rafts or caveolae, it is possible that the inhibitory effect of cholesterol sequestering agents or dominant negative caveolin-1 may be the result of interference with tyrosine kinase signaling.
Examination of the localization of the CCV over time showed that at 4–5 h after infection, C. jejuni localized to the perinuclear region [56]. To gain more insight into the specific localization of the CCV in relation to other organelles, we investigated by immunoflurescence microscopy the position of the CCV in relation to the Golgi apparatus over time using an antibody directed against GM130, a Golgi resident protein. Two hours after infection, intracellular C. jejuni were evenly distributed around the cell (Figure 8A). However by 6–8 h of infection, >85% of the CCVs were seen in close association with the Golgi apparatus (Figure 8B and 8E), close to the microtubule organizing center (Figure S8). The close association of the CCV and the Golgi does not represent a default pathway for any internalized particle traveling to a perinuclear position since internalized S. typhimurium invA (invasin) did not show association to the Golgi despite the fact that these phagosomes were also located in a perinuclear region (Figure 8C). Electron microscopy analysis confirmed the intimate association of the CCV and the Golgi apparatus (Figure 8D). However, the CCV did not acquire Golgi markers (data not shown), indicating that despite their close association, the two compartments remain distinct.
We then investigated the mechanism by which the CCV reaches its perinuclear destination. Many intracellular bacteria in membrane-bound compartments traffic along microtubule tracks to reach their destination within the cell. We first investigated the role of microtubules in the localization of the CCV. When nocodazole was added to disrupt the microtubule network after infection of Cos-1 cells, the CCVs were observed distributed throughout the cell and did not reach a perinuclear location (Figure 8F). These data show that intact microtubules are necessary for the CCV to reach its final destination. Phagosomes most often move along microtubules through the action of specific motors [57,58]. Cytoplasmic dynein is a minus-end directed motor that is responsible for moving cargo away from the periphery and toward the microtubule organizing center [59] and is therefore a candidate motor to move the CCV to its final destination. We investigated this hypothesis by overexpressing GFP-dynamatin p50, a subunit of the dynactin complex that when overexpressed, blocks the function of dynein [60]. Immunofluorescence analysis showed that overexpression of dynamatin p50 effectively disrupts the localization of C. jejuni at 6 h post-infection (Figure 8G). These results are consistent with a previous observation indicating that addition of orthovanadate, a rather non-specific inhibitor of dynein, inhibits the movement of the CCV to a perinuclear position [56]. Taken together, these results indicate that subsequent to internalization, the CCV travels to a perinuclear position in the immediate vicinity of the Golgi apparatus and that the movement of the CCV requires both microtubules and the molecular motor dynein.
Similar to other enteric pathogens, C. jejuni has evolved the ability to gain intracellular access to non-phagocytic intestinal epithelial cells and this process has been implicated in pathogenesis [7,14,20,61,62]. Although most work to date has focused on C. jejuni entry into host cells, the intracellular fate of this pathogen has been largely uncharacterized. We have shown here that C. jejuni survives within intestinal epithelial cells, although over time it acquires a metabolic state that renders it unculturable under standard culture conditions. However, C. jejuni recovered from within epithelial cells could be cultured if subjected to conditions of severe oxygen limitation. These results suggest that once within epithelial cells, C. jejuni may either become oxygen sensitive or may alter its respiration mode so that it can no longer be cultured in the presence of oxygen. The recently completed nucleotide sequence of the genome of the C. jejuni strain 81–176 used in this study revealed the presence of genes involved in additional respiratory pathways, including electron acceptors that may be utilized for alternate modes of respiration [63]. Thus, these additional respiration genes may contribute to the ability of C. jejuni 81–176 to survive within intestinal epithelial cells.
We showed here that C. jejuni survives within intestinal epithelial cells within a compartment that is distinct from lysosomes (Figure 9). CCVs are not accessible to endocytic tracers indicating that they are functionally separated from described endocytic pathways leading to lysosomes. In fact, when targeted into lysosomes after internalization via the Fc receptor, C. jejuni was unable to survive within epithelial cells. These results indicate that C. jejuni has evolved specific adaptations to traffic within host cells and avoid delivery into lysosomes. These adaptations may be important to faciliate colonization of the host by providing a safe-heaven where C. jejuni can avoid innate immune defense mechanisms. However, those adaptations must not be able to operate in macrophages since, in these cells, C. jejuni is targeted to lysosomes and therefore cannot survive.
Our results suggest that C. jejuni deviates from the canonical endocytic pathway shortly after internalization (Figure 9). The CCV appears to interact with early endosomal compartments since it associates with early endosomal markers such as EEA-1, Rab5, Rab4, and PX-GFP (which labels PI3P). However, this interaction is transient and does not lead to progression within the canonical endocytic pathway. The presence of markers of lipid-associated rafts and caveolae on the CCV suggests that C. jejuni may reside in a compartment that is functionally distinct from early endosomes. In fact, C. jejuni was still able to target properly in the presence of dominant interfering mutants of Rab5 or Rab7, which control early events in the endocytic pathway [43].
C. jejuni has unusual cytoskeletal requirements to gain intracellular access to intestinal epithelial cells since its internalization is dependent on microtubules but not on the actin cytoskeleton [10], as is usually the case for most intracellular bacteria [22]. Consistent with previous observations suggesting that caveolae are required for C. jejuni entry [44,45], we have shown here that bacterial internalization is dependent on caveolin-1. However, we showed that the entry process is independent of dynamin, whose function is essential for clathrin and caveolae-mediated endocytosis [51–53]. In fact, expression of a dominant-inhibitory form of dynamin resulted in a reproducible increase in the ability of C. jejuni to enter cells. We therefore hypothesize that a caveolin-1-stabilized lipid membrane domain may be required for proper signaling through tyrosine kinases, which are also required for C. jejuni internalization rather than for endocytosis. In fact, efficient signaling through receptor tyrosine kinases requires lipid rafts or caveolae [64,65] and the surface availability of many these is regulated by dynamin [66]. In this context, we hypothesize that the enhancement of C. jejuni internalization observed when inhibiting dynamin function, may be the result of an increase in the availability of putative surface “receptor” for this pathogen resulting in enhanced signaling for entry.
What is the nature of the CCV? We showed that the CCV contains lamp-1, although this compartment is unique and clearly distinct from lysosomes since it does not colocalize with the lysosomal protein marker cathepsin B and it is not accessible to endocytic tracers. In fact, the acquisition of lamp-1, which occurs very early in the CCV maturation, must occur by an unusual mechanism that does not require the GTPases Rab5 or Rab7. Interestingly, S. typhimurium resides within a vacuole that is apparently segregated from the canonical endocytic pathway [27,67] and also harbors lamp-1, although in this case acquisition of this marker appears to require Rab7 [68]. Another unique property of the CCV is its close association with the Golgi, which requires microtubules and the motor protein dynein. More studies will be required to better define the nature of this compartment and the precise mechanisms by which C. jejuni modulates vesicular trafficking.
In summary, we have established that C. jejuni has evolved specific adaptations to survive within intestinal epithelial cells by avoiding delivery into lysosomes. This survival strategy does not appear to operate in BMDM since C. jejuni is rapidly killed in these cells. We hypothesize that C. jejuni's unusual entry mechanism may be central to its ability to avoid delivery into lysosomes since when internalized via a different pathway (e.g., via the Fc receptor), C. jejuni could not avoid delivery into lysosomes. Its diversion from a pathway leading to lysosomes must therefore occur upon entry. Understanding the mechanism by which this bacterium survives within host cells may provide new insights into C. jejuni pathogenesis as well as reveal undiscovered paradigms in host cellular trafficking.
Wild-type C. jejuni 81–176 has been described previously [69]. C. jejuni were routinely grown on tryptic soy broth agar supplemented with 5% sheep blood (BA) or in brain heart infusion (BHI) broth at 37 °C under 10% CO2, or where indicated, in an anaerobic chamber under low oxygen conditions (BD-Diagnostic Systems GasPak Plus Anaerobic System Envelopes with Palladium Catalyst, catalog number 271040, New Jersey), or with BBL CampyPacks (BBL Microbiology Systems, Cockeysville, MD). S. typhimurium invA has been described previously [70] and was transformed with invasin-encoding plasmid pRI203, which mediates mammalian cell entry via αβ1 integrins [71]. A S. typhimurium invA strain expressing the dsRed protein was constructed as follows. The plasmid DsRed.T3_S4T, which expresses the dsRed protein under the control of an arabinose-inducible promoter [72], was digested with EcoRI and ScaI to release a fragment containing dsRed and the paraABC promoter. This fragment was ligated into pACYC184 and resulting plasmid, pSB3082, was transformed into S. typhimurium invA (pRI203). S. typhimurium invA expressing invasin and dsRed was routinely grown in LB containing tetracycline (10 μg ml−1), ampicillin (50μg ml−1), and 0.1% arabinose to induce DsRed expression.
T84, a human intestinal epithelial cell line, and Cos-1, a monkey kidney epithelial cell line, were obtained from the American Type Culture Collection (Rockville, MD) and were grown in DMEM supplemented with 10% fetal bovine serum containing penicillin (100 U ml−1) and streptomycin (50 μg ml−1). Bone marrow–derived macrophages (BMDM), were obtained as previously described [73]. Briefly, femurs and tibias were excised and flushed with DMEM containing 10% fetal bovine serum (FBS), penicillin (100 U ml−1), and streptomycin (50 μg ml−1). Cells were spun down and resuspended in BMDM differentiation medium [DMEM containing 20% FBS, 30% L-cell supernatant, penicillin (100 U ml−1) and streptomycin (50 μg ml−1)] and plated onto non-tissue culture treated 10-cm2 plastic dishes. The cells were fed fresh BMDM differentiation medium on day 3–4 to allow further differentiation until day 6–7. BMDM were then seeded in the appropriate tissue culture dishes to be used in infection experiments.
C. jejuni was harvested from a fresh BA plate and grown in BHI broth under under 10% CO2 until mid log phase (OD600 = 0.7–0.8). To prepare the inoculum, bacteria were pelleted at 20,000 × g in a microfuge for 2 min and directly resuspended in Hank's Balanced Salt Solution (HBSS) (Invitrogen). The inoculum was diluted in HBSS to adjust for different multiplicity of infections (MOI). Serial dilutions of the inoculum were plated onto BA plates to determine the number of bacteria. T84 cells were split to 70% confluence (∼105 cells per well) in a 24 well dish. BMDM were seeded at 2 × 105 cells per well in a 24 wells dish. After washing 3X with HBSS, T84 cells and macrophages were infected with an MOI of 50 or 20, respectively. The plates were centrifuged at 200 × g for 5 min to maximize bacteria-cell contact and incubated for 1 or 2 h at 37 °C 5% CO2. Following the incubation, the monolayers were washed 3X with HBSS and incubated with complete media containing gentamicin (100 μg ml−l) for 2 h. This concentration of gentamicin was found to be optimal to kill extracellular bacteria without affecting the viability of intracellular bacteria. Plating of the infection medium determined that no significant number of c. f. u. were detected after this treatment. For experiments involving longer time points, the media was replaced with complete media containing gentamicin (10 μg ml−1). Again, plating of the infection medium determined that no significant number of c. f. u. were present after this treatment. After 3 additional washes, the infected cells were lysed at the designated time points and the samples were prepared for colony forming unit (CFU) determination or FACS analysis. To quantify the number of intracellular bacteria, cells were lysed in PBS with 0.1% deoxycholate and the CFU were enumerated after plating serial dilutions grown at 37°C with 10% CO2, or in GasPak jars (BBL Microbiology Systems, Cockeysville, MD) with BBL CampyPacks (BBL Microbiology Systems, Cockeysville, MD). For anaerobic incubations, plates were incubated in an anaerobic chamber with a GasPak (BD-Diagnostic Systems GasPak Plus-Anaerobic System Envelopes with Palladium Catalyst, Catalog number 271040, New Jersey) for 48 h and incubated further at 37 °C with 10% CO2. For immunofluorescence studies, Cos-1 cells were seeded on glass cover slips in 24 well plates. C. jejuni was cultured as described above. S. typhimurium invA expressing invasin was cultured in LB containing ampicillin (30 μg ml−1) overnight and subcultured 1:20 for 3 h prior to infection. For time courses, cells were infected with an MOI of 100 and 50 for C. jejuni and S. typhimurium invA (invasin), respectively, centrifuged from 5 min at 1,000 × g to maximize bacteria-host cell contact, and incubated for an additional 15 min at 37 °C 5% CO2. Wells were washed three times in PBS and either fixed in 4% PFA for a 15 min time point or the media was replaced with DMEM + 10% FBS with gentamicin (100 μg ml−1) to kill the extracellular bacteria and prevent additional bacterial internalization. At later time points cells were washed an additional three times in PBS, and fixed in 4% paraformaldehyde. After 4 h or longer infection times, no significant number of bacteria that had remained extracellular but attached to the cell were detected using the gentamicin treatment described (i.e., initial addition of 100 μg/ml for 2 h and subsequent addtion of 10μg/ml for the remainder of the experiment, see above) (Figure S9). For quantitation of intracellular bacteria in transfected cells, Cos-1 cells were infected with an MOI of 25 for 1 h as described above. Cells were washed and incubated for 2 additional hours in DMEM + 10% FBS with gentamicin (100 μg ml−1) to kill the extracellular bacteria and prevent additional bacterial internalization. The cells were then fixed and processed for immunofluorescence as described below.
Where indicated, cells were incubated with Nocodazole (10 μM), Cytocholasin D (5 μM), Genistein (10 μM and 100 μM), dissolved in DMSO (Sigma), or methyl-beta cyclodextrin (MβCD) dissolved in PBS (1mM-10mM) 30 min prior to infection and kept throughout the duration of the incubation period. All control cells were treated with the appropriate solvent for the same length of time.
T84 cells and BMDM from wild-type mice were seeded at density of 105 cells per well on a 24-well dish and infected with an MOI of 50 and 20, respectively. Following a 1 h incubation at 37 °C and 5% CO2, the cells were washed with HBSS and DMEM containing 10% FBS and 100 μg ml−1 gentamicin was added to each well. Cells were washed again and lysed at the designated time points in 500 μl of 0.05% sodium deoxycholate in PBS. The cell lysates were collected and subjected to a low speed spin (1,000 rpm) for 2 min to remove large cell debris. Supernatants were collected and intracellular bacteria were isolated by a 2 min high-speed spin (10,000 rpm). The isolated bacterial pellet was resuspended in 500 μl filter-sterilized staining buffer (PBS containing 1mM EDTA and 0.01% Tween). The bacteria were then stained with the reagents of a cell viability kit (BD Biosciences, San Jose, CA), which distinguishes live and dead cells by using a thiazole orange (TO) solution, which stains all bacteria, and propidium iodide (PI), which only stains dead bacteria. TO and PI were added to final concentrations 53 nM and 11 μM, respectively, in accordance with the manufacturer's instructions. After 5 min of staining, bacteria were pelleted, washed once in PBS, resuspended in 1 ml of PBS and analyzed by flow cytometry. The absolute count of live/dead bacteria was carried out by addition of 50 μl of a liquid suspension of a known number of fluorescent beads (supplied in the kit, BD Biosciences, San Jose, CA) following the manufacturer's instructions. Samples were analyzed on a FACS calibur flow cytometer. TO fluoresces primarily in FL1 and FL2; PI primarily in FL3. An SSC threshold was used, and cells and beads were gated using scatter and FL2, which detects the TO fluorescence and therefore the total bacterial population. In order to best discriminate between live and dead bacteria, an FL1 versus FL3 plot was used and live and dead populations were gated within this plot (dead cells, FL3+; live cells, FL1+). To determine the bacterial concentration, the following equation was used: # events in cell region/ # events in bead region x # beads/test/test volume × dilution factor = concentration of cell population. A plot was generated after using this equation to calculate the number of viable bacteria (in triplicate wells) in both T84 and BMDM at each time point.
Cos-1 cells were infected as described above and at the designated time points, cells were washed three times in PBS and fixed in 4% paraformaldehyde (PFA) for 13 min at room temperature (RT). The fixed cells were washed three times in PBS and permeabilized by incubating them in PBS containing 3% non-fat milk and 0.05% saponin (PBS-MS) (Calbiochem). Cover slips were incubated in primary antibody diluted in PBS-MS for 30 min. The cover slips were then washed three times in PBS and incubated in secondary antibody. After two washes in PBS and two washes in deionized water, the cover slips were mounted onto glass slides using Prolong Gold antifade reagent (Molecular Probes). Images were acquired on a Nikon TGE2000-U Eclipse inverted microscope fitted with a Micromax Princeton digital camera controlled by the Metamorph software package, version 6.1 (Universal Imaging Corp., Downingtown, PA). When needed, inside-out staining was used to differentiate extracellular from intracellular bacteria. Briefly, before permeabilization with saponin, extracellular bacteria were stained with rabbit anti-C. jejuni antiserum in PBS containing 3% milk followed by Alexaflour 350-conjugated anti-rabbit antibodies (Molecular Probes). Cells were washed three times, permeabilized, and the total bacterial population was stained with rabbit anti-C. jejuni followed by Alexaflour 594-conjugated mouse anti-rabbit antibodies (Molecular Probes). After two washes in PBS and two washes in deionized water, the cover slips were mounted onto glass slides using Prolong Gold antifade reagent (Molecular Probes). Where indicated, nuclei stained with DAPI (Invitrogen).
Rabbit antibodies against C. jejuni were obtained by repeated immunization of rabbits with a mixture of equal amounts of formaldehyde and heat-killed whole cell C. jejuni. Anti-S. typhimurium (rabbit) antibodies were purchased from DIFCO Laboratories, Detroit, Michigan. Mouse monoclonal antibodies against EEA-1, Lamp-1, beta-tubulin, and GM130 were acquired from BD Biosciences Pharmingen. A mouse anti cathepsin B antibody was a gift from the laboratory of Dr. Ira Mellman, Yale University, New Haven, CT. Secondary antibodies used were: Alexa 596- Alexa-488, Alexa 350-conjugated goat anti-rabbit and Alexa 596- Alexa 488- Alexa 350-conjugated goat anti-mouse IgG antiserum (Molecular Probes).
Eukaryotic expression vectors encoding GFP-tagged wild-type Rab5 and Rab7 as well as their dominant negative mutants have been previously described [74–76]. GFP-tagged Rab4, PX, dynamin II, dynamin IIK44A, caveolin-1, and caveolin-1Y14F as well as murine Fc-receptorII (FcRII) expressing eukaryotic vectors have been described elsewhere [77–81]. The human flotillin-1 gene was amplified from a human cDNAs library by polymerase chain reaction (PCR) using primers fwd (5′-TAGCTCGAGCCATGTTTTTCACTTGTGGCCC-3′) and rev (5′-TCTAGAATTCCGGCTGTTCTCAAAGGCTTGA-3′). The PCR product was then cloned into pEGFP-N1 (Clontech, Oxford, UK) using XhoI and EcoR1. The resulting plasmid, pSB3111, yielded flotillin-1 fused to the N-terminus of GFP. pFlot1-FLAG was a generous gift from Dr. Rosanna Paciucci, Unitat de Recerca Biomedica, Barcelona, Spain [82]. Plasmid DNA was purified using the Maxiprep kit (Qiagen) and used for transfection of cells with LipofectAMINE 2000 (Invitrogen) according to the manufacturers instructions.
To quantitate the percentage of CCV containing different endocytic fluid tracers or cellular markers, infected cells were visualized directly in the fluorescence microscope. Using the Metamorph software package a series of images were taken, including internalized bacteria, total bacteria, and the cellular marker. Overlayed fluorescent images were analyzed by determining the number of CCVs that contained the corresponding marker. A minimum of one hundred vacuoles was analyzed per cover slip for each treatment and designated post-infection time. Each experiment was completed in triplicate wells/cover slips and expressed as an average. CCVs were considered positive for the presence of a marker when they contained detectable amounts of the staining probe/antibody. The same methodology was used to quantitate the number of intracellular bacteria within transfected cells. Briefly, Cos-1 cells seeded on cover slips were transfected with plasmids encoding GFP-tagged proteins, infected, and processed for immunofluorescence. A minimum of 50 transfected cells were randomly selected and imaged as described above and the number of intracellular bacteria was quantitated within each cell. Each experiment was completed in triplicate wells/cover slips and expressed as an average number of bacteria per transfected cell and where designated normalized to control cells. Statistical analysis of the results was carried out by the student t-test.
Cos-1 cells were grown on coverslips in 24-well dishes and infected with C. jejuni or S. typhimurium invA (invasin) and dsRed at an MOI of 25 and 10 respectively. After spinning for 5 min at 200 × g and allowing the infection to proceed for 1 h, the cells were washed three times and fresh media containing gentamicin was added to kill the extracellular bacteria. After an additional 3 h the cells were incubated with Texas Red or Alexafluor-488 labeled dextran (1 mg/ml; Molecular Probes) for 1 h. The cells were washed three times and incubated with fresh media containing gentamicin for an additional 2 h. The cover slips were then processed for immunofluorescence.
Cos-1 cells were plated in a 24 well dish at ∼40% confluency and transfected with a plasmid encoding FcRII [80]. After an overnight incubation, cells were incubated with C. jejuni that were opsonized with rabbit polyclonal anti-C. jejuni antibodies (1:1,000). Plates were centrifuged at 200 × g for 5 min and incubated at 37 °C 5% CO2 for 1 h, after which the cells were washed three times with HBSS, and incubated with complete media containing gentamicin (100 μg ml−1). Cells were then lysed at the designated time points and the number of viable intracellular bacteria was assessed as described above.
Cos-1 cells were grown on life cell imaging dishes (MatTek Corp.) and transfected with flotillin-1-GFP or PX-GFP [78] encoded eukaryotic expression vectors as described above. C. jejuni was cultured as described above and fluorescently labeled using PKH26 Red Flourescent Cell Linker Kit according to manufacturer's instructions (Sigma). The labeling protocol did not affect C.jejuni's viability or its ability to enter into non-phagocytic cells (data not shown). Cells were infected with an MOI of ∼100. Time-lapse series were acquired using a Nikon TGE2000-U Eclipse inverted microscope fitted with a Micromax Princeton digital camera controlled by the Metamorph software package, version 6.1 (Universal Imaging Corp., Downingtown, PA). Filters allowed the simultaneous detection of GFP and rhodamnine, respectively. The acquired images were merged as RGBs (for two color movies) and converted into QuickTime movies using Metamorph. For the flotillin-1-GFP three dimensional movie, cells transfected with flotillin-1-GFP were infected for 15 min, fixed, and processed for imunoflourescence using inside-out staining. Z-stacks were acquired using a Zeiss Axio Imager Upright fluorescent Microscope fitted with an Apotome and the AxioCam MRc5 digital camera controlled by the Axiovision software package, version 4.2 (Carl Ziess MicroImaging, Inc.). The cropped images were reconstructed and converted into a 3D QuickTime movie using Axiovision.
Depletion of flotillin-1 was performed using a pool of three target-specific 20–25 nucleotide siRNAs designed to knock down gene expression (Santa Cruz Biotechnology, Inc., Santa Cruz, CA). The siRNA pool was transfected into Cos-1 cells using LipfectAMINE 2000 (Invitrogen). To mark transfected cells, flotillin siRNA was cotransfected with pEGFP-N1 at a ratio of 5:1. Intracellular bacteria within GFP-expressing cells were quantified as described above. RNAi silencing efficiency and specificity were analyzed at the protein level by Western blot analysis 72 h after cotransfection of pFlot1-FLAG [82] with the siRNA pool. Protein depletion by RNAi was normalized to endogenous levels of actin using rabbit anti-actin antibodies (Sigma-Aldrich).
Cos-1 cells or BMDM were plated at approximately 8 × 106 cells on 10 cm2 plastic tissue culture dishes. After 3x wash with warm HBSS, C. jejuni, cultured as described above, were used to infect cells at an MOI of 100. At 1.5 h post-infection, the extracellular bacteria were washed off three times with warmed HBSS and the plates were incubated with warmed culture media containing gentamicin to kill the extracellular bacteria. For standard EM, the infected cells were incubated at 37 °C 5% CO2 for and additional 4.5 h. For BSA-gold experiments, after a 3.5 h incubation, the infected cells were incubated for 1 h with BSA-gold-containing complete media. After three additional washes cells were fixed in situ with a freshly made solution of 1% glutaraldehyde (from an 8% stock from Electron Microscopy Sciences (EMS), Fort Washington, PA) 1% OsO4 in 0.05 M phosphate buffer at pH 6.2 for 45 min. After fixation, the cells in petri plates were rinsed three times with cold distilled water and en bloc stained with uranyl acetate overnight. The petri plates were then dehydrated in ethanol then placed into hydroxypropyl methacrylate (EMS), which does not react with the plastic in the petri dish, and embedded in L 112, an epon substitute (Ladd, Burlington, VT). Following polymerization of the epon, the block was cut out and mounted and thin sections were cut through their exposed surfaces. Thin sections were collected on naked grids stained with uranyl acetate and lead citrate and examined in a Philips 200 electron microscope. At least fifty vacuoles were analyzed for the presence of BSA-gold for each condition. |
10.1371/journal.ppat.1004340 | A Novel Signal Transduction Pathway that Modulates rhl Quorum Sensing and Bacterial Virulence in Pseudomonas aeruginosa | The rhl quorum-sensing (QS) system plays critical roles in the pathogenesis of P. aeruginosa. However, the regulatory effects that occur directly upstream of the rhl QS system are poorly understood. Here, we show that deletion of gene encoding for the two-component sensor BfmS leads to the activation of its cognate response regulator BfmR, which in turn directly binds to the promoter and decreases the expression of the rhlR gene that encodes the QS regulator RhlR, causing the inhibition of the rhl QS system. In the absence of bfmS, the Acka-Pta pathway can modulate the regulatory activity of BfmR. In addition, BfmS tunes the expression of 202 genes that comprise 3.6% of the P. aeruginosa genome. We further demonstrate that deletion of bfmS causes substantially reduced virulence in lettuce leaf, reduced cytotoxicity, enhanced invasion, and reduced bacterial survival during acute mouse lung infection. Intriguingly, specific missense mutations, which occur naturally in the bfmS gene in P. aeruginosa cystic fibrosis (CF) isolates such as DK2 strains and RP73 strain, can produce BfmS variants (BfmSL181P, BfmSL181P/E376Q, and BfmSR393H) that no longer repress, but instead activate BfmR. As a result, BfmS variants, but not the wild-type BfmS, inhibit the rhl QS system. This study thus uncovers a previously unexplored signal transduction pathway, BfmS/BfmR/RhlR, for the regulation of rhl QS in P. aeruginosa. We propose that BfmRS TCS may have an important role in the regulation and evolution of P. aeruginosa virulence during chronic infection in CF lungs.
| The rhl quorum-sensing (QS) system allows P. aeruginosa to regulate diverse metabolic adaptations and virulence. However, how rhl QS system is regulated remains largely unknown. Here, we report that two-component sensor BfmS controls rhl QS system by repressing its cognate response regulator BfmR, which directly suppresses the expression of rhl QS regulator RhlR gene and reduces the production of QS signal molecule N-butanoyl-L-homoserine lactone (C4-HSL). We find that BfmS is critical to the ability of P. aeruginosa to modulate the expression of virulence-associated traits and adapt to the host. Intriguingly, although wild-type BfmS is a repressor of BfmR, naturally occurring missense mutation (L181P, L181P/E376Q, or R393H) can convert its function from a repressor to an activator of BfmR, leading to BfmR activation, which in turn reduces the level of rhl QS signal C4-HSL. These results, therefore, provide important and novel insight into the regulation and evolution of P. aeruginosa virulence.
| Pseudomonas aeruginosa is an important opportunistic pathogen that accounts for 10% of all hospital-acquired infections [1], [2]. Most notably, P. aeruginosa is the leading cause of chronic pulmonary infections and mortality in cystic fibrosis (CF) patients [3]. The success of P. aeruginosa relies on the production and precise coordination of numerous virulence-associated factors such as lipopolysaccharide, flagella, type IV pili, exopolysaccharide alginate, toxins, proteases, lipases, pyocyanin, and rhamnolipids, which are primarily controlled by regulatory systems such as the quorum-sensing (QS) system and the two-component system (TCS) [4]–[12].
P. aeruginosa has two well-characterized acyl-homoserine lactone (acyl-HSL)- based QS systems, las (LasR-LasI) and rhl (RhlR-RhlI) [4]–[6], [8]–[10]. In addition, a third Pseudomonas quinolone signal (PQS) acts as a link between the las and rhl QS systems, although PQS is not involved in sensing cell density [4]–[6], [8]–[10]. The synthase of LasI catalyzes the synthesis of N-(3-oxododecanoyl) homoserine lactone (3O-C12-HSL), whereas RhlI catalyzes the synthesis of N-butyryl-homoserine lactone (C4-HSL), which induces their respective cognate transcriptional regulators LasR and RhlR, responsible for the activation of numerous QS-controlled genes [4]–[6], [8]–[10]. The transcriptional regulator LasR is generally considered to sit at the top of the QS hierarchy in P. aeruginosa. LasR/3O-C12-HSL activates the transcription of rhlR, and RhlR/C4-HSL activates the transcription of rhlI and various virulence-associated genes [4]–[6], [8]–[10]. However, RhlR is able to control the expression of LasR-specific factors independent of LasR [13]. 2-(2-hydroxylphenyl)-thiazole-4-carbaldehyde (IQS) could activate the rhl system in a LasR-independent manner [14]. Thus, the regulation of the rhl QS system is much more complicated than previously thought. So far, LasR and Vfr are the two transcriptional regulators known to regulate the expression of rhlR directly, other than RhlR itself [4]–[6], [8]–[10], [15].
Pathogenic bacteria, including P. aeruginosa, probes its surrounding environment constantly and makes appropriate decisions during infection [7], [11], [12], [16]–[18]. An important molecular device to achieve sampling of environmental signals is the two-component system (TCS) [12], [16]. Classically, two-component systems are composed of an inner membrane-bound sensor, which is able to detect an environmental stimulus, and a response regulator, which is phosphorylated by the sensor kinase and which, in turn, modulates the expression of genes necessary for the appropriate physiological response [16]. Approximately 130 genes encoding for TCS components have been identified in the genome of P. aeruginosa [1], [7], [11]. This provides P. aeruginosa with a sophisticated capability to regulate diverse metabolic adaptations, virulence and antibiotic resistance processes [7]. In fact, a large number of TCSs or TCS components, such as GacSA, PhoPQ, SadARS, RetS, and LadS have been described as having a key role during the infection process [7], [8], [11]. However, the direct links between TCS and QS remain poorly understood [6]–[8], [11], [19]–[22].
The observation that the expression of BfmRS TCS was dramatically up-regulated in the lungs of cystic fibrosis patients compared to in vitro growth intrigued us [23]. We sought to determine the roles of BfmRS in virulence regulation in P. aeruginosa. In this study, we showed that BfmRS TCS directly controls rhl QS system and modulates the ability of P. aeruginosa to adapt to the host. We demonstrate that BfmRS TCS may play important roles in the regulation and evolution of bacterial virulence during long-term bacterial adaptation to lungs afflicted with cystic fibrosis.
In P. aeruginosa, BfmS (PA4102) is a putative two-component sensor kinase with uncharacterized functions although its cognate response regulator BfmR (PA4101) has been reported to play an important role in biofilm maturation [24], [25]. To probe the biological roles of BfmS, we generated a bfmS null mutant strain (ΔbfmS) as described in the Materials and Methods section and in our previous studies [26]. Interestingly, ΔbfmS strain was defective in green pigment and rhamnolipids (Figure 1A), which can be complemented by introducing the native copy of bfmS (Table S1 in Text S1) into the ΔbfmS strain (Figure 1A). Quantitative analysis of pyocyanin and rhamnolipids indicates that the deletion of bfmS results in a 3.5-fold decrease of pyocyanin production and a 5-fold decrease in rhamnolipid production respectively (Figure S1A and S1B). Given that rhamnolipids promote the swarming motility of P. aeruginosa [27], we next examined the swarming motility of a wild-type MPAO1 strain, a ΔbfmS strain and its complementary strain (ΔbfmS/p-bfmS). As shown in Figure S1C, deletion of bfmS abolished swarming whereas both wild-type MPAO1 strain and the complementary strain swarmed on the surface of plates at 36 h.
The rhlAB operon is required for rhamnolipid synthesis [4]–[6], [9], [10]. We therefore constructed an rhlA promoter-lux fusion (rhlA-lux, Table S1 in Text S1) and measured its activity in a wild-type MPAO1 strain, a bfmS deletion strain (ΔbfmS), and its complementary strain (ΔbfmS/p-bfmS). The expression of rhlA-lux fusion in ΔbfmS was significantly lower than those of other strains when bacteria were grown in an M8-glutamate minimal medium supplemented with 0.2% glucose (Figure S1D). This result suggests that the decreased expression of rhlAB in ΔbfmS strain is likely responsible for the reduction in rhamnolipid production. Since the expression of rhlAB is positively controlled by the rhl quorum-sensing system in P. aeruginosa [4]–[6], [9], [10], we next sought to measure the RhlI-dependent autoinducer C4-HSL content in the wild-type MPAO1 strain, the ΔbfmS strain, and the complementary strain (ΔbfmS/p-bfmS). We used the pDO100 (pKD-rhlA) system [28] (Table S1 in Text S1) that carries a lux reporter fused with a rhlA promoter. As a result, supernatants prepared from either the wild-type MPAO1 strain or the complementary strain (ΔbfmS/p-bfmS), but not the ΔbfmS strain, markedly promoted the luminescence values and thereby C4-HSL levels (Figure 1B). We also observed that deletion of bfmS results in decreased rhlI promoter activity (Figure S1E). Based on these results, we conclude that BfmS positively controls rhl QS system in P. aeruginosa.
To further study the roles of bfmS, we used microarray analysis in order to compare the transcriptome of the ΔbfmS strain with that of the wild-type MPAO1 strain. As a result, we identified 131 genes with increased transcript levels (≥2-fold) (Table S2 in Text S1) and 71 genes with decreased transcript levels (≤2-fold) (Table S3 in Text S1) in the ΔbfmS strain versus wild-type MPAO1 strain. These 202 genes represent ∼3.6% of the total number of annotated genes in the P. aeruginosa PAO1 genome. Of those 202 genes, 42% encode hypothetical proteins of unknown functions (Tables S2 and S3 in Text S1). Grouping these genes according to their annotated function shows that they belong to several functional categories, primarily transport of small molecules, carbon compound catabolism, translation, and adaptation (Tables S2 and S3 in Text S1).
Among the 131 genes whose expression is up-regulated in the ΔbfmS strain, 7 genes were up-regulated more than 10-fold. Interestingly, these 7 genes, including PA4100, bfmR, PA4103, PA4104, PA4105, PA4106, and PA4107, are located at or near the bfmRS (PA4101-PA4102) loci (Figure S2A, Table S2 in Text S1). These microarray-based expression data are consistent with the operon predictions for P. aeruginosa, which suggested that PA4103 and PA4104 are organized into PA4103 operon (PA4103-PA4104) while PA4105, PA4106 and PA4107 are organized into PA4107 operon (PA4107-PA4106-PA4105) (www.pseudomonas.com). Among these genes, PA4100 encodes a dehydrogenase of unknown function, and bfmR encodes a two-component response regulator that acts as a biofilm maturation regulator, whereas PA4103, PA4104, PA4105, PA4106, and PA4107 encode hypothetical proteins. Although PA4103 contains a ferric reductase like transmembrane component (pfam01794) and PA4107 contains a calcium binding motif (cd00051), their biological functions are unknown. Further characterization of the functions of these genes may provide insight into the roles of bfmS in P. aeruginosa.
There are 11 genes whose expressions are down-regulated more than 10-fold in ΔbfmS strain (Figure S2, Table S3 in Text S1). Among them, 5 genes (rhlA, rhlB, antA, antB, and antC) are already known to be controlled by the rhl QS system [29], [30]. In addition, we observed that deletion of bfmS decreases the transcription of rhlI by approximately 66% (Table S3 in Text S1). We also found a moderate, 20% decrease in rhlR transcription in the ΔbfmS mutant compared with the parent, which is consistent with the results of an rhlR-lux reporter gene analysis (Figure S3A). These results further suggest that bfmS positively controls the rhl QS. To further confirm the differentially expressed genes identified by the microarray analysis, 12 genes representing a range of microarray signal intensity and expression profiles were subjected to real-time (RT) PCR analyses. There was a high degree of consistency among data generated by these two methods (Table S4 in Text S1), which assures the reliability of microarray analysis in determining transcriptional changes.
Since deletion of bfmS led to the over-expression of bfmR (>90-fold) (Figure S2A, Table S2 in Text S1), we next sought to determine if the elevated bfmR contributes to the phenotypes observed in the ΔbfmS strain. We generated a bfmRS double mutant strain (ΔbfmRS) (Table S1 in Text S1) and performed phenotypic analysis. Interestingly, the ΔbfmRS strain and wild-type MPAO1 strain display similar phenotypes when bacteria are grown in Pyocyanin production broth (PPB) or on a cetyltrimethylammonium bromide (CTAB plate) (Figure 1C). The introduction of a wild-type bfmR gene (p-bfmR, Table S1 in Text S1) into ΔbfmRS strain restored its phenotypes similar to ΔbfmS strain (Figure 1C). These results suggest that the effect of bfmS deletion on either the pigment production or rhamnolipids production in P. aeruginosa is likely mediated through the over-expression of bfmR. Moreover, qRT-PCR analysis indicates that the expressions of at least 12 genes were significantly affected by the deletion of bfmS, whereas their altered expression levels caused by bfmS deletion were suppressed by additional deletion of bfmR. Thus, bfmR may mediate most, if not all, the output of bfmS.
We next tested if BfmS could affect the expression of BfmR. We constructed a bfmR promoter-lux fusion (bfmR-lux, Table S1 in Text S1) and then measured its activity in a wild-type MPAO1 strain, a bfmS deletion strain (ΔbfmS), its complementary strain (ΔbfmS/p-bfmS), a bfmRS double deletion strain (ΔbfmRS), and a ΔbfmRS strain harboring p-bfmR, p-bfmS or p-bfmRS (Table S1 in Text S1). As shown in Figure 1D, the activity of bfmR-lux in ΔbfmS strain was about 60-fold higher than that of the wild-type MPAO1 strain. Complementation with p-bfmS in the ΔbfmS strain restored the activity of bfmR-lux to levels similar to the wild-type strain (Figure 1D). In addition, the activity of bfmR-lux in the ΔbfmRS strain was similar to that observed in the wild-type MPAO1 strain; however, the introduction of p-bfmR, but not p-bfmS or p-bfmRS, to the ΔbfmRS strain dramatically increased the activity of bfmR-lux (>46-fold) (Figure 1D). Hence, BfmR can activate its own gene promoter in the absence of BfmS.
We next evaluated if the absence of BfmS causes an accumulation of BfmR in P. aeruginosa. To this end, we constructed an integration vector mini-ctx-BfmR-Flag (Table S1 in Text S1) and the resulting clone was mobilized into the wild-type MPAO1 and ΔbfmRS strain, yielding an MPAO1::BfmR-Flag strain and ΔbfmRS::BfmR-Flag strain, respectively. The ΔbfmRS::BfmR-Flag strain displayed a pigment-deficient phenotype as observed for either the ΔbfmS strain or the ΔbfmRS/p-bfmR strain (Figure S4A). The cell lysates of MPAO1::BfmR-Flag strain, ΔbfmRS::BfmR-Flag strain and its complementary strain (ΔbfmRS::BfmR-Flag/p-bfmS) were subjected to Western blot analysis using anti-FLAG antibodies. As shown in Figure S4B, a large amount of BfmR-Flag in the ΔbfmRS::BfmR-Flag strain was detected. In contrast, no detectable signal was obtained for the BfmR-Flag generated from the MPAO1::BfmR-Flag strain or the complementary strain (ΔbfmRS::BfmR-Flag/p-bfmS) (Figure S4B). Therefore, the absence of BfmS leads to an accumulation of its cognate response regulator BfmR.
Since BfmR activates its own gene promoter, we next aimed to test if BfmR could bind its own promoter. We performed electrophoretic mobility shift assay (EMSA) using 6His-BfmR protein and DNA fragments containing bfmR, rhlA, and rhlC promoter regions, respectively. We found that 6His-BfmR could shift the bfmR promoter DNA, although it failed to shift the rhlA or rhlC promoter DNA (Figure 2A). We further determined the specific DNA sequence that BfmR can recognize in the bfmR promoter region by using a dye-primer-based DNase I footprint assay. We uncovered three BfmR-protected regions in the bfmR promoter DNA (Figure 2B). Interestingly, all three BfmR-protected regions contain a consensus sequence GATACAnnGC (where n is any nucleotides, Figure 2C).
Using RAST (http://rsat.ulb.ac.be/rsat/), we found that 41 promoters (−1 bp to −400 bp of the coding region) of P. aeruginosa PAO1, including the PA4017 promoter, contain a putative BfmR-binding motif (GATACAnnGC) (Table S5 in Text S1). As expected, BfmR could shift the PA4107 promoter DNA (Figure S5A) in our EMSA analysis, although it failed to shift the rhlI promoter DNA (Figure S5A). Interestingly, we also observed that BfmR is able to bind to PA4103 promoter DNA (Figure S5B) that lacks a canonical BfmR-binding motif (GATACAnnGC) (Table S5 in Text S1). Using a dye-primer-based DNase I footprint assay, we uncovered that the BfmR-protected region of PA4103 promoter DNA contains a non-canonical BfmR-binding motif (GATACAnnAC, the mismatch is underlined) (Figure S5C), which is subsequently determined to be required for the BfmS-mediated regulation of PA4103 promoter activity (Figure S5D). Thus, it is likely that BfmR directly control the expression of the PA4107 and PA4103 operon.
As aforementioned, BfmR negatively controls the rhl QS system of P. aeruginosa, which does not need to bind promoter of rhlI (Figure S5A and S5B). These observations prompted us to determine if BfmR binds to the promoter of rhlR. EMSA analysis indicated that 6His-BfmR bound to the rhlR promoter DNA but not to the promoter region of rhlC that serves as a negative control (Figure 3A). Dye-primer-based DNase I footprint assay indicated that there were three BfmR-protected regions in the promoter of the rhlR (Figure 3B). Interestingly, protected region I (−220 to −193 from the start codon of rhlR) harbored a putative BfmR-binding motif (GATACTnnGC) with one mismatch (underlined) (Figure 3B), oriented in the opposite direction of the transcription of rhlR. Protected region II extends from nucleotide −151 to nucleotide −171 while protected region III extends from nucleotide −69 to nucleotide −85, relative to the start codon of rhlR, respectively (Figure 3B).
There are 44 bfmS-regulated genes harbor a consensus sequence (GATACAnnGC with or without one mismatch) in their promoter region (−1 bp to −400 bp of the coding region) (Tables S2 and S3 in Text S1). Additionally, there are 984 promoters (−1 bp to −400 bp of the coding region) harbor the consensus sequence (GATACAnnGC without or with one mismatch) in the PAO1 genome. These observations suggest that BfmR may serves as a global regulator affecting expression of a large number of genes.
We next elucidated if BfmR regulates the expression of rhlR. To do this, we measured rhlR promoter-lux fusion activity in the wild-type MPAO1 strain, the ΔbfmS strain, the complementary strain (ΔbfmS/p-bfmS), the ΔbfmRS strain, and the ΔbfmRS strain harboring p-bfmR (ΔbfmRS/p-bfmR). The low-phosphate (0.32 mM Pi) M8-glutamate minimal medium supplemented with 0.2% glucose, used as phosphate limitation, served to stimulate the expression of rhlR [21]. As shown in Figure 3C and S3B, the activity of rhlR-lux in the ΔbfmS strain was more than 8-fold lower than that observed in the wild-type MPAO1 strain. Complementation with p-bfmS in the ΔbfmS strain restored the activity of rhlR-lux similar to that of the wild-type strain (Figure 3C). Moreover, ΔbfmRS strain exhibited rhlR-lux activity similar to that observed in the wild-type MPAO1 strain, while the introduction of p-bfmR into the ΔbfmRS strain caused a 3.8-fold decrease in rhlR-lux activity (Figure 3C). Thus, it is likely that bfmS activates the expression of rhlR by repressing BfmR, which acts to negatively regulate rhlR expression and rhl QS. This notion was further substantiated by the observations that under low-phosphate growth conditions ΔbfmS also exhibits decreased rhlI (Figure S3C) and rhlA (Figure S3D) promoter activity, and lowered C4-HSL content (Figure S3E) as compared to wild-type MPAO1.
To determine if the putative BfmR-binding motif (GATACTnnGC) (Figure 3B) is involved in the BfmR-mediated inhibition of rhlR-lux activity, we deleted the first five residues (GATACT) in the consensus sequence (yielding rhlR-D-lux, Table S1 in Text S1), and examined the ability of the mutant sequence to permit the inhibition of the reporter gene in ΔbfmS strain. As shown in Figure 3D, the rhlR-lux activity in ΔbfmS strain was approximately 8-fold lower than that observed in the wild-type MPAO1 strain or in the ΔbfmRS strain. However, the rhlR-D-lux activity was about 2-fold lower in the ΔbfmS strain compared to the wild-type MPAO1 strain or the ΔbfmRS strain (Figure 3D). Thus, the five residues (GATACT) are required for the full inhibition of rhlR-lux activity in ΔbfmS strain, demonstrating the importance of these conserved binding-site residues. However, besides GATACT sequence elements, additional regulatory sequence elements within the promoter region of rhlR are most likely involved in BfmR-mediated inhibition of rhlR-lux activity, given that the rhlR-D-lux activity in ΔbfmS strain is still decreased, although to a much lesser extent than that of rhlR-lux (Figure 3D).
Like many other response regulators [31], BfmR can be phosphorylated by acetyl phosphate (Figure S6A) and hence activated in vitro (Figure S6B). We further observed that the predicted phosphorylation site, aspartate residue D55, is required for the activation of BfmR in vitro and in vivo (Figure S7). As acetyl∼P is an intermediate in the acetate kinase (AckA)-phosphate acetyltransferase (Pta) pathway [31], we hypothesized that the AckA-Pta pathway may be involved in the activation of BfmR. Thus, we constructed a mutant strain (ΔbfmSΔackA-pta, Table S1 in Text S1) with deletion of both the bfmS gene and ackA-pta operon and measured the activity of bfmR-lux as well as the C4-HSL content in this strain and the ΔbfmS strain, respectively. As shown in Figure 4, the expression of bfmR-lux was lower (Figure 4A) while the C4-HSL content was higher (Figure 4B) in the ΔbfmSΔackA-pta strain than that of the ΔbfmS strain, respectively. The introduction of wild-type ackA-pta operon (p-ackA-pta, Table S1 in Text S1) into the ΔbfmSΔackA-pta strain was able to restore either the activity of bfmR-lux (Figure 4A) or the C4-HSL content to the level of the ΔbfmS strain (Figure 4B). Therefore, in the ΔbfmS strain, acetyl phosphate or the component that is dependent on the Acka-Pta pathway is required for the full activity of BfmR.
As BfmS modulates the rhl QS system that contributes significantly to the virulence of P. aeruginosa [4]–[6], [9], [10], we infected romaine lettuce leaves with P. aeruginosa to see if BfmS controls bacterial virulence. The pathogenicity assay revealed a significant difference in the manifestation of infection symptoms caused by the ΔbfmS strain compared to wild-type MPAO1 strain. Relative to wild-type MPAO1, the ΔbfmS strain failed to cause severe necrotic lesions of the leaves, which can be complemented by introducing the wild-type bfmS gene into the ΔbfmS strain (Figure 5A). In addition, the ΔbfmRS strain exhibited a virulence phenotype similar to that of a wild-type MPAO1 strain, while the introduction of p-bfmR into the ΔbfmRS strain led to a low virulence phenotype (Figure S8). Moreover, the constitutive expression of rhlR in ΔbfmS strain could restore either the virulence of ΔbfmS strain to the level of the wild-type MPAO1 strain (Figure 5B), suggesting that the decreased expression of rhlR is likely responsible for the attenuated virulence of ΔbfmS strain in the lettuce leaf model of P. aeruginosa infection.
Since cytotoxicity and invasion of P. aeruginosa are useful traits for this pathogen [32], we further characterized BfmS to check if it regulates the cytotoxicity or the invasion of P. aeruginosa in a murine lung epithelial cell line (MLE-12), a widely used in vitro model for studying host-pathogen interactions [33]–[35]. Using an MTT assay, we found that about 50% MLE-12 cells were killed when challenged with wild-type MPAO1 strain; however, only 5% of MLE-12 cells were killed after inoculation with the ΔbfmS strain (Figure 5C). Using a colony forming unit (CFU) assay, we showed that deletion of bfmS significantly increases (p<0.01) the internalization of P. aeruginosa by approximately 50% (Figure 5D). Further, the invasive and cytotoxic phenotypes of ΔbfmS strain could be completely restored to the wild-type levels by the introduction of p-bfmS (Figure 5C and 5D). Thus, deletion of bfmS causes a loss of cytotoxic capacity while it enhances the invasion of P. aeruginosa MPAO1 to MLE-12 cells.
To further determine the virulence regulated by BfmS, a mouse model of acute pneumonia was used as described in our previous study [26]. C57BL/6J mice were intranasally infected with approximately 5×106 CFU of wild-type MPAO1, bfmS null mutant ΔbfmS, and its complementary strain ΔbfmS/p-bfmS. Figure 5E shows the CFU of bacteria recovered from the lungs compared to the initial inoculum at 18 h post infection, with a geometric mean indicated for each group. In this assay, wild-type MPAO1 was recovered in numbers approximately at 3.13% of the inoculum dose from lungs with a result 5.6-fold higher than that (0.56%) of the ΔbfmS strain. Further, bacteria of complementary strain (ΔbfmS/p-bfmS) were recovered from lungs with 3.43%, similar to that of the wild-type MPAO1 strain (Figure 5E). These results indicate that deletion of bfmS decreases P. aeruginosa survival in the mouse lungs in this model and thus reduced virulence.
The P. aeruginosa DK2 lineage is highly successful and has been isolated from ∼40 cystic fibrosis (CF) patients since the start of the sampling program in 1973 [36]. We noted that the DK2 lineage-specific mutations in BfmS are point mutations that cause two amino acid substitutions, proline replaces leucine 181 (L181P), and glutamine replaces glutamic acid (E376Q). Among them, L181P was fixed in the DK2 lineage before the year 1979, while E376Q was subsequently fixed in the DK2 lineage after 1991 [36]. We next investigated the regulatory effect associated with these amino acid substitutions observed in the BfmS. We created p-bfmSL181P, p-bfmSE376Q, and p-bfmSL181P/E376Q plasmids (Table S1 in Text S1) and introduced them to the ΔbfmS strain, and tested their effects on bfmR-lux activity. Interestingly, L181P and E376Q substitutions in BfmS caused a 16-fold and a 2-fold increase in the activity of bfmR-lux, respectively (Figure 6A). More significantly, the combined substitution (L181P/E376Q) led to a 27-fold increase in the activity of bfmR-lux (Figure 6A). Accordingly, L181P or L181P/E376Q substitutions in BfmS decreased the RhlI-dependent autoinducer C4-HSL content (Figure 6B). These data strongly suggest that mutations in specific residues of BfmS result in activation of BfmR.
The DK2 lineage-specific mutations (L181P, L181P/E376Q) may abolish the negative regulatory effects of BfmS on BfmR or alternatively, it may transform BfmS into a positive regulator of BfmR and therefore, activate BfmR (Figure 6A and 6B). To discriminate between these two possibilities, we tested the effect of these amino acid substitutions in BfmS on the activity of bfmR-lux when bacteria were grown in M8-glutamate minimal medium supplemented with sodium acetate. BfmS did not function as a negative regulator of BfmR when bacteria were grown in this media, given that the ΔbfmS strain exhibited similar bfmR-lux activity as the MPAO1 strain (Figure 6C). However, the ΔbfmS/p-bfmSL181P strain and ΔbfmS/p-bfmSL181P/E376Q strain displayed bfmR-lux activity 2.5-fold and 6-fold higher than that the activity observed in the reference strain (ΔbfmS/p-bfmS), respectively (Figure 6C). These results suggest that the L181P or L181P/E376Q amino acid substitution render BfmS as a positive regulator of BfmR. Further, the introduction of p-bfmSL181P or p-bfmSL181P/E376Q to the bfmRS double deletion mutant strain (ΔbfmRS) failed to increase the activity of bfmR-lux (Figure S9A), suggesting that the effect of L181P or L181P/E376Q substitutions in BfmS on the bfmR-lux activity is mediated through the activation of BfmR.
BfmS is a member of the HisKA subfamily of bacterial histidine kinases and it is predicted that the conserved H238 residue is required for its kinase activity [37]. We next investigated the regulatory role associated with this residue by changing His to Ala. We created p-bfmSH238A and p-bfmSL181P/E376Q/H238A plasmids (Table S1 in Text S1) and introduced them to the ΔbfmS strain, and examined their effects on the bfmR-lux activity. As shown in Figure 6D, H238A substitution in BfmS has no significant effect on the activity of bfmR-lux when bacteria were grown in M8-glutamate minimal medium supplemented with 0.2% glucose. However, H238A substitutions in the BfmSL181P/E376Q abolished its ability to induce the bfmR-lux activity (Figure 6D), which suggests that the H238 residue or the kinase activity is required for BfmSL181P/E376Q to activate BfmR. This hypothesis is further supported by the fact that amino acid substitution L181P/E376Q, but not L181P/E376Q/H238A, causes overproduction of phosphorylated and unphosphorylated BfmR (Figure 6E).
Furthermore, we noted that the RP73-specific mutation in bfmS causes the substitution of arginine by histidine at the codon 393 (R393H) [38]. The P. aeruginosa strain RP73 was isolated after 16.9 years of chronic lung infection in a CF patient [38]. Like the L181P substitution or the L181P/E376Q substitution, the R393H substitution in BfmS resulted in increased bfmR-lux activity (Figure S9B and S9C) and decreased level of C4-HSL (Figure S9D), suggesting that BfmSR393H activates BfmR. As the regulatory effects of these BfmS variants (BfmSL181P, BfmSL181P/E376Q, and BfmSR393H) on BfmR have been changed from negative to positive, we therefore term them “reverse function” mutants. Although we failed to obtain either the soluble full-length BfmS protein or the cytoplasmic region of BfmS that prevented us from the assays of the phosphatase/kinase activities of BfmS in vitro, our genetic analyses clearly indicate that specific missense mutations (L181P, L181P/E376Q, or R393H) can convert BfmS from a repressor to an activator of BfmR.
In this study, we uncovered a novel signal transduction pathway, BfmS/BfmR/RhlR, for the regulation of the rhl QS system in P. aeruginosa. We demonstrated that BfmS has profound effects on the expression of virulence-associated traits and the ability of P. aeruginosa to adapt to the host. In addition, we found that deletion of bfmS leads to a dramatic increase in biofilm formation and that BfmR mediates this effect (Figure S10). This appearance is consistent with previous observations that BfmR is a biofilm maturation regulator [24], [25]. Intriguingly, BfmS is able to switch its function from a repressor to an activator of BfmR in cystic fibrosis (CF) isolates such as DK2 strains and RP73 strain. A proposed model for signal transduction by BfmRS TCS is shown in Figure 7.
The prototypical two-component regulatory system is composed of a sensor kinase and a response regulator. In general, the sensor kinase senses an environment change and communicates it via phosphorylation to its cognate response regulator, and hence activates the response regulator's function. We demonstrated that BfmS functions as a negative regulator of BfmR (Figure 1D and Figure S4), whose activation requires phosphorylation on D55 (Figure S7). Therefore, BfmS might act as a phosphatase as opposed to a kinase for BfmR under our experimental conditions. In fact, many two-component sensors are bifunctional, catalyzing both the phosphorylation and dephosphorylation of their cognate response regulator [37], [39]–[41]. For some sensor kinases, the phosphatase activity may be the critical function in vivo [42]–[45]. Interestingly, the BfmS homologue in Pseudomonas syringae, RhpS, has also been shown to be a negative regulator of its cognate regulator RhpR in our previous studies [44], [45].
In the absence of BfmS, the Acka-Pta pathway can modulate the activity of BfmR (Figure 4). These observations suggest that the activation of BfmR is shaped by BfmS as well as by the nutritional status of P. aeruginosa. However, the expression of bfmR-lux in the ΔbfmSΔackA-pta strain was still about 12-fold higher than that in the wild-type MPAO1 strain (Figure 4A), indicating that acetyl phosphate or the component that is dependent on the Acka-Pta pathway is likely not the sole trigger of BfmR activation. This notion was further substantiated by the observation that the deletion of ackA-pta operon only partially alleviates the inhibition of QS signal C4-HSL production caused by the absence of bfmS (Figure 4B).
The rhlR gene encodes the transcriptional regulator RhlR, which has a central role in the quorum-sensing response, and is therefore very important for P. aeruginosa to co-ordinate its virulence in order to establish a successful infection [4], [6], [8]–[10], [46], [47]. We found that BfmR binds to and represses the rhlR promoter (Figure 3). A DNase I footprint analysis demonstrated that the BfmR-protected region (binding site I) of the rhlR promoter has a putative BfmR-binding motif (GATACTnnGC) that is crucial to the BfmR-mediated inhibition of rhlR-lux activity (Figure 3D), thereby reinforcing the likelihood that BfmR directly regulates rhlR. However, the detailed effect of BfmR on the rhlR gene expression awaits further study, since the rhlR promoter harbors multiple transcription start sites [15], regulatory sequences [15], and at least three BfmR-protected regions (Figure 3B). To our knowledge, this is the first evidence of a two-component regulator regulating rhlR in a direct manner. However, this finding is in contrast to a previous report suggesting that BfmR functions independently of QS signaling [24]. The exact cause of this discrepancy remains unknown. In the previous report [24], Petrova et al drew the conclusion based on the fact that deletion of bfmR has no significant effect on the transcript abundance of rhlA and lasB. Consistent with this, we observed that deletion of bfmRS has no significant effect on rhl-dependent phenotypes (Figure 1A and 1B) and the expression of rhlA (Table S4 in Text S1). However, deletion of bfmS alone leads to the activation of BfmR (Figure 1D and S4B), which in turn directly binds to the promoter and decreases the expression of the rhlR (Figure 3), causing the inhibition of the rhl QS system (Figure 1 and S3). Additionally, the decreased expression of rhlR in ΔbfmS strain (Figure 3C and 3D) may contribute to the attenuated virulence in lettuce leaves (Figure 5A) and the reduced production of QS signal C4-HSL, pyocyanin and rhamnolipids (Figure 1 and S1), since the constitutive expression of rhlR in ΔbfmS strain could restore these phenotypes to wild-type levels or higher (Figure 5B and S11). Moreover, the expression of rhlI in the ΔbfmS strain was significantly lower (>2-fold) than that of the wild-type strain (Tables S3 and S4 in Text S1). Thus, the bfmS deletion, which results in activation of BfmR, affects all aspects of the rhl QS system.
Deletion of bfmS impacts the expression of 202 genes that comprise 3.6% of the P. aeruginosa genome (Tables S2 and S3 in Text S1). These observations suggest that BfmS acts as a global regulator in P. aeruginosa. Besides regulating rhl quorum sensing, BfmS also regulates the expression of a large number of genes such as PA4103, PA4104, PA4105, PA4106 and PA4107, whose transcripts are likely independent of the quorum-sensing regulated [48]. Thus, it is not surprising that BfmS has a profound effect on the expression of virulence-associated traits and the ability of P. aeruginosa to adapt to the host (Figure 5). Interestingly, BfmS has a positive impact on acute virulence phenotypes (Figure 5A and 5E), but a negative effect on biofilm formation (Figure S10) that acts as a major virulence-associated trait contributing to chronic infections [49]. This formation suggests that BfmS may play an important role in mediating the switch between the acute and chronic infection lifestyles of P. aeruginosa.
P. aeruginosa can cause serious acute and chronic infections in humans and it underwent numerous genetic adaptations during evolution in the CF airways, resulting in remodeling of the regulatory networks to match the fluctuations in the environment of CF lung [36], [50]–[52]. bfmS in P. aeruginosa CF isolates are often found to undergo missense mutations. For instance, L181P (point mutations causing the substitution of leucine by proline at codon 181) or L181P/E376Q in at least 10 DK2 strains [36], R393H in RP73 strain [38], A21P/T120K/L164F in either LESB58 or LES431strain, L164F in c7447m strain, A4T in CIG1 strain, T120K/L163V/L164F in C3719 strain, D295N in CF5 strain, and P6S/L164F in CF614 strain (http://www.ncbi.nlm.nih.gov/). We found that specific missense mutations in bfmS gene (L181P, L181P/E376Q, and R393H) result in elevated BfmR activity (Figure 6 and Figure S9), which contributes to biofilm formation (Figure S10) [24], [25] as well as to the inhibition of the rhl QS system (Figures 1, 3, and S3). It is well known that biofilm formation enables P. aeruginosa to cause persistent infections [49] while the loss of quorum sensing is one of the dominating changes that occur during the adaptive process of the P. aeruginosa in the CF lung [36], [50]. Thus, the naturally occurring missense mutations in BfmS may provide a selective advantage to either DK2 strains or RP73 strain during the course of chronic infection in CF lungs. However, it should be noted that the P. aeruginosa community in the CF lung is very dynamic [50], [51], [53] and only a fraction of the isolates will most probably possess these mutations. Therefore, the role of these missense mutations in the chronic lung infection awaits further investigation.
Intriguingly, although BfmS functions as a negative regulator of BfmR (Figure 1D and S4B), the naturally occurring missense mutations in bfmS gene (L181P, L181P/E376Q, and R393H) can produce BfmS variants that no longer repress, but instead activate BfmR (Figure 6 and Figure S9). These “reverse function” mutants of BfmS may exhibit an elevated ratio of kinase to phosphatase activity on BfmR, given that the activation of BfmR requires the phosphorylation on D55 (Figure S7). In agreement with this notion, we found that H238, the conserved histidine residues predicted to be involved in the autophosphorylation of BfmS, is required for BfmSL181P/E376Q to activate BfmR (Figure 6D and 6E). The occurrence of BfmS “reverse function” mutants is not strain dependent, as evidenced by the fact that either DK2 lineage-specific mutations or RP73-specific mutation in bfmS could reverse its function against BfmR. The L181P mutation was located in the HAMP domain of BfmS, while the E376Q and the R393H mutation were located in ATP-binding domain. Currently, it is not clear how these missense mutations change the function of BfmS. Although further studies are needed to elucidate this elegant mechanism, our genetic analyses clearly indicated that naturally occurring missense mutations in P. aeruginosa gene could result in reverse of function, rather than simply loss (weakened) or gain (strengthened) of function.
Noticeably, bfmRS operon and BfmR-activated transcripts such as PA4103-4104 and PA4107-4106-4105 (Figure S2, Tables S2 and S4 in Text S1), were dramatically up-regulated in the lungs of cystic fibrosis patients compared to in vitro planktonic bacteria, indicating that BfmRS system is likely activated during chronic infection in CF lungs [23]. These genes also exhibit much higher gene expression levels in some P. aeruginosa CF isolates such as DK2-lineage strains (late stage infection isolates) [54] and E601 strain [55] compared to wild-type laboratory strain PAO1. These observations and results from the current study suggest that BfmRS TCS may sense and respond to environmental stress in CF lungs. We envision that further studies aimed at the characterization of the stimuli that BfmRS and/or its variants detect within the host could be of great importance to a full understanding of the mechanisms that make P. aeruginosa a successful pathogen and to the development of novel strategies to limit its infections.
Animal experiments were performed in strict accordance with the Regulations for the Administration of Affairs Concerning Experimental Animals approved by the State Council of People's Republic of China (11-14-1988). All animal procedures were approved by the Institutional Animal Care and Use Committee (IACUC) of Shanghai Public Health Clinical Center (Permit Number: 2013P201). The laboratory animal usage license number is SYXK-HU-2010-0098, certificated by Shanghai Committee of Science and Technology.
The bacterial strains and plasmids used in this study are listed in Table S1 in Text S1. Unless noted otherwise, P. aeruginosa MPAO1 [56] and its derivatives were grown in Luria-Bertani (LB) broth, Pyocyanin production broth [57] (PPB: 20 g peptone, 1.4 g MgCl2, 10 g K2SO4, 20 ml glycerol per liter; pH 7.0), or M8-glutamate minimal medium [27] (6 g Na2HPO4, 3 g KH2PO4, 0.5 g NaCl, 0.24 g MgSO4, 0.5 g glutamate per liter; pH 7.4) supplemented with 0.2% glucose, as indicated. E. coli cultures were grown in Luria-Bertani (LB) broth. All cultures were incubated at 37°C with shaking (250 rpm). For plasmid maintenance in P. aeruginosa, the medium was supplemented with 100 µg/ml carbenicillin or 100 µg/ml kanamycin when required. For plasmid maintenance in E. coli, the medium was supplemented with 100 µg/ml carbenicillin, 50 µg/ml kanamycin, 300 µg/ml trimethoprim, or 10 µg/ml gentamicin, as appropriate. For marker selection in P. aeruginosa, either 30 µg/ml gentamicin or 10 µg/ml tetracycline were used when required.
For gene replacement, a SacB-based strategy [58] was employed as described in our previous study [26]. To construct the bfmS null mutant (ΔbfmS), polymerase chain reactions (PCRs) were performed in order to amplify sequences upstream (1,574 bp) and downstream (1,562 bp) of the intended deletion. The upstream fragment was amplified from MPAO1 genomic DNA using primers BfmSupF (with EcoRI site) and BfmSupR (with XbaI site), while the downstream fragment was amplified with primers, BfmSdownF (with XbaI site) and BfmSdownR (with HindIII site). The two PCR products were digested and then cloned into the EcoRI/HindIII-digested gene replacement vector pEX18Ap, yielding pEX18Ap::bfmSUD. A 1.8 kb gentamicin resistance cassette was cut from pPS858 with XbaI and then cloned into pEX18Ap::bfmSUD, yielding pEX18Ap::bfmSUGD. The resultant plasmid, pEX18Ap::bfmSUGD, was electroporated into MPAO1 with selection for gentamicin resistance. Colonies were screened for gentamicin sensitivity and loss of sucrose (5%) sensitivity, which typically indicates a double-cross-over event and thus marks the occurrence of gene replacement. The ΔbfmS strain was further confirmed by PCR.
A similar strategy was used to construct the ΔbfmRS strain as described above. Briefly, the upstream fragment (1,832 bp) of the intended deletion was amplified with primers BfmRupF (with EcoRI site) and BfmRupR (with XbaI site). The downstream fragment (1,562 bp) was amplified with primers, BfmSdownF (with XbaI site) and BfmSdownR (with HindIII site). A 1.8 kb gentamicin resistance cassette was cut from pPS858 with XbaI and then cloned into pEX18Ap::bfmRSUD, yielding pEX18Ap::bfmRSUGD.
Again, a similar strategy was used to construct the ΔbfmSΔackA-pta strain. Primers Acka-up-F (with KpnI site) and Acka-up-R (with BamHI site) amplified the upstream fragment (2,245 bp) of the intended deletion of ackA-pta operon in ΔbfmS. Primers Pta-domn-F (with BamHI site) and Pta-domn-R (with HindIII site) amplified the downstream fragment (1,709 bp). A 2.3 kb tetracycline resistance cassette was amplified from the integration vector mini-CTX-lacZ with primers, Mini-TC-F (with BamHI site) and Mini-TC-F (with BamHI site). The resultant plasmid, pEX18Ap::acka-ptaUTD, was electroporated into ΔbfmS strain with selection for tetracycline resistance. Colonies were screened for tetracycline sensitivity and loss of sucrose (5%) sensitivity, which typically indicate a double-cross-over event and thus mark the occurrence of gene replacement. PCR further confirmed the deletion of pta-acka loci.
Primers bfmRflag-F (with HindIII site) and bfmRflag-R (with BamHI site) (Table S6 in Text S1) were used to perform PCR of the BfmR gene that was meant to fuse with the Flag-tag. A 1,586-bp PCR product covering the region from 848 bp upstream and the BfmR gene (not including the stop codon) was generated. The HindIII- and BamHI-digested PCR product was cloned into the HindIII and BamHI sites of the mini-CTX-lacZ [59] to generate mini-ctx-BfmR-Flag. The resulting plasmid was conjugated into P. aeruginosa MPAO1 and ΔbfmRS strains and the construct was integrated into the attB site as described previously though a diparental mating using E. coli S17 λ-pir as the donor, yielding a MPAO1::BfmR-Flag strain and a ΔbfmRS::BfmR-Flag strain, respectively (Table S1 in Text S1). In these mutant strains, parts of the mini-CTX-lacZ vector containing the tetracycline resistance cassette were deleted using a flippase (FLP) recombinase encoded on the pFLP2 plasmid.
To construct the plasmid for constitutive expression of bfmR, a 806 bp PCR product covering 15 bp of the bfmR upstream region, the bfmR gene, and 50 bp downstream of bfmR was amplified using primers BfmR(comp)Fwr (with HindIII site) and BfmR(comp)Rev (with BamHI site). The product was digested with HindIII and BamHI and ligated into PAK1900 [60] in the same orientation as plac to generate p-bfmR.
To construct the plasmid for the constitutive expression of bfmS, a 1,385 bp PCR product covering 30 bp of the bfmS upstream region, the bfmS gene, and 50 bp downstream of bfmS was amplified using primers BfmS(comp)Fwr (with HindIII site) and BfmS(comp)Rev (with BamHI site), and then cloned into PAK1900, yielding p-bfmS.
To construct the plasmid for the constitutive expression of bfmRS, a 2,107 bp PCR product covering 15 bp of the bfmR upstream region, the bfmRS operon, and 50 bp downstream of bfmS was amplified using primers BfmR(comp)Fwr (with HindIII site) and BfmS(comp)Rev (with BamHI site) and then cloned into PAK1900, yielding p-bfmRS.
To construct the plasmid for the constitutive expression of rhlR, a 770 bp DNA fragment covering 44 bp of the rhlR upstream region and the rhlR was amplified using primers RhlR-OE-F (with HindIII site) and RhlR-OE-R (with HindIII site) and then cloned into PAK1900. The construct with rhlR in the same orientation as plac was confirmed by DNA sequencing, yielding p-rhlR.
To construct the plasmid for constitutive expression of acka-pta operon, a 3,563 bp DNA fragment covering 136 bp of acka upstream region, the acka-pta operon, and a 65 bp downstream of pta was amplified using primers Acka-comp-F (with HindIII site) and Pta-comp-R (with BamHI site) and then cloned into PAK1900, yielding plasmid p-ackA-pta.
The five mutations, p-bfmRD55A, p-bfmSL181P, p-bfmSE376Q, and p-bfmSL181P/E376Q, and p-bfmSR393H were obtained using the QuikChange II site-directed mutagenesis kit (Stratagene). For generating p-bfmRD55A, the primer pair BfmR(D55A)-F/BfmR(D55A)-F was used. For generating p-bfmSL181P, primer pair PA4102L181P-F/PA4102L181P-R was used. For generating p-bfmSE376Q, the primer pair PA4102E376Q-F/PA4102E376Q-R was used. For generating p-bfmSL181P/E376Q, primer pairs PA4102L181P-F/PA4102L181P-R and PA4102E376Q-F/PA4102E376Q-R were used. For generating p-bfmSR393H, the primer pairs R393H-F/R393H-R was used.
All constructs were sequenced to ensure that no unwanted mutations resulted.
Full-length of bfmR was cloned into pET28a with a thrombin-cleavable N-terminal His-tag. Primers bfmR-F (with NdeI site) and bfmR-R (XhoI) were used to amplify the bfmR gene from P. aeruginosa MPAO1 chromosomal DNA. The amplified fragments were ligated into similarly cut pET28a (Novagen) in order to produce the plasmids pET28a-6His-BfmR. pET28a-6His-BfmRD55A was obtained by using the primer pair BfmR(D55A)-F/BfmR(D55A)-R and a QuikChange II site-directed mutagenesis kit (Stratagene). The protein was expressed in the E. coli strain BL21 star (DE3) and purifications were performed as described in our previous studies [26], [61], [62]. Briefly, bacteria were grown at 37°C overnight in 10 ml of LB medium (containing 50 µg/ml kanamycin) with shaking (250 rpm). The next day, the cultures were transferred into 1 L of LB medium (containing 50 µg/ml kanamycin) incubated at 37°C with shaking (250 rpm) until the OD600 reached 0.6, and then IPTG (isopropyl-1-thio-β-d-galactopyranoside) was added to a final concentration of 1.0 mM. After 4 h incubation at 30°C with shaking (250 rpm), the cells were harvested by centrifugation and stored at −80°C. The cells were lysed at 4°C by sonication in lysis buffer [10 mM Tris (pH 7.4), 300 mM NaCl, 1 mM PMSF, and 2 mM DTT]. Clarified cell lysate was loaded onto a HisTrap HP column (Amersham Biosciences), washed with Ni-NTA washing buffer, and eluted with Ni-NTA elution buffer. The fractions containing 6His-BfmR or 6His-BfmRD55A were concentrated and loaded onto a Superdex-200 gel filtration column with a running condition of 10 mM Tris (pH 7.4), 300 mM NaCl, and 2 mM DTT. The purified protein was >90% pure as estimated by a 12% (wt/vol) SDS/PAGE gel.
The DNA sequence of the extracellular sensory domain of BfmS consisting of 121 residues (Gln34-Trp154) was amplified from MPAO1 genomic DNA with the primers PA4102-EX-F (with NcoI) and PA4102-EX-R (with BamHI) by PCR and was subsequently cloned into pET28b using NcoI and BamHI as the restriction enzymes. Following confirmation by DNA sequencing, the recombinant plasmid (pET28b-bfmS34–154) was transformed into E. coli strain BL21 star (DE3). The extracellular sensory domain of BfmS (designated BfmS34–154) was expressed and purified as described above with some modifications. Briefly, bacteria were grown at 37°C overnight in 10 ml of LB medium (containing 50 µg/ml kanamycin) with shaking (250 rpm). The next day, the cultures were transferred into 1 L of LB medium (containing 50 µg/ml kanamycin) incubated at 37°C with shaking (200 rpm) until the OD600 reached 0.6, and then IPTG was added to a final concentration of 1.0 mM. After 16 h incubation at 16°C with shaking (200 rpm), the cells were harvested by centrifugation and stored at −80°C. The cells were lysed at 4°C by sonication in lysis buffer [50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10% glycerol 1 mM PMSF, and 2 mM DTT]. Clarified cell lysate was loaded onto a HisTrap HP column (Amersham Biosciences), and eluted with Ni-NTA elution buffer (50 mM Tris-HCl, pH 8.0, 100 mM NaCl, 20 mM imidazole 10% glycerol, and 1 mM DTT,). The fractions containing BfmS34–154 were concentrated and loaded onto a Superdex-75 gel filtration column with a running condition of 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 10% glycerol, 1 mM DTT. The purified protein was >90% pure as estimated by a 12% (wt/vol) SDS/PAGE gel.
The plasmid pMS402 [63] carrying a promoterless luxCDABE reporter gene cluster was used to construct promoter-luxCDABE reporter fusions of the bfmR as described previously [28], [64]. For bfmR-lux, the bfmR promoter region (−463 to +18 of the start codon) was amplified by PCR using the primers PMS402-bfmR-F (with XhoI site) and PMS402-bfmR-R (with BamHI site). For rhlA-lux, the rhlA promoter region (−526 to −20 of the start codon) was amplified by PCR using the primers pms402-rhlA-F (with XhoI site) and pms402-rhlA-R (with BamHI site). For rhlR-lux, the rhlR promoter region (−450 to +19 of the start codon) was amplified by PCR using the primers pms402-rhlR-1F (with BamHI site) and pms402-rhlR-R (with BamHI site). The promoter oriented in the same direction as luxCDABE was selected for further analysis. To generate rhlR promoter mutant rhlR-D (deletion of GATACT, which is the central part of the putative BfmR-binding site on the reverse DNA strand), the DNA fragment was amplified using primers pms402-rhlR-1F/pms402-rhlR-R and subsequently cloned into pGEM-T vector. rhlR-D (rhlR promoter lacking putative BfmR-binding site) was obtained using a QuikChange II site-directed mutagenesis kit (Stratagene) and primer pair pms402-rhlR(D1)F/pms402-rhlR(D1)R. For 4103-lux, the PA4103 promoter region (−659 to +19 of the start codon) was amplified by PCR using primers pms402-p4103-F (with XhoI site) and pms402-p4103-R (with BamHI site). To generate PA4103 promoter mutant 4103-M (GATACA was mutated to ATATAT), primer pair p4103-mutation-F/p4103-mutation-R was used as described above. The promoter regions were cloned into the XhoI-BamHI site or BamHI site (for rhlR-lux) upstream of the lux genes on pMS402 and the cloned promoter sequences were confirmed by DNA sequencing. The constructs were transformed into MPAO1 or its derivatives by electroporation. Use of these lux-based reporters, gene expression under different conditions was measured as counts per second (cps) of light production with a 2104 EnVision Multilabel Plate Readers or Synergy 2 (Biotek). Relative light units were calculated by normalizing CPS to OD600.
The bacterial growth and the extraction of total RNAs were performed as described above. The total DNase-treated RNA (5 µg) was reversely transcribed to synthesize cDNA using the PrimeScript RT reagent Kit (Takara) with random primers according to the manufacturer's protocol. The resulting cDNA were diluted by 1∶2, 1∶4, and 1∶8 respectively. Triplicate quantitative assays were performed on 1 µl of each cDNA dilution with the THUNDERBIRD SYBR qPCR Mix and 300 nM primers using an Applied Biosystems 7500 Fast Real-Time PCR System. Dissociation curve analysis was performed in order to verify product homogeneity. The gene-specific primers used for Quantitative real-time PCR for PA4100, PA4103, PA4107, PA4108, ntrB, oprH, phoB, hmgA, rhlA, antA, nasA, and rhlI are listed in Table S6 in Text S1. The amplicon of 16S rRNA was used as an internal control in order to normalize all data. Relative expression levels of interest genes were calculated by the relative quantification method (ΔΔCT) as previously described [65], [66].
P. aeruginosa was grown at 37°C for 24 h on M8-glutamate minimal agar plate (M8-glutamate minimal medium supplemented with 0.2% glucose, and solidified with 2% agar). To prepare cell lysates for the Phos-tag gel assay, bacteria cells were scraped from the plate and immediately resuspended in 60 µl of lysis buffer [50 mM Tris-Cl (pH 7.5), 150 mM NaCl, 1 mM MgCl2, 0.1% Triton X-100, 15 µg/ml DNaseI, 0.5 mM PMSF, 1 mM DTT) with 0.1% (vol/vol) Lysonase. Sufficient lysis was achieved by repeated pipetting up and down for 10 s followed by addition of 20 µl of 4×SDS loading buffer. Resulting cell lysates (10 µl) were immediately loaded onto a Phos-tag gel for electrophoresis.
BfmR-flag and BfmR-flag∼P were separated on 10% acrylamide gels containing 25 µM acrylamide-Phos-tag ligand (Wako Pure Chemical) and 50 µM MnCl2 as previously described [67]. Electrophoresis was performed at 30 mA at 4°C for 80 min in Tris-Glycine-SDS running buffer (25 mM Tris, 192 mM glycine, 0.1% SDS, pH 8.4). After electrophoresis, the Phos-tag gel was washed 10 min at RT with Transfer Buffer [20%(v/v) methanol, 50 mM Tris, 40 mM glycine] supplied with 1 mM EDTA to remove Zn2+ from the gel, then the gel was incubated at room temperature with gentle shaking for another 10 min in Transfer Buffer twice to remove EDTA.
Samples resolved on gels were transferred to PVDF (Bio-Rad) membranes through semi-dry transfer assembly (Bio-Rad) for 30 min at room temperature. BfmR-Flag proteins were detected by Western blot analysis using a mouse anti-Flag monoclonal antibody (Cat#: AGM12165, Aogma) followed by a secondary, sheep anti-mouse IgG antibody conjugated to horseradish peroxidase (HRP) (Code#: NA931, GE Healthcare). For detection of ClpP protein, anti-ClpP polyclonal antibody and anti-rabbit IgG antibody conjugated to horseradish peroxidase (HRP) (Code#: NA934, GE Healthcare) were used. Anti-ClpP polyclonal antibody, which cross-reacts with the ClpP of Pseudomonas aeruginosa, was prepared by immunizing a rabbit with a Staphylococcus aureus full-length ClpP protein (NWMN_0736). For detection of BfmS protein, anti-BfmS polyclonal antibody (prepared by immunizing a rabbit with a BfmS34–154 protein, Shanghai Immune Biotech CO., Ltd) and anti-rabbit IgG antibody conjugated to horseradish peroxidase (HRP) (Code#: NA934, GE Healthcare) were used. For detection of RNAP protein, anti-RNAP (Neoclone, #WP003) antibody and anti-mouse IgG antibody conjugated to horseradish peroxidase (HRP) (Code#: NA931, GE Healthcare). Immunoblots for ClpP and RNAP served as loading control. The membrane is exposed to X-ray film (Kodak) or the chemiluminescent is detected by a Imaging Quant LAS-4000 (GE), according to the manufacturer's recommendation.
Pyocyanin was extracted from culture supernatants and measured using previously reported methods [68]. Briefly, P. aeruginosa was grown in Pyocyanin production broth [57] (PPB: 20 g peptone, 1.4 g MgCl2, 10 g K2SO4, 20 ml glycerol per liter; pH 7.0) for 36 h at 37°C with shaking (250 rpm). The culture was subsequently centrifuged and filtered (pore size, 0.22 µm). 1.5 ml of chloroform was added to 2.5 ml of culture supernatant. After extraction, the chloroform layer was transferred to a fresh tube and mixed with 1 ml of 0.2 N HCl. After centrifugation, the top layer (0.2 N HCl) was removed and its absorption measured at 520 nm. Concentrations, expressed as micrograms of pyocyanin produced per ml of culture supernatant, were determined by multiplying the optical density at 520 nm (OD520) by 17.072.
Rhamnolipids production was estimated by inoculating strains on M8-based agar plates supplemented with 0.2% glucose(m/v), 2 mM MgSO4, 0.0005% (m/v) methylene blue, and 0.02% (m/v) cetyltrimethylammonium bromide, as described previously [68], [69]. The orcinol assay was used to directly assess the amount of rhamnolipids in the sample as previously described [28]. After a culture of 48 h in LB medium at 37°C with shaking (250 rpm), 1 ml of the culture supernatant was extracted twice with 2 ml of diethyl ether. The pooled ether fractions were evaporated to dryness and the remainder was dissolved in 100 µl of distilled water and mixed with 100 µl of 1.6% orcinol, and 800 µl of 60% sulfuric acid. After heating for 30 min at 80°C in the dark, the samples were cooled for 3 h at room temperature in the dark. Absorbance at 421 nm (A421) was measured. Rhamnolipid concentrations were calculated by comparing A421 values with those obtained for rhamnose standards between 0 and 1000 µg/ml, assuming that 1 µg of rhamnose corresponds to 2.5 µg of rhamnolipids.
The motility assay was carried out as described previously [27], [68]. Swarming medium was based on M8-glutamate minimal medium [27] (6 g Na2HPO4, 3 g KH2PO4, 0.5 g NaCl, 0.24 g MgSO4, 0.5 g glutamate per liter; pH 7.4), supplemented with MgSO4 (2 mM), glucose (0.2%), and Casamino acid (0.5%), and solidified with 0.5% agar. Bacteria were inoculated with a toothpick onto swarm agar plates. Swarm agar plates were incubated for 24 hours at 37°C and then incubated for more time at room temperature.
Phosphorylation of 6His-BfmR was detected by utilizing the Pro-Q Diamond phosphorylation gel stain as described by the manufacturer (Invitrogen). Purified 6His-BfmR and 6His-BfmRD55A were incubated with buffer (10 mM Tris pH 8.0; 1 mM DTT; 5 mM MgCl2; 10 mM KCl; 50 mM acetyl phosphate) at 37°C for 30 min. The acetyl phosphate-treated samples of 6His-BfmR and 6His-BfmRD55A were resolved on a 12% SDS polyacrylamide gel, and then the gel was immersed in fixing solution (10% acetic acid, 50% methanol) for 30 min and subsequently washed three times with deionized water each for 10 min. The gel was stained with Pro-Q Diamond phosphoprotein gel stain for 60 min, followed by washing with deionized water for 30 min. The entire procedure was conducted at room temperature. Fluorescent output was recorded using an Tanon-5200 multi.
The electrophoretic mobility shift experiments were performed as described in our previous studies with some modifications [26], [61], [62]. Briefly, 20 µl of the DNA probe mixture (30 to 50 ng) and various amounts of purified proteins in binding buffer (10 mM Tris-Cl, pH 8.0; 1 mM DTT; 10% glycerol; 5 mM MgCl2; 10 mM KCl) were incubated for 30 min at 37°C. When indicated, 50 mM acetyl phosphate was added to the solution. Native polyacrylamide gel (6%) was run in 0.5× TBE buffer at 85 V at 4°C. The gel was stained with GelRed nucleic acid staining solution (Biotium) for 10 min, and then the DNA bands were visualized by gel exposure to 260-nm UV light.
DNA probes were PCR-amplified from P. aeruginosa MPAO1 genomic DNA using the primers listed in Table S6 in Text S1. The probes for bfmR promoter, a 481 bp DNA fragment covering the promoter region of bfmR (from −463 to +18 of the start codon) was amplified using primers bfmR-F(EMSA) and bfmR-R(EMSA). For rhlR promoter, a 470 bp DNA fragment covering the promoter region of rhlR (from −450 to +20 of the start codon) was amplified using primers rhlR-F(EMSA) and rhlR-F(EMSA). For rhlI promoter, a 446 bp DNA fragment covering the promoter region of rhlI (from −444 to +2 of the start codon) was amplified using primers rhlI-F(EMSA) and rhlI-R(EMSA). For rhlA promoter, a 572 bp DNA fragment covering the promoter region of rhlA (from −591 to −19 of the start codon) was amplified using primers rhlA-F(EMSA) and rhlA-R(EMSA). For rhlC promoter, a 540 bp DNA fragment covering the promoter region of rhlC (from −549 to −9 of the start codon) was amplified using primers rhlC-F(EMSA) and rhlC-F(EMSA). For PA4103 promoter, a ca. 0.7 kb DNA fragment (4103-P) containing the promoter region of PA4103 (from −659 to +19 of the start codon) was amplified from plasmid 4103-lux DNA using primers pZE.05 and pZE.06. For PA4107 promoter, a 360 bp DNA fragment (4107-P) covering the promoter region of PA4107 (from −490 to −131 of the start codon) was amplified from P. aeruginosa MPAO1 genomic DNA using primers PA4108-F and PA4108-R. All PCR products were purified by using a QIAquick gel purification kit (QIAGEN).
The published DNase I footprint protocol was modified [70] in this study in the same way as described in our previous study [61]. Briefly, PCR was used to generate DNA fragments using the primer sets as detailed in Table S6 in Text S1. For amplification of bfmR promoter, primers bfmR-F(EMSA) and 6FAM-bfmR-R were used. For amplification of the rhlR promoter, primers rhlR-F(EMSA) and 6FAM-rhlR-R were used. For amplification of the PA4103 promoter, p4103-F (EMSA) and p4103-R-FAM were used. All PCR products were purified by with QIAquick gel purification kit (QIAGEN). 50 nM 6-carboxyfluorescein (6-FAM)-labeled promoter DNA and various amounts of 6His-BfmR (as indicated) in 50 µl of binding buffer (10 mM Tris-Cl, pH 8.0; 1 mM DTT; 10% glycerol; 5 mM MgCl2; 10 mM KCl; 50 mM acetyl phosphate) were incubated at room temperature for 10 min. 0.01 unit of DNase I was added to the reaction mixture and incubated for 5 more min. The digestion was terminated by adding 90 µl of quenching solution (200 mM NaCl, 30 mM EDTA, 1% SDS), and then the mixture was extracted with 200 µl of phenol-chloroform-isoamyl alcohol (25∶24∶1). The digested DNA fragments were isolated by ethanol precipitation. 5.0 µl of digested DNA was mixed with 4.9 µl of HiDi formamide and 0.1 µl of GeneScan-500 LIZ size standards (Applied Biosystems). A 3730XL DNA analyzer detected the sample, and the result was analyzed with GeneMapper software.
Overnight P. aeruginosa cultures were washed and diluted 100-fold in M8-glutamate minimal medium (6 g Na2HPO4, 3 g KH2PO4, 0.5 g NaCl, 0.24 g MgSO4, 0.5 g Glutamate per liter; pH 7.4) supplemented with glucose (2 g/L). The bacteria were subsequently grown at 37°C for 48 h (OD600≈1.0) with shaking (250 rpm). Total RNA was immediately stabilized with RNAprotect Bacteria Reagent (Qiagen, Valencia, CA) and then extracted using a Qiagen RNeasy kit following the manufacturer's instructions. The total DNase-treated RNA samples were then analyzed by CapitalBio Corp for Chip (Affymetrix) assay. Briefly, samples were labeled according to the manufacturer (Affymetrix, Santa Clara, CA) and then hybridized to the Affymetrix GeneChip P. aeruginosa genome array (catalog number AFF-900339) for 16 h at 50°C though the use of the GeneChip hybridization oven at 60 rpm. Washing, staining, and scanning were performed using the Affymetrix GeneChip system. The data were normalized using Robust Multi-array Average (RMA) [71]. Gene expression analysis was performed using three independent mRNA samples for each strain. Microarray data were analyzed using SAM (Significance Analysis of Microarrays) software [72]. Criterion such as cutoff limitation for fold change ≥2 or ≤0.5 and q-value ≤5% was used to select differential expression genes. All data were submitted to the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress) under accession number E-MTAB-1983.
The autoinducer of the rhl system, C4-HSL, was measured using an rhlA promoter-based P. aeruginosa strain, pDO100 (pKD-rhlA) [28]. This detection system was developed by fusing the C4-HSL-responsive rhlA promoter upstream of luxCDABE and introducing the construct into pDO100, a rhlI mutant strain [28]. Procedures were modified from the protocol described previously [28]. Briefly, the reporter strain pDO100 (pKD-rhlA) was grown in LB medium plus 100 µg/ml kanamycin overnight at 37°C with shaking (250 rpm) and diluted to an OD600 of 0.05 in fresh LB plus kanamycin. 90 µl was subsequently added to the wells of a 96-well microtitre plate. A 10 ml portion of the samples or medium control was added to the wells. The luminescence value was measured in a 2104 EnVision Multilabel Plate Readers or Synergy 2 (Biotek), and calculated from the luminescence value minus that of the medium control. The data are presented as CPS and are not normalized to OD600 of pDO100 (pKD-rhlA). In this assay, the growth curves of pDO100 (pKD-rhlA) are identical.
Different strains of bacteria were grown overnight in Luria-Bertani (LB) broth at 37°C with shaking. Then, the bacteria were subject to pelleting by centrifugation at 5,000 g and resuspended in 10 ml of fresh LB broth and allowed to grow until the mid-logarithmic phase. OD600 nm was measured, and the density was adjusted to 0.25 OD (0.1 OD = 1×108 cells/ml). Mammalian cells were washed once with PBS after overnight culture in full medium, and changed to a serum-free and antibiotic-free medium immediately before infection. Cells were infected by various strains at a multiplicity of infection (m.o.i) of 10∶1 bacteria-to-cell ratio at indicated time points for 2 h. The cells were washed three times with PBS to remove surface bacteria and incubated with 100 µg/ml polymyxin B for 1 h. The cells were lyzed to evaluate the internalized bacteria using CFU assay on agar dishes as described in our previous studies [73], [74].
The killing of MLE-12 cells by bacterial infection was performed by continuing incubation for 24 h and cell survival measured using MTT assay [75]. MLE-12 cells were cultured in 96-well plates as above. After incubation for 24 h, MTT dye was added to the cells in each well with at a final concentration of 1 µg/ml. Then, the cells were incubated at 37°C until the color developed. The yellow color may change to brown upon reduction by enzymes. The reaction was terminated by adding 100 µl of stop solution (10% DMSO, 10% SDS in 50 mM HEPES buffer). The plate was left at room temperature overnight. The next day, the 560-nm absorbance was read using a plate reader in order to quantify the dye conversion [76]. Background correction was done with controls containing only the dye.
A lettuce leaf virulence assay was performed as described previously [77]–[79]. Briefly, P. aeruginosa strains were grown aerobically overnight at 37°C with shaking (250 rpm) in PPB broth or PPB broth containing carbenicillin (100 µg/ml) when appropriate, washed, resuspended, and diluted in sterile MgSO4 to a bacterial density of 1×109 CFU/ml. Lettuce leaves were prepared by washing with sterile distilled H2O and 0.1% bleach. Samples (10 µl) were then inoculated into the midribs of Romaine lettuce leaves. Containers containing Whatman paper moistened with 10 mM MgSO4 and inoculated leaves were kept in a growth chamber at 37°C for five days. Symptoms were monitored daily. As a control, lettuce leaves were inoculated with 10 mM MgSO4.
All P. aeruginosa strains were grown at 37°C overnight in PPB medium with shaking (250 rpm), diluted 100-fold in fresh PPB medium, and incubated at 37°C for 2.5–3.0 h until the cultures reached OD600 0.8. Bacteria were collected by centrifugation, washed, and suspended in PBS buffer. Viable P. aeruginosa were enumerated by colony formation on Pseudomonas isolation agar (PIA) (Difco) plates in order to quantify the infectious dose. Mouse infections were carried out as described previously [26], using 8-week-old female C57BL/6J mice obtained from Shanghai SLAC Laboratory Animal Co. Ltd. and housed under specified pathogen-free conditions. Mice were anaesthetized with pentobarbital sodium (intraperitoneal injection, 80 mg/kg) and intranasally infected with c. 5×106 cfu of each bacterial isolate. After that, animals were sacrificed 18 h post infection. Lungs were aseptically removed and homogenized in PBS plus 0.1% Triton X-100 in order to obtain single-cell suspensions. Serial dilutions of each organ were plated on Pseudomonas isolation agar (PIA) (Difco) plates. Bacterial burden per organ was calculated and is expressed as a ratio of the inoculum delivered per animal. Statistical analysis was performed using Prism software (GraphPad).
Two-tailed Student's t tests or Mann–Whitney test was used to calculate p-values (two-tailed) using Prism software (GraphPad), as indicated. * p<0.05, ** p<0.01, *** p<0.001.
|
10.1371/journal.pcbi.1002460 | Genome-Scale Modeling of Light-Driven Reductant Partitioning and Carbon Fluxes in Diazotrophic Unicellular Cyanobacterium Cyanothece sp. ATCC 51142 | Genome-scale metabolic models have proven useful for answering fundamental questions about metabolic capabilities of a variety of microorganisms, as well as informing their metabolic engineering. However, only a few models are available for oxygenic photosynthetic microorganisms, particularly in cyanobacteria in which photosynthetic and respiratory electron transport chains (ETC) share components. We addressed the complexity of cyanobacterial ETC by developing a genome-scale model for the diazotrophic cyanobacterium, Cyanothece sp. ATCC 51142. The resulting metabolic reconstruction, iCce806, consists of 806 genes associated with 667 metabolic reactions and includes a detailed representation of the ETC and a biomass equation based on experimental measurements. Both computational and experimental approaches were used to investigate light-driven metabolism in Cyanothece sp. ATCC 51142, with a particular focus on reductant production and partitioning within the ETC. The simulation results suggest that growth and metabolic flux distributions are substantially impacted by the relative amounts of light going into the individual photosystems. When growth is limited by the flux through photosystem I, terminal respiratory oxidases are predicted to be an important mechanism for removing excess reductant. Similarly, under photosystem II flux limitation, excess electron carriers must be removed via cyclic electron transport. Furthermore, in silico calculations were in good quantitative agreement with the measured growth rates whereas predictions of reaction usage were qualitatively consistent with protein and mRNA expression data, which we used to further improve the resolution of intracellular flux values.
| Cyanobacteria have been promoted as platforms for biofuel production due to their useful physiological properties such as photosynthesis, relatively rapid growth rates, ability to accumulate high amounts of intracellular compounds and tolerance to extreme environments. However, development of a computational model is an important step to synthesize biochemical, physiological and regulatory understanding of photoautotrophic metabolism (either qualitatively or quantitatively) at a systems level, to make metabolic engineering of these organisms tractable. When integrated with other genome-scale data (e.g., expression data), numerical simulations can provide experimentally testable predictions of carbon fluxes and reductant partitioning to different biosynthetic pathways and macromolecular synthesis. This work is the first to computationally explore the interactions between components of photosynthetic and respiratory systems in detail. In silico predictions obtained from model analysis provided insights into the effects of light quantity and quality upon fluxes through electron transport pathways, alternative pathways for reductant consumption and carbon metabolism. The model will not only serve as a platform to develop genome-scale metabolic models for other cyanobacteria, but also as an engineering tool for manipulation of photosynthetic microorganisms to improve biofuel production.
| Cyanothece spp. are unicellular, diazotrophic cyanobacteria that temporally separate light-dependent oxygenic photosynthesis and glycogen accumulation from N2 fixation at night [1]. When grown under nutrient excess, Cyanothece sp. strain ATCC 51142 (thereafter Cyanothece 51142) cells can accumulate significant amounts of storage polymers including glycogen, polyphosphates, and cyanophycin [2]. The inter-thylakoid glycogen granules are significantly larger in size than those found in other cyanobacteria, which points at an unusual branching pattern and packaging of this compound. From a biotechnological perspective, this presents an intriguing theoretical possibility to accumulate substantially higher amounts of polyglucose without any significant increase in the number of granules [3]. Cyanothece 51142 is also of interest for bioenergy applications due to its ability to evolve large quantities of H2. Remarkably, H2 production in this organism can occur under light conditions in the presence of O2 and is mediated by nitrogenase [4], [5]
Sequencing of the Cyanothece 51142 genome [6] has enabled application of high-throughput genomic approaches to study the unique physiological and morphological features of this organism. Transcriptomic and proteomic studies have been conducted to analyze global gene expression patterns under a variety of environmental conditions and infer regulatory pathways that govern the organism's diurnal growth [7], [8]. The availability of genomic information also provides means to construct genome-scale constraint-based models of metabolism, which are powerful tools for systems-level analysis and prediction of biological systems response to environmental cues and genetic perturbations [9], [10]. Such models have been developed for a variety of biological systems [9] but only in a few studies has this approach been applied to photosynthetic microorganisms, including Synechocystis sp. PCC 6803 [11]–, Rhodobacter sphaeroides [14], and Chlamydomonas reinhardtii [15], [16]. However, the modeling of metabolism in oxygenic photoautotrophs is an intriguing problem due to the complexity of photosynthetic and respiratory electron transport chains, and the potential effects of two distinct photosystems upon the generation and fate of reductant and energy that drives the remainder of metabolism.
In this work, we developed the first genome-scale metabolic model of Cyanothece 51142 and used a combination of computation and experimental approaches to investigate how photosynthetic and respiratory fluxes affect metabolism. Discrete representation of PS II and PS I and their integration with multiple respiratory pathways enabled modeling of photon fluxes and electron flux distributions under conditions of variable light quality and intensity. The predicted changes in growth rates of Cyanothece 51142 in response to changes in light input were experimentally tested using a photobioreactor with controlled sources of monochromatic 630 and 680 nm light. We also carried out computational and experimental analyses of light- and nitrogen-limited chemostat growth of Cyanothece 51142 and used mRNA and protein expression data to constrain model-predicted flux distributions. Both in silico and experimental data suggest that respiratory electron transfer plays a significant role in balancing the reductant (NADPH) and ATP pools in the cells during photoautotrophic growth. This study is a first step towards a systems-level analysis of cyanobacterial metabolism, as it integrates information into a genome-scale reconstruction to understand metabolism qualitatively and quantitatively through a constraint-based analysis [9]. We also discuss strategies for improving internal flux distributions through integration of in silico simulations and data.
To build a constraint-based metabolic model of Cyanothece 51142, a genome-scale metabolic network was reconstructed using the genome annotation and data from NCBI [6], SEED [17], KEGG [18]–[20], and CyanoBase [21], [22]. The resulting iCce806 network contains 806 genes and 667 metabolic and transport reactions (see Dataset S1 and Tables S1, S2, S3 for network details). Most of the 42 reactions without genes associated with them were added to complete metabolic pathways needed for biomass production. The final reconstruction encompasses central metabolic pathways such as the Calvin-Benson cycle, the pentose phosphate pathway (PPP), reactions within the tricarboxylic acid (TCA) cycle, as well as, the complete set of anabolic pathways involved in biosynthesis of glycogen, cyanophycin, amino acids, lipids, nucleotides, vitamins, and cofactors. Pathways for glycolate synthesis (via ribulose-1,5-bisphosphate carboxylase/oxygenase, i.e., photorespiration), glycolate conversion to serine, and glycerol catabolism are also included. Photosynthetic electron transfer associated with the thylakoid membrane is represented as a set of four separate reactions, including light capture by photosystem II (PS II) and photosystem I (PS I), electron transfer between the two photosystems, and cyclic electron transfer around PS I. Similarly, respiratory electron transfer is represented by reactions catalyzed by terminal cytochrome c oxidase (COX), quinol oxidases (QOX, both bd- and bo-types), NADH dehydrogenases (NDH, type 1 and 2), and succinate dehydrogenase. In addition, two reactions (NADP+- and ferredoxin- requiring) for flavin-dependent reduction of O2 (i.e., Mehler reactions) were included. A simplified scheme of the photosynthetic and respiratory electron transfer reactions in iCce806 is shown in Figure 1.
For initial testing, we examined the ability of the constraint-based model of iCce806 to predict growth under photoautotrophic (using light and fixing CO2), heterotrophic (using glycerol in the dark), and photoheterotrophic (using glycerol and light) conditions with different nitrogen sources. In silico calculated biomass yields, which simulated carbon or light- limited growth (Figure S1), qualitatively agreed with previously reported growth data for Cyanothece 51142 [1], [2], [23]. Other non-growth conditions that were simulated with the model, included nitrogen fixation as occurs during the dark phase of Cyanothece's ciracadian cycle [1]. In this case, the oxidation of glycogen provides reductant and ATP for nitrogenase, and we examined the model's ability to quantitatively predict the amount of nitrogen (N2) that could be fixed and stored in the dark, by maximizing cyanophycin production from glycogen. Although H2 is an obligate co-product of the nitrogenase reaction, no H2 was produced in the initial simulations under dark N2-fixing conditions, contradicting experimental observations. Model examination revealed that all of the nitrogenase-generated H2 was utilized by hydrogenases to reduce NAD(P) and ferredoxin, which ultimately increased cyanophycin production. When the three hydrogenase reactions (HDH_1, HDH_2, and UPHYDR) were eliminated from the model, the predicted ratio of fixed N2 to consumed glycogen depended on the non-growth associated ATP requirement (NGAR), and was estimated to be 0.3 (NGAR = 2.8) or 0.67 (NGAR = 0) mole N2/mole glycogen, which was in accordance with an experimentally measured value of 0.51 [2]. Under this condition, the model predicted that H2 production would have same yields as fixed N2 (0.3 to 0.67 mole H2/mole glycogen) due to the stoichiometry of the nitrogenase reaction.
We also evaluated how fluxes through electron transfer reactions are affected by the nitrogenase flux under N2-fixing dark conditions. With glycogen being the sole source of reductant for both ATP-generating oxidative phosphorylation and N2 reduction, a balance between fluxes through respiratory pathways and nitrogenase reaction is needed. In the absence of the hydrogenase reactions, the model predicted that O2 reduction via COX, QOX, or Mehler reactions are required to consume NADH resulting from glycogen catabolism (Figure S2). The model predicts that the COX reaction is required to achieve the maximum N2 fixation rate since it generates more ATP than the QOX or Mehler pathways (∼9 O2 are needed per N2 fixed). This is consistent with the results from recent proteomic studies showing the CoxB1 (cce_1977) subunit of COX is more predominant during the dark [24], [25]. These results suggest terminal oxidases are important under dark N2-fixing conditions not only to generate an intracellular anaerobic environment for nitrogenase, but also to provide ATP for nitrogenase activity.
As photosynthesis and respiratory electron transport chains are interconnected in cyanobacteria [26], these pathways were allowed to interact in the iCce806 model. To perform model robustness analysis, we computationally explored the impact of key photosynthetic and respiratory pathways on growth rate and intracellular flux distributions under varying photon uptake flux for PS I, while the photon uptake flux for PS II was fixed at 20 mmol·g−1 AFDW·h−1 (Figure 2). First, the model was evaluated assuming only linear photosynthetic electron transfer. In this case, all alternative reductant sinks including the proton and O2 reduction as well as cyclic photosynthetic reactions around PS I were eliminated from the model (Figure 2A). Under this condition, growth only occurred at one value of photon uptake flux for PS I and extracellular organic products (ethanol, lactate and/or alanine with trace amounts of formate) would have to be secreted in order to generate enough ATP to support biomass production. Second, when cyclic photosynthetic reactions were added back, the photon uptake flux for PS I could vary with a fixed photon uptake flux for PS II, but significant amounts of extracellular products were still formed until the photon uptake flux for PS I exceeded ∼85 mmol·g−1 AFDW·h−1 (Figure 2B). No growth occurred unless PS I photon uptake flux was greater than or equal to the photon uptake flux for PS II. Only when the model was allowed to use both cyclic photosynthesis and O2 reduction reactions were no extracellular products predicted and the photon uptake flux for PS I could be less than that for PS II (Figure 2C). Since experimental data does not indicate that any by-products including H2 or organic acids are produced by Cyanothece 51142 at a detectable level during photoautotrophic growth with excess ammonium, a plausible mechanism for balancing growth through the generation of additional ATP may involve activity of the cytochrome oxidases.
The discrete representation of PS II- and PS I-mediated reactions and their interactions with multiple respiratory reactions in iCce806 enabled further in silico analysis of growth and electron flux distributions under photoautotrophic conditions of variable light quality and intensity. In this case, the complete model was used to explore which reactions would be used to support maximal photoautotrophic growth rates for different levels of PS II and PS I photon uptake fluxes. To predict the corresponding growth rates under light-limited conditions, we constrained the photon uptake fluxes (ranging from 0 to 60 mmol·g−1 AFDW·h−1) through each photosystem. The resulting phenotypic phase plane (PhPP) contained three distinct regions (Figure 3A): in two regions growth was limited only by fluxes through PS II (region 1) or PS I (region 3), while in region 2 growth was limited by both PS II and PS I photon uptake fluxes (i.e., increases in either flux would improve growth rate). By adding artificial ATP or NADPH generating reactions (ADP+HPO4+H→ATP+H2O and NADP+H→NADPH) to the model and analyzing changes in predicted maximal growth rates, we were able to identify that in regions 1 and 3 growth was NADPH/reductant-limited, while in region 2 it was limited by energy supply (Figure 3A).
To analyze the effect of photon uptake rates on electron flux distributions, we calculated the flux ranges using flux variance analysis (FVA) for all photosynthetic and respiration reactions within each PhPP region (Figure 3B). In this instance, PhPP FVA was run with constraints that restrict the model to a given region and to the maximum growth for each point in the region (in contrast, standard FVA is used at a single point in a region). Using PhPP FVA, we identified active (both minimum and maximum flux values are positive or negative), inactive/blocked (minimum and maximum fluxes are both zero), and optional (which could have at least one zero and one non-zero flux value somewhere in the region) reactions leading to optimal solutions in each PhPP region. This new analysis technique allowed classification of reaction usage across entire regions of the PhPP and is not restricted to fixed points within a region. While linear photosynthesis was active and Mehler reactions were blocked across the entire PhPP, there were differences in the usage of photosynthetic and respiratory reactions observed within all three regions (Figure 3B). Surprisingly, while generation of NADPH from reduced ferredoxin via linear photosynthesis is the key source of reductant, ferredoxin-NADP+ oxidoreductase (FNR) was predicted to be active in region 2, but optional in regions 1 and 3. Closer examination of in silico calculated electron flux distributions revealed that, in addition to FNR, the model utilized a cycle involving glutamine synthetase, glutamate synthase and transhydrogenase, resulting in ATP-driven NADPH production. In regions 1 and 3, the model predicts there is excess ATP, and so this cycle can be used instead of FNR to move electrons from ferredoxin to NADPH. However, this cycle is unlikely to be of any physiological relevance since there has been no experimental data supporting this route for making NADPH, and FNR is essential for photoautotrophic growth in unicellular cyanobacteria such as Synechococcus 7002 [27]. Differences in the predicted usage of respiratory reactions were also found. In region 1, where growth is limited by the flux through PS I, at least one of the COX and QOX reactions must be active to oxidize excess electron carriers (Pc, Cyt c6, or Pq) generated from PS II. Similarly, in region 3 under PS II flux limitation, excess electron carriers (Pq, Fd) must be reduced via NDH-1 or –2 or ferredoxin-dependent cyclic electron transfer (FdPq). Conversely, due to ATP limitation in region 2, the model favored reactions with higher proton pumping capacities and so both the QOX and FdPq reactions were inactive. The usage of COX was optional in region 2 and depended on photon uptake rates (e.g., COX reaction was inactive at the boundary between regions 2 and 3).
The model predictions (Figure 3A) were compared to batch growth experiments in the LED-photobioreactor which allowed instantaneous measurements of initial growth and photon uptake rates by Cyanothece 51142 cells exposed to different intensities and ratios of 630 and 680 nm light (Table 1). When Cyanothece 51142 cultures were illuminated with both 630 nm and 680 nm light, initial growth rates generally correlated with the total photon flux through PS II and PS I, with higher growth rates observed at 80 mmol·g−1 AFDW·h−1 total photon flux and 630 nm∶680 nm light ratio of 2∶1. When cultures were exposed to only a single wavelength of light (batch experiments 6–10), i.e., either 630 or 680 nm, Cyanothece 51142 cells displayed a similar trend with higher growth rates observed at higher photon flux intensities. The predicted growth rates were within 7% of the experimentally measured values, except for the two cases where single 630 nm wavelength irradiances were used (Table 1). The reasons for this are unclear but may be due to other physiological and/or biochemical phenomena such as state transitions that are not contained within the model but are operating in vivo.
Data from these batch experiments (batch experiments 1–5, Table 1) were also used to estimate the growth (GAR) and non-growth (NGAR) associated ATP requirements. NGAR is the amount of energy spent to maintain the cell (i.e., maintenance energy). GAR is defined as energy expenditures used on protein and mRNA turnover or repair, proton leakage, and maintenance of membrane integrity; it does not include ATP spent on polymerization reactions, which are already accounted for in the macromolecular synthesis pathways of the network. The time-averaged growth and photon uptake rates were used to constrain the model and the maximal amount of ATP hydrolysis was calculated (Figure S3) for each batch experiment. A plot of growth rate versus maximum ATP hydrolysis flux was generated and a linear fit used to estimate the GAR and NGAR values [28]. Specifically, the slope of the fitted line is the GAR (544 mmol·g−1 AFDW·h−1), and the y-intercept is NGAR (2.8 mmol·g−1 AFDW·h−1). The estimated GAR value is significantly higher than those reported from other bacteria [29]; however, these model estimates assume that all absorbed photons lead to photosynthetic fluxes (100% quantum efficiency) and that the overall efficiency of ATP production via all electron transfer reactions (photosynthetic and respiratory) are accurate. Depending on the growth condition the quantum yields can change, and for Cyanothece 51142 this value was reported to be between ∼70–100% for photoautotrophic growth [23]. Upon further analysis, we found the estimated Cyanothece ATP requirements were most sensitive to reductions in quantum efficiency and the amount of ATP generated by photosynthesis and respiration (Table S4). Since neither quantum efficiency nor combined photosynthetic and respiratory ATP production were experimentally measured for Cyanothece 51142, the original estimates, GAR = 544 and NGAR = 2.8 were used in all growth simulations.
Chemostat cultures grown under light and ammonium limitations were used to calculate metabolic fluxes and further understand reductant partitioning pathways in Cyanothece 51142. The differences in biomass composition between these growth conditions indicated a major shift in carbon partitioning pathways (Figure 4; and Table S5). In ammonium limited cultures, carbohydrates comprised almost half of cell biomass; in contrast, under light limitation, Cyanothece 51142 cells contained higher amounts of protein, nucleic acids, and approximately 10% cyanophycin. The quantitative biomass composition measurements were used to generate two separate biomass equations for the metabolic model; experimentally measured growth rate, photon uptake rates, and O2 production rates were included as additional model constraints (Table S6, in this case no mRNA or protein expression data is used by the model). Using FBA and through minimization of the overall flux magnitude, we calculated representative flux distributions under light and ammonium limitations (values listed in Table S1). As expected, changes in flux values were consistent with differences in measured biomass compositions used in the simulations: under light limitation, fluxes increased for reactions involved in biosynthesis of amino acids, nucleotides and cyanophycin, while ammonium limitation resulted in flux increases for glycogen biosynthesis.
Comparisons of global transcriptome profiles displayed by Cyanothece 51142 during ammonium- and light-limited chemostat growth also reflected the rewiring of cellular metabolism (Table S7). Under ammonium limitation, significant increase in relative mRNA abundances was observed for genes involved in N2 fixation (cce_0198, cce_0545–0579), iron acquisition (cce_0032–0033, cce_1951, cce_2632–2635), respiratory electron transport (cce_1665, cce_3410–3411, cce_4108–4111, cce_4814–4815) as well as peptide transport, synthesis, and protein repair (cce_0392, cce_1720, cce_3033, cce_3054–3055, cce_3073–3075). Among the most highly expressed genes in ammonium-limited Cyanothece 51142 cells was the one encoding 6-phosphogluconate dehydrogenase (cce_3746), a key PPP enzyme. Under light limitation, the major changes in the transcriptome of Cyanothece 51142 included upregulation of genes encoding: components of the photosynthetic apparatus and electron transport chain (cce_0776, cce_0989–0990, cce_1289, cce_2485, cce_2959, cce_3176, cce_3963); pigment biosynthesis (cce_0920, cce_1954, cce_2652–2656, cce_2908, cce_4532–4534); CO2 uptake and fixation machinery (cce_0605, cce_3164–3166, cce_4279–4281); ATP synthase (cce_2812, cce_ 4485–4489), and protein synthesis machinery (cce_ 4016–4030) (Table S7).
Global proteome profiles of Cyanothece 51142 corroborated the shifts in gene expression (Table S8). The abundance of proteins from central metabolism (glycolysis, TCA, and pentose phosphate pathway) all had significant differences between cells grown under ammonium- and light-limited conditions. Enzymes of the oxidative PPP branch, namely glucose-6-phosphate dehydrogenase (cce_2535–2536), 6-phosphogluconolactonase (cce_4743) and 6-phosphogluconate dehydrogenase (cce_3746), showed increased abundances under ammonium limited conditions. Similarly, two-fold increase in abundance levels was observed for gluconeogenesis proteins, including fructose 1,6-bisphosphatase (cce_4758), glucose-6-phosphate isomerase (cce_0666), glyceraldehyde-3-phosphate dehydrogenase (cce_3612), and phosphoglycerate kinase (cce_4219). In contrast, relative abundances of proteins catalyzing the conversion of glycerate-3P to pyruvate (cce_1789 and cce_2454) were unchanged or up-regulated (pyruvate kinase cce_3420) in light-limited cells. Consistent with the results from global mRNA profiles was the up-regulation of Cyanothece 51142 proteins involved in photosynthesis and carbon fixation under light-limited conditions (Table S8). Notably, two key components involved in the electron transfer to PS I, namely plastocyanin (cce_0590) and cytochrome b6 (cce_1383), displayed elevated peptide abundances in light-limited cells.
Since there may be more than one flux distribution that is consistent with the experimentally measured rates of growth, photon uptake, and O2 production we used FVA to identify required (flux must be non-zero), optional (flux may or may not be zero), or inactive (flux must be zero) reactions under light- and ammonium-limited growth conditions. As our initial simulations (Table 2) produced a large number of optional reactions (170 out of 667 for both growth conditions), that represent uncertainty regarding usage, we subsequently used the transcriptome and proteome data (TPD) to further constrain the model. Using a modification to a previously developed approach [30], we obtained a flux distribution that was consistent with measured rates and TPD while reducing the overall flux magnitude (Table S1). In this analysis, flux was favored through reactions for which proteins were detected and disfavored through reactions associated with undetected proteins and transcriptome data less than a given threshold (e.g., log2 of mRNA expression level is less than 8). The model constrained by TPD predicted that the majority of reactions in central metabolism would be active under both chemostat conditions (Figure 5). In addition, we subsequently applied FVA employing additional constraints arising from the TPD. Comparison between FVA results with and without TPD constraints demonstrated a significant decrease in the number of ambiguities (the optional reaction set) when TPD is used (Table 2).
While the number of optional reactions was reduced by incorporating TPD into the model, the flux spans (difference between maximum and minimum values) of individual fluxes was still large (>30 mmol·g−1 AFDW·h−1 for some central metabolic reactions, Table S1). These large flux spans could arise from cycles or alternative pathways in the model, and deleting these features from the model could subsequently reduce the flux spans. FVA was repeated using measured growth, photon uptake, and O2 release rates under light-limited conditions as constraints and with optional reactions were deleted (similar results were found for ammonia limited conditions, data not shown). Flux spans for reactions in central metabolism (Figure 5) were then calculated for a series of single or double reaction deletions in silico. The purpose of this analysis was to identify those reactions that exert the greatest impact on the flux span in central metabolism (Figure 6A). Single deletions of glyceraldehyde-3-phosphate dehydrogenase (GAPD or GAPD_NADP) or hydrogenase (HDH_1) reduced the average central metabolic flux span the most (from 74 to 22 mmol·g−1 AFDW·h−1). Other single deletions with significant effects included FNR and NDH-1, which are involved in photosynthesis and respiration. The reaction deletions shown in Figure 6A all had a larger impact on reducing average central metabolic flux span than did imposition of constraints based on TPD. There were cases where single deletions had large effects on other specific reactions, but only modest effects on overall central metabolic flux spans. For example, a single deletion in phosphogluconate dehydrogenase (PGDHr) reduced the span for glucose-6-phosphate isomerase flux (PGI) to 0 (Figure 6B), but only reduced the average central metabolic flux span by ∼0.7 mmol·g−1 AFDW·h−1. The in silico analysis of double reaction deletions did not yield any new double deletions that would reduce the average central metabolic flux span significantly. However, some double deletions strategies did reduce flux spans of individual reactions.
Several cyanobacterial metabolic models (all for Synechocystis PCC 6803) have been published, which represented photosynthesis as two lumped reactions [12], [31] for linear (PSII, Cyt b6f, PSI, and FNR) and cyclic (PS I and Cyt b6f) pathways. In this study, we modeled photosynthesis as a larger set of separate reactions [13] as this structuring allowed analysis of the effects of different illumination on the production and partitioning of reductant through photosynthetic and respiratory reactions, as well as the contribution of different electron transfer pathways to growth. Our PhPP FVA results showed how different photosynthetic and respiratory electron transport chain components are used to maximize biomass production under different lighting regimes. It was not surprising that linear photosynthesis was active in all three regions because the cell needs photons from both PSI and PSII to grow under photoautotrophic conditions. However, the Mehler reactions were inactive in all three regions when we only consider maximal growth rate solutions. In regions 1 and 3, reducing equivalents (e.g., NADPH) limit growth and the Mehler reactions would lower the amount of reducing equivalents available for growth. The Mehler reactions are less energetically efficient than NADH dehydrogenase and cytochrome oxidase so the model would not use them in region 2, where ATP is limiting. So while the Mehler reactions can carry flux in the model, using these reactions lowers the maximum growth rate making them inactive (blocked reactions) in our PhPP analysis. A recent study showed that the Mehler reactions are operational in Synechocystis sp. PCC 6803, serving as a sink for excess electrons [32]. These reactions are also likely to be active in Cyanothece 51142, since the associated proteins were detected in the proteomic data (Table S8). As a result the model only predicted non-zero Mehler fluxes when the proteomic data were used to constrain the model (Table S1).
In the absence of cyclic photosynthesis, other products including water (produced by COX, QOX or Mehler reactions), H2 (via hydrogenase), or small organic compounds (alanine, ethanol, lactate and formate) were predicted to be necessary in order to balance the electrons and ATP needed to support growth. In the presence of linear and cyclic photosynthesis reactions, these products must also be produced unless significant amounts of cyclic photosynthesis occurs (>3 times the amount of linear photosynthesis). Since H2 and small organic compounds are not generally produced under photoautotrophic conditions with excess ammonium, any additional energy is most likely supplied by cytochrome oxidase activities that reduce photosynthetically produced O2. Interestingly, in the absence of cytochrome oxidase activities in the model, the PS I fluxes must always be greater than or equal to the PS II fluxes. It was shown that the marine cyanobacteriium Synechococcus has a PS I/PS II protein ratio >1, which has been explained as a mechanism to protect PS II from photo-damage [33]. Under conditions with high levels of PS II activity, cytochrome oxidase activity may ensure an adequate supply of oxidized plastoquinone (needed for PS II) and reduce O2 concentrations to limit photorespiration.
Similarly, cyclic electron flow via NADH dehydrogenase- or ferredoxin-dependent routes have also been experimentally demonstrated to play important roles in balancing the amount of NADPH and ATP produced via photosynthesis. Synechocystis 6803 mutants lacking ndhD genes (encoding subunits of NDH-1) had significantly lower cyclic photosynthesis activity [34]. Although the mechanism of electron transfer from ferredoxin to the plastoquinone pool (without using NDH) is still unclear, its activity has been demonstrated in green algae [35] and higher plants [36]. Our computational simulations also showed that, under light-limited photoautotrophic conditions, cyclic electron transfer involving NADH dehydrogenase (NDH-1) is needed for maximal growth if ATP (rather than NADPH) is limiting. In an environment where PS I photon availability is high relative to PS II, cyclic electron transport is needed (Figure 2) to increase availability of PS I substrates (reduced PC or Cyt c6) and protect against photo-damage. Cyclic electron flow has been experimentally shown to help protect the photosynthetic apparatus from photo-damage [37]–[39]
In addition to studying the interactions between components of the photosynthetic and respiratory components computationally, we also experimentally evaluated cells grown under continuous light conditions in light- and ammonia-limited chemostats. The measured 630 nm and 680 nm photon uptake and O2 production rates suggests that reductant was being directed towards O2 via the Mehler, QOX, and/or COX reactions. In both chemostat conditions, the model predicted that steady-state growth rate could have been achieved using lower photon uptake rates by decreasing the amount of reductant generated by PS II that was predicted to reduce O2.
A limitation to flux balance analysis is that a wide range of flux values may be consistent with the constraints in the computational model. An iterative application of computational and experimental methods is an important strategy to improve the comprehensive understanding of cyanobacterial metabolism. We have begun to apply this iterative approach, by including mRNA and protein expression datasets as additional constraints beyond biomass composition and physiological rate measurements. Experimentally-measured TPD were successfully used to further constrain the model, and thereby reduce uncertainty and increase the number of required (that is, metabolically active) reactions (Table 2). However, there remained discrepancies in that the model did not predict flux through all reactions for which proteins were experimentally detected. Such discrepancies can be used to subsequently improve the model with previously developed approaches [40]–[42]. For example, an earlier version of the model did not predict flux through proline oxidase, even though proteome data demonstrated that proline oxidase was synthesized. This prediction arose because the model did not contain a reaction in which FADH2 (a product of the proline oxidase reaction) could be reoxidized to FAD. After experimental confirmation that proline can be used as a nitrogen source (implying activity of proline oxidase) by Cyanothece 51142 (data not shown), a FADH2 recycling reaction was included in the final iCce806 model.
Even with these additional TPD constraints, a wide range of flux values remained feasible (Figure 6). We should note that we did not take real enzymatic activities into account (which can be affected by post-translational modifications), as we did not have this type of data for the two conditions examined. Such data, if available, could be used as additional factors for determining whether to favor or disfavor fluxes through associated reactions (See Material and Methods). Other constraint-based methods for incorporating gene expression data use similar Boolean on/off type of constraints to restrict fluxes [30], [43], [44] and would be expected to yield results similar to those described herein. Thus, novel computational methods which can more quantitatively constrain the metabolic flux values are still needed. The strategy of evaluating fluxes for reaction deletions in silico can be used to identify knockout mutants that can potentially improve the resolution of intracellular flux distributions. A flux that is well resolved would have a small span meaning we can more definitively state its value. If the mutants show no growth defects then the corresponding reactions may not be used under the conditions tested, or alternative pathways not included in the model may occur. Either way, this information could be used to better resolve the intracellular flux distribution or improve the metabolic model. For Cyanothece 51142, this would require development of a genetic system (such a system already exists for another Cyanothece strain [45]) as experiments with mutants would have the most potential to improve resolution of central metabolic fluxes during photoautotrophic growth. Also, as a complement to the in silico reaction knockouts that our simulations predict would reduce the flux spans associated with central metabolic reactions, the photobioreactor employed here provides a system whereby cultivation conditions can be rigorously controlled and some aspects of physiological state monitored continuously. In addition, cells from steady-state or perturbed cultures can be interrogated via physiological or biochemical analyses to experimentally test the predictions of the computational models for wild type or mutants. As the number of available cyanobacterial models continues to grow, cross-species physiological, genomic, and metabolic comparisons will enable the identification of core networks and contribute towards improving our understanding of metabolic processes in cyanobacteria.
Cyanothece 51142 was grown in modified ASP2 medium [46] amended with 0.75 mM K2HPO4, 0.03 mM FeCl3•6H2O, and 17 mM NH4Cl which substituted NaNO3 as the nitrogen source. Routinely, the cells were maintained under continuous white light illumination (50 µmol photons·m−2·s−1) in 1-L Roux bottles sparged with CO2-enriched air (0.3% vol/vol). Culture purity was confirmed by plating on DIFCO Bacto Tryptic Soy Broth and DIFCO Luria-Bertani solid media (BD Diagnostic Systems, Franklin Lakes, NJ) as well as by phase contrast or acridine orange fluorescent microscopy.
Controlled batch and chemostat cultures of Cyanothece 51142 were grown in a 7.5-L borosilicate glass vessel operated at 5.5-L working volume under the control of New Brunswick Bioflo 3000 bench-top bioreactor (New Brunswick Scientific, Edison, NJ). The vessel was housed in a custom-made black anodized aluminum enclosure equipped with light-emitting diodes (LED) generating 680 nm and 630 nm light for the preferential excitation of chlorophyll a and phycobilin pigments, respectively. Built-in sensors allowed for automatic adjustment of incident and transmitted light intensities using custom-designed control module. Both hardware and software components of the LED enclosure and the control module were developed at Pacific Northwest National Laboratory (US Patent Application # 20100062483; http://appft1.uspto.gov). All experiments were carried out under continuous illumination in modified ASP2 medium sparged with CO2-enriched argon (Ar) (0.2% vol/vol). Agitation, temperature, pH, and gas flow rates were maintained at 250 rpm, 30°C, 7.4, and 2.8 L/min, respectively. Incoming and off-gas composition was constantly monitored by an in-line mass spectrometry based gas analyzer MGA iSCAN (Hamilton Sundstrand, Pomona CA). Cell density was monitored spectrophotometrically at 625, 678, and 730 nm.
To establish a light-limited chemostat culture, the photobioreactor was inoculated with 10 mL of mid-log phase Cyanothece 51142 cells and maintained as a batch culture under 630 nm and 680 nm illumination at 40 and 70 µmol photon·m−2·s−1, respectively, until the culture reached late logarithmic stage. Chemostat mode was initiated by continuous inflow of medium at a dilution rate of 0.05 hr−1 that resulted in a steady-state optical density (OD730) of 0.20. Similarly, a nitrogen-limited continuous culture of Cyanothece 51142 was established using low-nitrogen ASP medium containing 0.75 mM NH4+. The ammonium-limited chemostat was maintained under identical operating conditions in regard to the culture dilution rate and optical density under incident light at 38.5 and 73.5 µmol photon·m−2·s−1 for 630 nm and 680 nm wave lengths, respectively).
The light uptake fluxes (mmol·g−1 AFDW·h−1) were determined by multiplying the light consumption rates (µmol photon m−2s−1) by the surface area of cell culture exposed to light (m2) and dividing by the amount of biomass in the reactor (g AFDW). The light consumption rates were determined by subtracting the transmitted light intensity from the values of incident light intensity after corrections were made for the abiotic consumption of light to account for the gas bubbles and probes in the reactor. Cells in the 5.5 L working volume were assumed to be equally exposed to the light at all times. Based on the inner diameter and height of the liquid culture at working volume, the surface area was 0.1403 m2. The amount of biomass in the reactor was determined from the working volume and biomass concentrations.
Biomass ash-free dry weight (AFDW) was measured using centrifuged (11,000× g, 4°C) cell pellets as described previously [29]. Total protein, reducing carbohydrates, RNA, and DNA were assayed using standard analytical techniques [47]–[49]. The total lipid fraction was measured gravimetrically after an extraction from a known volume of freeze-dried culture using previously published methodology [50]. Total reducing carbohydrates were quantified using the anthrone method [51] with glycogen as the standard. Chlorophyll concentrations were measured as described elsewhere [52], [53]. Amino acid composition was analyzed in acid-phenol hydrolyzed samples prepared using Eldex hydrolysis/derivatization station (Eldex Laboratories, Inc., Napa, CA) [29]. The derivatized samples were resolved on a 4-µm AccQ-Tag Nova-Pak C-18 column (3.9 mm×150 mm, Waters Corp., Milford, MA, USA), eluted using a linear gradient of acetonitrile (from 1.2% to 4.2% over 15 min.; from 4.2% to 6% over 4 min.; from 6% to 20% over 12 min.; at 20% over 1 min.; from 20% to 60% over 1 min.) with a flow rate of 1.0 ml/min at 37°C, and detected at 254 nm (HPLC system and UV detector by Shimadzu, Tokyo, Japan). Cyanophycin was estimated based on relative amino acid values and total protein measurements (see Text S1 for details).
Previously developed whole-genome oligonucleotide microarrays of Cyanothece 51142 [7] were manufactured by Agilent Technologies (Santa Clara, CA). RNA isolation, labeling, hybridization, and data analysis were performed by MOgene, LC (St. Louis, MO) using published protocols [7].
Cell lysis and tryptic digestion followed a previously described “global protein preparation” scheme [54]. A reference peptide database was prepared using strong cation exchange fractionation (10 fractions) of a portion of each global digest, as previously reported [55]–[57]. The methods for capillary liquid chromatography and mass spectrometry have been described in detail elsewhere [54], [58], [59]. Here, the HPLC mobile phase was 0.1% formic acid in water (A) and 0.1% formic acid in acetonitrile (B). A Finnigan LTQ ion trap mass spectrometer (ThermoFinnigan, San Jose, CA) was used for MS/MS analysis of SCX fractions and an LTQ-Orbitrap (Thermo) was used for high-resolution MS analysis of the global unfractionated samples. Each of the 10 SCX fractions was analyzed once, while each global digest was injected four times.
To build an accurate mass and time (AMT) tag database, SEQUEST analysis software was used to match the MS/MS fragmentation spectra to sequences from the annotation of the Cyanothece 51142 proteome [6]. Peptide identifications from the SCX fractions were combined with identifications from unfractionated samples to create a reference database of calculated mass and normalized elution time for each identified peptide. This database was used for subsequent high-sensitivity, high-throughput analysis of Cyanothece 51142 samples using the AMT tag approach [60]. LC-MS features from the unfractionated global samples were matched to the rich database built from the fractionated samples to give accurate peptide IDs. The area of each LC elution peak was used as a measure of peptide abundance.
Data from the AMT output were imported into the software MDART (Burnum et al., unpublished results), for filtering using a mass error tolerance of <5 parts per million, delta match score >0 (a measure of peptide uniqueness), match score >−1, and absolute normalized elution time error <10,000. The resulting 7450 peptides were imported into the software tool DAnTE [61] for further filtering and analysis. Peptide abundances were transformed to log base 2 and mean-centered. A linear regression-based normalization method available in DAnTE was then applied within each replicate category. Peptide abundances were used to infer the corresponding protein abundances through the ‘Rrollup’ algorithm in DAnTE [61]. During the Rrollup step, peptides were excluded if not present in at least 3 of the eight datasets, and Grubbs' outlier test was applied with a P-value cutoff of 0.05 to further remove outlying peptides. For increased confidence in protein identifications, each protein was required to be identified by at least 2 unique peptides, resulting in a total of 865 proteins. The minimum observed relative protein abundance value (14,465) was imputed as a crude surrogate for missing data for statistical calculations. Statistical differences between the two samples (4 technical replicates of each) were determined using ANOVA with a P-value cutoff of 0.05 (q<0.03) in DAnTE [61].
A draft metabolic network of Cyanothece 51142 was reconstructed in SimPheny (Genomatica, San Diego, CA) using a previously described automated model-building process [62]. Metabolic reactions and gene-to-protein-to-reaction (GPR) associations from other models were incorporated into the reconstruction if good BLAST hits could be found between genes in Cyanothece 51142 and genes in other modeled organisms. Additional reactions were added as necessary to produce known biomass constituents or utilize known nutrients; detailed literature, database, and BLAST searches were then carried out to find genes encoding these reactions in Cyanothece 51142 genome. This resulted in several new GPR associations that were incorporated into the reconstruction.
Based on the metabolic reconstruction, a constraint-based metabolic model for Cyanothece 51142 was developed as described in [63]. Fluxes are limited based on several different types of constraints: steady-state mass balance constraints (Eq. 1), enzyme capacity and thermodynamic constraints (Eq. 2) [10], given by:(1)(2)where S is a stoichiometric matrix for the reaction network, v is a flux vector, and α and β are parameters that limit the capacity and directionality of individual reactions. Flux balance analysis (FBA) uses these constraints to identify a flux distribution which maximizes or minimizes an objective function, such as growth rate [10]. Flux variability analysis (FVA) can also be used to determine the range of values each flux can take that are consistent with Eq. 1 and 2, by maximizing and minimizing each flux individually [64].
To further constrain the models based on mRNA or protein expression data, a modified version of the method developed by Shlomi et al. [30] was used. Here, we identified a single flux distribution that best agreed with measured transcriptome and proteome data (TPD) and minimized flux usage. Reactions with experimentally measured fluxes belong to set RE (which included biomass production and exchange fluxes for oxygen, 630 nm and 680 nm photons) and were constrained to their measured values. Reactions associated with detected proteins were included in the high reaction set (RH). Reactions associated with undetected proteins and genes with low mRNA expression levels (whose mRNA expression was less than the lowest mRNA expression of detected proteins) were included in the low reaction set (RL). The method finds a flux distribution that maximizes the number of active reactions (v≠0) and inactive reactions (v = 0) in reaction sets RH and RL, respectively. For reactions in set RH, binary variables x and y indicate whether a reaction is active, meaning its flux is greater than a positive threshold ε (x = 0 and y = 1), or smaller than a negative threshold -ε (x = 1 and y = 0) for reversible reactions. If both x and y are zero then the reaction is inactive and its flux value is zero. Likewise, a binary variable z is used for reactions in set RL such that if z = 1 then the reaction is inactive (v = 0). The original method [30] has alternate solutions, which can contain unrealistically high flux values due to the presence of cycles (e.g., futile cycles and circulations) in the network. To identify a solution that minimizes the use of these cycles, the objective function was modified to also minimize the sum of squared fluxes through the network.
The mixed integer quadratic programming formulation to identify a flux distribution that best matches TPD while minimizing flux magnitude is given below (Eq. 3).(3)Additionally, to find the flux ranges consistent with the TPD, flux variability analysis (FVA) was performed by minimizing and maximizing the flux through each reaction in the network. In these FVA simulations, the same constraints described above were included and the binary variables (x, y, and z) were further constrained by their optimal values (xopt, yopt, and zopt) found in the original problem (formulation below, Eq. 4). In this study, all model simulations were performed in GAMS software (General Algebraic Modeling System, GAMS Development Corporation, Washington, D.C.(4)
|
10.1371/journal.pntd.0004212 | Environmental Transmission of Typhoid Fever in an Urban Slum | Enteric fever due to Salmonella Typhi (typhoid fever) occurs in urban areas with poor sanitation. While direct fecal-oral transmission is thought to be the predominant mode of transmission, recent evidence suggests that indirect environmental transmission may also contribute to disease spread.
Data from a population-based infectious disease surveillance system (28,000 individuals followed biweekly) were used to map the spatial pattern of typhoid fever in Kibera, an urban informal settlement in Nairobi Kenya, between 2010–2011. Spatial modeling was used to test whether variations in topography and accumulation of surface water explain the geographic patterns of risk.
Among children less than ten years of age, risk of typhoid fever was geographically heterogeneous across the study area (p = 0.016) and was positively associated with lower elevation, OR = 1.87, 95% CI (1.36–2.57), p <0.001. In contrast, the risk of typhoid fever did not vary geographically or with elevation among individuals less than 6b ten years of age.
Our results provide evidence of indirect, environmental transmission of typhoid fever among children, a group with high exposure to fecal pathogens in the environment. Spatially targeting sanitation interventions may decrease enteric fever transmission.
| Typhoid fever, a serious bloodstream infection caused by the bacterium Salmonella Typhi, is commonly associated with direct, person-to-person transmission as a result of improper hygiene and unsafe food/water handling practices. Recent evidence, however, suggests that individuals may be indirectly exposed to typhoid through contact with fecal contamination in their immediate environment. In this study we investigated the role of environmental sources in the transmission of typhoid fever across an urban slum in Kenya by mapping the occurrence of cases in both children and adults. We tested the hypothesis that cases (relative to non-cases) cluster in low elevation areas as a result of the downstream flow and accumulation of fecal waste. We found that cases of typhoid fever among children tended to be concentrated in the downstream area. In adolescents and adults, on the other hand, there was little evidence of a geographic pattern in the risk of typhoid fever. These results provide evidence that environmental transmission of typhoid fever contributes to the risk of disease in children but not adults and adolescents, an observation most likely attributed to the fact that children are more likely to be exposed to fecal contamination through outside play. Interventions to improve local sanitation may therefore provide particular benefit to children who are at most risk of exposure to and acquisition of typhoid fever from environmental sources.
| Typhoid fever is a systemic, enteric disease caused by Salmonella enterica serovars Typhi and Paratyphi and has an estimated annual global incidence of 26.9 million cases, and causes 200,000 deaths per year [1]. Morbidity and mortality due to typhoid fever occurs primarily in young children in Africa and Asia [2, 3]. Children lack natural immunity and experience high levels of exposure to fecal pathogens [4]. If untreated, the case fatality can exceed 10%, although appropriate antibiotic treatment can reduce case fatality to 1% or less [5].
Transmission of typhoid fever depends primarily on direct contact with the stool of an infected individual [6–8], and risk is highest in densely populated areas that lack proper sanitation and access to safe drinking water [3, 9]. Household-level hygiene and food/water safety and handling practices, as well as close contact with an index case, are associated with the direct transmission of typhoid in endemic areas [9–11]. Because S. Typhi is exclusively human host-adapted, reservoirs of infection exist solely within groups of infected humans, a small number of which (1–6%) develop a chronic carrier state [12, 13], which has allowed the disease to persist during inter-epidemic periods [14].
Recent evidence suggests that environmental reservoirs of infection may also support disease transmission. The risk of typhoid fever is associated with environmental factors, including proximity to open sewers and highly contaminated water bodies, residence in low elevation areas, and rainy season [3, 15–17]. Major outbreaks of S. Typhi have been linked to contaminated municipal water sources, and suggest waterborne transmission as an important environmental pathway [10, 18]. Whether environmental sources contribute to endemic transmission during non-outbreak periods is unclear.
There is limited study on the epidemiology and environmental drivers of S. Typhi infection in Africa [19], where the incidence in some urban areas parallels that of high burden regions of Asia [2]. Additional data are needed to investigate the role that environmental reservoirs play in the endemic transmission of typhoid fever in Africa, particularly among children who are at an elevated risk of infection [2]. Such information can be used to predict where risk is greatest and can inform targeting for vaccination programs, water and sanitation improvements, or other community interventons [20].
We utilized a spatial modeling framework with climatic and remotely sensed data to estimate the geographic distribution of typhoid fever risk among a large disease surveillance cohort in Kibera, a densely populated, urban informal settlement in Nairobi, Kenya. We examined the contribution of environmental exposures to transmission by testing for associations between typhoid fever risk and variations in the hydrologic landscape. These data suggest that environmental transmission is an important contributor to the risk of typhoid fever in young children but may not be important in adults and adolescents.
The protocol was reviewed and approved by the Institutional Review Boards of the United States Centers for Disease Control and Prevention (US-CDC) and the Kenya Medical Research Institute (KEMRI). All data analyzed were anonymized to ensure confidentiality of the study participants.
Kibera is an informal urban settlement with between 250,000–500,000 residents in Nairobi, Kenya [21]. The area lacks adequate sanitation infrastructure, as indicated by open sewers and limited access to clean water [22], and has a large burden of many infectious diseases, including an adult HIV prevalence at 12.6% [23]. The wet season in Kibera is characterized by long rains from March–May and short rains from October–November; dry season runs from June–September and December–February. The average monthly temperature is 19°C [24].
Disease surveillance data were obtained from an ongoing, KEMRI/CDC-Kenya, population-based, household and clinic surveillance system in Kibera. Details of the surveillance have been described previously [2, 24]. Briefly, about 28,000 individuals have been followed biweekly since 2006 by trained community interviewers, and those with fever during these household visits were advised to seek medical attention at the surveillance site clinic. Blood cultures were conducted on all consenting individuals presenting to the clinic with an axillary temperature of ≥38.0 degrees C. The surveillance area covers 0.40 km2, has a high population density (70,000 individuals/km2) [2], and lies at an altitude of 1700–1740 meters, with terrain gently sloping towards the Motoine River in the southeastern portion of the study area (Fig 1).
Incident cases of symptomatic typhoid fever were ascertained between Jan 1, 2010 and Dec 31, 2011 at Tabitha Clinic, the primary surveillance clinic in the study area providing free care to those enrolled in the Kibera cohort. Tabitha Clinic is centrally located within 600 meters of the entire study population. Cases were defined as individuals who presented to Tabitha Clinic with acute febrile illness ≥ 38°C and from whom S. Typhi was isolated through blood culture. Only the first episode of typhoid fever during the 2-year study period was included for any individual with more than one occurrence. Cases whose residential location could not be verified 14 days prior to a typhoid fever episode were excluded from the analysis.
Controls were randomly selected from the underlying surveillance cohort (~28,000 individuals) in order to estimate the geographic distribution of the population at risk. A pool of control households with verified GPS coordinates was compiled for the years 2010–2011 (study years) and all individuals residing in those households during that time were eligible for control selection if they met the following criteria: 1) not identified as a case of typhoid fever between 2010 and 2011 (whether or not fever had been reported), 2) enrolled in the study with at least one household study visit between 2010 and 2011, and 3) non-missing GPS coordinates at the time of study visit. Four controls per case were selected at random, and one study visit from each control was randomly selected from the series of study visits. The rationale for selecting controls randomly was to estimate the spatial distribution of the underlying population at risk. Controls were not matched to cases on potential confounders such as age and location in order to allow for analysis of those co-variates.
Handheld Garmin GPSMap76CSx units were used to capture Global Positioning Systems (GPS) points on case households 14 days prior to diagnosis (to account for the average incubation period), and at the time of interview for the control households. Household elevation was estimated for each case and each control using a 90 X 90 meter digital elevation model (DEM) downloaded from NOAA.gov [25]. The DEM was used to generate a continuous 90 X 90 meter flow accumulation surface using the ArcHydro Toolset in Arc GIS 10.0 [26], with values at each location corresponding to the upstream contributing area draining to that point (e.g., lower-values are higher in the watershed and have less upstream area draining to that location, whereas high-value cells are lower in the watershed and have more upstream contributing area). DEM data are freely available, provide good approximations of the dominant flow direction of surface water at the watershed scale [27], and have been used to model the hydrologic diffusion and geographic distribution of different infectious diseases [28–31]. We used DEM-derived elevation and flow accumulation surfaces to approximate the level of exposure to accumulated fecal contamination at each location across the study area. We also measured the Euclidean distance from each case/control to the nearest point along two heavily polluted streams that bound the study area to test for associations between typhoid fever risk and exposure to point source contamination along the streams.
Rainfall data recorded at a weather station at Nairobi’s Jomo Kenyatta International Airport (JKA) between 2010 and 2011 were downloaded from https://data.noaa.gov/dataset/global-surface-summary-of-the-day-gsod, “The Global Surface Summary of Day” product, produced by the National Climatic Data Center (NCDC) [32]. The total accumulation of rainfall over 3, 7, and 30 day periods was calculated 14 days prior to each case diagnosis (to account for the average incubation period for typhoid following exposure) and 14 days prior to each control interview for comparability.
We conducted a spatial case-control analysis to identify geographic and environmental risk factors for S. Typhi infection. We compared the distribution of demographics (age, gender, household size) and environmental/hydrological variables (elevation, flow accumulation, distance to stream, and daily rainfall) between cases and controls in both univariate and multivariate logistic regression models. Multivariate analyses were stratified on age group (children over ten and adults/adolescents under ten years of age) in order to test for differential effects of environmental risk factors on typhoid transmission in each age-stratified risk group. Children under ten years of age experience the highest incidence of typhoid fever in our study area [2], and may also have unique exposure pathways that warrant further investigation. Multivariate analyses were adjusted for confounding factors identified a priori. All logistic regression analyses were performed using STATA 11 (STATACo, Texas 77845 USA).
We used spatial regression models with logit links to compare the geographic distribution of cases relative to controls among groups of individuals over and under 10 years of age, adjusting for the spatial confounding effects of individual-level co-variates. Under the null hypothesis, the distribution of cases relative to controls is expected to be equivalent to the age-group-specific case-control ratio at every location across the study area. Maps of the adjusted odds were produced for each age group (> and < 10 years) using a locally weighted regression smoother in a general additive model (GAM) framework for case-control data [33] using a logistic link function and a non-parametric component for the residual spatial surface:
Logit(P)=α0+β1(x1)+β2(x2)+S(lat,lon)
Where P is the probability of being a case versus control, x1 is age in years, x2 is the number of individuals in the household, and S is a thin plate spline used to smooth latitude and longitude over the geographic extent of the study area. We adjusted for individual-level factors in the model (age and household size) to estimate the residual spatial surface and test for its significance. When plotted on a map, the residual spatial surface is expressed as the residual log-odds of being a case versus control–where a value of zero corresponds to the expected case-control ratio. A log-odds above (below) zero indicates a larger (smaller) than expected proportion of cases to controls. We tested for significance of the log-odds surfaces against the null hypothesis of no spatial heterogeneity. Multiple transmission events within the same household were considered independent observations and were thus modelled without adjusting for spatial autocorrelation at the household level. The residual spatial surface models both small scale household-level transmission events as well as larger scale environmental heterogeneity in risk.
All spatial analyses were performed in the R-statistical software [34] using the mgcv package for fitting GAMs (methods described elsewhere [35]) and visualized using the splancs package[36]. GAM plots were stratified on age greater than and less than ten years.
A total of 118 cases of typhoid fever were confirmed at Tabitha clinic between Jan 1, 2010 and Dec 31, 2011. Of those, 111 had confirmed residence and household-level GPS coordinates at the time of infection and were included in the analysis. One case was excluded as it was a repeat episode of typhoid on the same individual within a one month period, which we consider a recrudescent case as opposed to two separate transmission events. The 110 incident cases included were unique individuals who resided in 103 unique households, with 7 secondary cases within the same household. Among the 7 secondary cases at the household level, 5 occurred within 30 days of an index case in the same household, which we classify as intra-household transmission events. Four hundred and forty controls were randomly selected from the underlying population at risk, comprising 416 households, with interview dates spanning from January 6, 2010 to December 29, 2011.
The distribution of demographic and environmental characteristics between cases and controls is shown in Table 1. Cases were significantly younger than controls (mean age of 8.4 years versus 16.4 years, respectively, (p<0.001). More than half of cases (56.4%) were under age ten versus 34.6% of controls (p<0.001, Fig 2). Cases resided in households with more individuals (70.0% of cases lived in households with more than 5 individuals compared to 48.6% of controls, p<0.001). There was no evidence that cases differed from controls with respect to gender (52.7% female versus 49.8% female, respectively, p = 0.579).
Comparison of spatial, temporal, and climatic characteristics of cases and controls
The spatial distribution of the 110 incident cases and 440 controls is displayed in Fig 1. Compared to controls, cases were concentrated in the eastern, lower elevation region of the study area. Only one case was observed in the westernmost part of the study area. The incidence of typhoid fever over the 2-year study period did not follow any seasonal pattern, nor was there any discernable association between monthly rainfall and risk (Fig 3).
Individuals with typhoid fever resided in lower elevation areas (19.1% of cases resided in the lowest elevation area, 1,695–1,707 meters, compared to 12.5% of controls, p<0.001) with higher flow accumulation (33.6% of cases resided in areas with the highest flow accumulation area compared to 18.0% of controls, p<0.001). There was no evidence that cases differed from controls with respect to season (43.6% of cases occurred in the wet season versus 51.1% of controls, p = 0.161), or total precipitation in the three days prior to infection/interview (68.2% of cases and 65.2% of controls occurred after three days with no precipitation, p = 0.205).
Household size (measured as number of inhabitants) was positively associated with risk of typhoid in both children (under age ten) and adults/adolescents (over age ten), OR = 1.27, 95% CI (1.11–1.46), p < 0.001; OR = 1.20, 95% CI (1.10–1.31), p < 0.001, respectively. Among children under age ten, those who resided at lower elevation had significantly greater risk of typhoid fever compared to those in higher elevation areas, OR = 1.87, 95% CI (1.36–2.57), p <0.001, corresponding to a ten meter decrease in elevation (Table 2), shown graphically in Fig 4B. Similarly, those children who resided in areas with higher flow accumulation had greater risk of typhoid fever compared to children in areas with less flow accumulation, OR = 1.27, 95% CI (1.00–1.62), p = 0.05. Among adults/adolescents over age ten, there was no evidence of an association between typhoid risk and elevation, OR = 1.23, 95% CI (0.89–1.71), p = 0.205 or flow accumulation, OR = 1.11, 95% CI (0.91–1.37), p = 0.305. Long periods of dryness (>30 days of no rain) prior to exposure/interview were positively associated with higher risk of typhoid in adults/adolescents, OR = 2.88, 95% CI (1.05–7.87), p = 0.040], with no evidence of an association in children.
Among children less than ten years of age, the risk of typhoid fever was geographically heterogeneous across the study area (p = 0.016) and generally followed a linear geographic gradient, with risk increasing from the west to east (Fig 4A). There was a weaker, and non-statistically significant (p = 0.150) spatial pattern in the risk of typhoid fever among adolescents/adults greater than ten years of age.
In this large, population-based, case-control study, risk of typhoid fever was spatially heterogeneous across a small geographic area. The observed spatial pattern in risk was especially pronounced among children under ten years of age, who experienced an almost doubling of risk for every ten meter decrease in elevation. In contrast, the spatial distribution in the risk of typhoid fever among adults and adolescents varied only slightly across the study area. Our results suggest distinct modes of transmission of typhoid fever between children and adolescents/adults in a Kenyan urban slum. Transmission in children may be more environmentally mediated than that in adults.
Low elevation areas aggregate bacterial pathogens from diffuse sources upstream via the overland flow of surface waters [37–42], and may serve as environmental reservoirs for a number of water-borne and water-related infectious diseases in children, including soil transmitted helminthes and schistsomiasis [30]. Our results build on evidence from previous studies that have shown environmental heterogeneity in the risk of typhoid fever [16, 43] by highlighting differences in environmental drivers of risk between children and adults.
The global health implications of environmental transmission of typhoid fever among children in Africa are significant. Young children are at the greatest risk of typhoid fever in densely populated urban areas with poor hygiene and sanitation infrastructure [2, 9]. Unlike adults and adolescents, children have underdeveloped natural immunity conferred by previous infection and are unable to fight systemic bacterial colonization [44]. Furthermore, young children are likely to be exposed to fecal contamination in the immediate environment surrounding their household [45]. Based on estimates from 2006–2009 in Kibera, children between 2–4 years of age and 5–9 years of age experienced the highest incidence of typhoid fever (2242.6 cases per 100,000 person years and 1,788 per 100,000 person years, respectively), compared to 821.5 cases per 100,000 person years among 0–1 year olds. The higher incidence among children two and over is consistent with behaviors associated with outdoor play and exposure to fecal pathogens in the environment. Based on our results, it is plausible that in typhoid-endemic areas children contract S. Typhi from environmental reservoirs at substantially higher rates than do adults. Environmental transmission may therefore account for a large part of the burden of typhoid fever in children and may in turn play an important role in its continued transmission.
We observed a negative association between precipitation and risk of typhoid fever in adults/adolescents greater than ten years of age, who experienced a 3-fold higher risk of infection after long dry periods. The effect of rainfall on risk of enteric pathogens, including typhoid fever, is unclear. Heavy rainfall is associated with increased risk of enteric disease transmission as a result of washing fecal pathogens into drinking water sources [46, 47]. One study from Nepal showed an increase in typhoid fever incidence in parallel with seasonal peaks associated with the rainy season [15]. On the other hand, long dry periods may increase the risk of diarrheal pathogens as a result of the limited availability of clean water for proper hygiene [48, 49].
A primary strength of our study was the use of population-based controls to estimate the underlying geographic distribution of the population at risk. The selection of controls in a clinic or hospital-based setting, a practice used in many spatial epidemiology studies, can obscure true disease-exposure associations if the controls do not adequately represent the geographic distribution of the underlying population at risk. A second strength is the stratification of our data into under and over ten years of age, which allowed us to compare the effects of environment on typhoid transmission across age groups.
The results of our study must also be considered in light of certain limitations. First, we lacked individual-level socioeconomic status (SES), which is needed to adjust for small-scale variation in access to resources, improved sanitation, and hygiene practices. Considering that our population was restricted to a small geographic area with a relatively homogenous distribution of SES and sanitation, we do not expect strong confounding by SES across the study area. Second, though Tabitha Clinic was the primary point of care for individuals with syndromes consistent with Typhoid fever (60–70% of individuals with acute fever seek care at Tabitha Clinic), there are inevitably missed cases [2]. Not all individuals with symptoms consistent with Typhoid in the cohort uptake care at Tabitha Clinic. We do not, however, expect bias due to residential proximity to the clinic. Tabitha’s location is closer to the upstream, low incidence area than the downstream, high incidence area and any confounding by proximity to Tabitha would therefore attenuate our estimates towards the null. Next, though DEM-derived surfaces have been used to model hydrological drivers of waterborne and water-related infectious diseases [30, 50], they are limited in their ability to capture fine scale heterogeneity in the flow of water and thus only crudely measure the accumulation of water-borne contamination. Finally, our study did not confirm environmental transmission via bacteriologic evidence indicating the presence of S. Typhi in the environment.
In summary, this study provides evidence of environmental transmission of typhoid fever in young children in an urban slum in Africa. Implementation of interventions to reduce transmission should include targeted sanitation improvements in areas of high geographic risk, particularly in low elevation areas where fecal waste tends to concentrate. Children living in these areas are at increased risk of typhoid transmission and may benefit from environmental interventions and targeted vaccination campaigns, as has been emphasized previously [2, 20].
|
10.1371/journal.pntd.0005144 | Maternal Filarial Infection Influences the Development of Regulatory T Cells in Children from Infancy to Early Childhood | Children born from filarial infected mothers are comparatively more susceptible to filarial infection than the children born to uninfected mothers. But the mechanism of such increased susceptibility to infection in early childhood is not exactly known. Several studies have shown the association of active filarial infection with T cell hypo-responsiveness which is mediated by regulatory T cells (Tregs). Since the Tregs develop in the thymus from CD4+ CD25hi thymocytes at an early stage of the human fetus, it can be hypothesized that the maternal infection during pregnancy affects the development of Tregs in children at birth as well as early childhood. Hence the present study was designed to test the hypothesis by selecting a cohort of pregnant mothers and children born to them subsequently in a filarial endemic area of Odisha, India.
A total number of 49 pregnant mothers and children born to them subsequently have been followed up (mean duration 4.4 years) in an area where the microfilariae (Mf) rate has come down to <1% after institution of 10 rounds of annual mass drug administration (MDA). The infection status of mother, cord and children were assessed through detection of microfilariae (Mf) and circulating filarial antigen (CFA). Expression of Tregs cells were measured by flow cytometry. The levels of IL-10 were evaluated by using commercially available ELISA kit. A significantly high level of IL-10 and Tregs have been observed in children born to infected mother compared to children of uninfected mother at the time of birth as well as during early childhood. Moreover a positive correlation between Tregs and IL-10 has been observed among the children born to infected mother.
From these observations we predict that early priming of the fetal immune system by filarial antigens modulate the development of Tregs, which ultimately scale up the production of IL-10 in neonates and creates a milieu for high rate of acquisition of infection in children born to infected mothers. The mechanism of susceptibility and implication of the results in global elimination programme of filariasis has been discussed.
| Lymphatic filariasis caused by thread like filarial worms involves asymptomatic to acute and/or disfiguring chronic conditions like lymphoedema, elephantiasis and scrotal swelling. Infection occurs when filarial parasites are transmitted to humans through mosquitoes. Adult worms lodge in the lymphatic system and disrupt the immune system that causes the disease. Nonetheless the infection if present during pregnancy, it affects the immune system of the unborn child in such a way that they become more susceptible to infection. But how the immune system of a fetus is affected by the maternal filarial infection is not known. Since regulatory T cells are responsible for development of hyporesponsiveness, a condition that supports the active filarial infection, and develops in thymus at an early stage of the human fetal development, we hypothesized that maternal filarial infection might be affecting the development of Tregs cell. Because Tregs secret IL-10, a regulatory cytokine, we have also measured its level in children born to infected and uninfected mother and correlate it with Tregs. We have observed a significantly high as well as a positive correlation between Tregs and IL-10 levels in children born to infected mother than the children of uninfected mother at the time of birth as well as early childhood indicating that Tregs and IL-10 contribute to immune modulation during pregnancy. Since ongoing MDA excludes pregnant mothers and children below 2 years of age, hence implementation of supervised therapy at the time of adolescent through MDA may help the programme in achieving the target of global elimination of LF by 2020.
| Lymphatic filariasis (LF) is a major cause of chronic morbidity in the tropics and sub tropics. According to a recent estimate more than 1.4 billion people across the world are at the risk of infection [1]. To eliminate LF globally by 2020, WHO has introduced annual mass drug administration (MDA) in different endemic countries since one and half decades. But studies have shown that the infection remains highly prevalent among children below five years of age even after several rounds of MDA [2–4]. Here question arises what makes these children more susceptible to infection even though infection levels have come down below threshold in these endemic areas. It is known that besides host genetics and environmental factors, maternal filarial infection plays some role to increase the susceptibility and outcome of the disease. Since pregnancy and early childhood are critical periods during which the inherited immune system of a child is shaped by the environment, the disease outcome in older age is possibly determined both in in-utero and at birth [5]. But it is not exactly known how in-utero exposures to parasite antigens affect immune responses and ultimately the outcome of disease in early childhood. The mechanism of such effects deserves to be explored since our previous findings suggest supervised therapy before pregnancy can reduce the infection rate among children [6].
Induction of regulatory T cells (Tregs) by pathogen is regarded as one of the mechanism of immune evasion in human. It is known that T cell hyporesponsiveness is associated with the active filarial infection, which is partly mediated by regulatory T cells [7]. The immune suppressive capacities of Tregs are due to production of down regulatory cytokines to inhibit inflammatory responses and facilitate the parasite survival [8,9]. Moreover a highly skewed Th2-type cytokine pattern, with a prominent role for the regulatory cytokine interleukin-10 (IL-10) has also been marked in neonates born to helminth-infected mothers [10]. In case of patent filarial infection the state of immune hyporesponsiveness has been observed to be associated with decreased proliferative responses and increased anti-inflammatory cytokines such as IL-10 and TGF-β [8,11]. As the Tregs develop in the thymus at an early stage of the human fetal development from CD4+CD25hithymocytes [12], the question arises that whether maternal infection during pregnancy affects the development of Tregs in children during their early life. Here we have made an attempt to find out the answer by evaluating the infection status, level of Tregs and regulatory cytokine IL-10 in a cohort of children born to filarial infected and non infected mothers.
Healthy pregnant women and their offspring born in Khurda District Headquarter Hospital of Odisha, India were enrolled in this mother-child cohort study. The study has received the approval from human ethical committee of the institute with a clause to obtain informed verbal consent from the research participants. The purpose of this research study has been explained in detail to all enrolled mothers in local language in presence of an unbiased witness of the community like Auxiliary Nurse Midwife (ANM) / Accredited Social Health Activist (ASHA) / Anganwadi Workers (AWW). All participants have given face to face oral consent for themselves and their children without a sign consent form to participate over the entire period of study. The name and detailed address of each consent participant has been recorded in data sheet both at the time of enrolment and during follow-up. The oral consent was preferable because (i) the project involves no risk while giving service to the public and benefits to the ongoing LF elimination programme and (ii) linguistic or literacy demands of the written format which requires signature or thumb impression.
This is a cohort study conducted in District Headquarter Hospital of Khurda, Odisha, India, known to be endemic for filarial (Wuchereria bancrofti) infections. The district has experienced 10 rounds of MDA with > 85% coverage since 2004 and reported 0.34% Mf in 2013 against 12% in 2004. The pregnant mothers admitted in the hospital for delivery during 2009–2011 without any complications, free from other chronic diseases and belongs to this region have been selected for the study. The pregnant mothers and their subsequently born children enrolled in the study live in 8 adjacent villages. The mother’s age, parity status, levels of formal education, clinical history of filariasis and history of drug consumption in MDA were recorded after enrollment. None of the mothers had signs/symptoms of clinical filariasis at the time of admission. All enrolled mothers have affirmed consumption of anti-filarials distributed during the annual MDA before pregnancy but not during pregnancy since the drugs are not recommended during pregnancy. At the time of delivery blood samples were collected from both mother and cord aseptically and aliquot in different sized tubes to avoid the chance of mislabeling. Serum was separated after centrifugation and stored at—70°C until further use.
Enrolled mothers having healthy full-term children were followed up in a house-to-house visit in the year 2014–15. During follow up along with detailed clinical history 1ml of venous blood sample was collected aseptically from each enrolled mothers and her children. On the basis of the availability of the baseline immunological parameters 49 mother-child pairs were identified for follow-up out of 158 mother-newborn pairs enrolled during 2009–2011. Amongst those 49 children, 28 are within 2–4 years of age and 21 within 4–7 years of age. Infection status of the mother-cord pair at the time of delivery and mother- child during follow-up was determined by diagnosing the presence of microfilaria and/or circulating filarial antigen in the peripheral blood collected at night between 20:30 to 22:30. The Mf (W. bancrofti) was determined by microscopy by examining the Giemsa stained thick blood smear and CFA was evaluated in serum samples using commercially available Og4C3 antigen detection assay kit (Trop BioMed, Townsville, Australia) following the manufacturer’s instructions.
The identification of Tregs (CD4+ and CD25hi T-cells) was determined by using fluorescently labeled antibodies specific to surface markers (CD4 and CD25). Briefly, 50 μl of heparinized blood collected from mother, cord and children were incubated in dark with 10 μl of anti-human CD4-FITC (BD-Bioscience), anti-human CD25-PE (BD-Bioscience) for 30 minutes at 4°C followed by addition of 2ml of lysing solution and incubation for 10 minutes at room temperature. The samples were then centrifuged at 250 X g for 10 mins and cell pellets were washed twice with 2 ml of sheath fluid (BD Bioscience). Finally the cell pellets were re-suspended in 0.5 ml of sheath fluid and subjected to flow cytometric analysis. Data were acquired by using BD FACS calibur flow cytometer and analyzed using cellquest pro software. The gating strategy for Tregs (CD4+CD25+ hi) cells is displayed in Fig 1.
The level of IL-10 was determined using IL-10 assay kit (Sigma Aldrich, USA) according to the instructions supplied by the manufacturer. Briefly, 100 μl of plasma and standards were added to each well of the antibody coated ELISA plate. The plate was sealed and incubated overnight at 4°C with gentle shaking followed by (i) 4 x wash with wash buffer and incubation with 100 μl of biotinylated detection antibody for 1 hour at room temperature, (ii) 4 x wash and incubation with 100 μl of HRP—streptavidin conjugate for 45 minutes at room temperature and (iv) 4x wash and incubation with 100 μl of colorimetric TMB reagent for 30 minutes. Finally50 μl of stop solution of 0.2M H2SO4was added and read in ELISA reader at 450 nm.
The statistical analysis was performed using GraphPad Prism software (version 4). Mann-Whitney test was used to analyze the difference between two groups of unpaired data and Wilcoxon signed rank test for paired data. Fisher's exact test was used to compare the difference of proportions between two groups. Kruskal-Wallis test with the addition of Dunn test was used to analyze the difference between more than two independent groups. The associations between Tregs and IL-10 levels were analyzed using Pearson’s correlation analysis. The level of significance was set at 0.05.
The summary of the enrolment and follow up of participants is depicted in Fig 2. A total number of 179 pregnant women admitted to hospital for delivery from July 2009 to July 2011 were evaluated for inclusion in this study. Twenty one (11.7%) of them was excluded because of complication during delivery or infant death or unwillingness. Finally 158 mother-new born pairs were enrolled for the study. At the time of enrolment 11.8% of the mother were microfilariae positive (3–210 per 60μl blood), whereas 44.5% of pregnant mothers were CFA positive (GM: 1925, range: 630–16596). Interestingly, 24.5% of infected mothers have shown transplacental transfer of filarial antigen to their cord, while none of the cord blood from CFA negative mother was CFA positive. Similarly the cord blood of neither CFA +ve nor CFA–ve mother was positive for Mf. During the study period total 109 mother-child pairs have been dropped because they are either non traceable, decline to participate, death of the children, moved out of study area or non availability of immunological parameter. Finally 49 pregnant mothers and their subsequently born children have been followed up during 2014-15.The mean duration of follow-up was 4.4 years (range, 2–7 years).
The characteristics of follow-up mothers and children have been described in Table 1. Amongst 49 follow up mothers 28 were CFA positive and 21 were CFA negative at the time of recruitment. Of the total 28 CFA positive mothers, only 3 were Mf positive at the time of enrollment. All of the study participants were living in rural areas and majority of them (83.3%) were house wives by occupation with primary level of school education (77.5%). Except filarial infection status, no difference was noticed in terms of age in years, multiparity status and educational level among the CFA +ve and CFA-ve mothers during follow-up. Amongst the CFA positive (n = 28) follow-up mothers, 18 mothers are still harbouring filarial infection (CFA +ve but Mf–ve) without any clinical symptoms of filariasis, 4 mothers have cleared CFA but have developed acute symptoms of filariasis (episodic attack of fever associated with inflammation of lymph nodes and lymphatics of legs/arms) and 6 mothers have cleared CFA without development of any clinical symptoms of filariasis. Whereas none of the CFA negative mothers had acquired filarial infection or developed any clinical sign/symptoms of filariasis.
Out of 28 children born to the infected mothers, 12 (42.8%) children have acquired filarial infection and become CFA positive. In contrast one of the children (1/21, 4.7%) born to the uninfected mothers has acquired filarial infection and become CFA positive. (OR = 15, 95% CI: 1.75–127.9, Z = 2.47, p = 0.013). Amongst the infected children 7 children were in the 2–4 years of age and 6 children were in 5–7 years of age. Out of the 12 CFA positive children 5 were from mothers who continued to be CFA positive where as 7 were from mothers those cleared CFA. While analyzing the infection status of cord of these 28 children it was observed that 21.4% (6/28) of them were CFA positive. Amongst those 6 cords positive children only 2 have become CFA positive during follow up. Statistically no significant difference (p = 0.67) was observed in acquiring infection among children born from CFA +ve mothers having CFA +ve (2/6, 33.3%) and CFA-ve (10/22, 45.4%) cord at the time of delivery. Interestingly none of the cord from uninfected mother was CFA positive at the time of enrollment. Also none of the children born to either infected or uninfected mother have detectable microfilariae and/or with any clinical signs/symptoms of filariasis. Besides presence of CFA no difference was observed in age, gender in children born to infected and uninfected mother. Based on the presence/absence of CFA in mothers and children during follow-up, the children of CFA positive mothers have been divided into 4 sub-groups i.e. group I: both mother and child are CFA positive (M+Ch+, n = 5), group II: mother positive but child negative for CFA (M+Ch-, n = 13), group III: mother negative but child positive (M- Ch+, n = 7) and group IV: both mother and child negative for CFA (M- Ch-, n = 3).
The expression of Tregs in infected mother–cord pairs was significantly high as compared to mother-cord pairs of uninfected mother (mother: p = 0.016, cord: p<0.001). Similarly Tregs cell expression was significantly high (p < 0.0001) in children born to enrolled CFA positive group of mothers in comparison to children born to enrolled CFA negative group of mothers (Fig 3A). Further we have observed a decreasing trend in the level of Tregs in children born to both infected and uninfected mother as compared to the cord blood (p <0.0001 for CFA+ve and p < 0.0001 for CFA-ve).
To evaluate the impact of maternal infection on development of Tregs in children during their early childhood, we have analyzed the Tregs in mothers as well as children born to two groups i.e. CFA positive and CFA negative group during follow up. Irrespective of the CFA status of mother at the time of follow-up, Tregs cells were significantly high (p = 0.01) in mothers who were CFA positive at the time of enrollment compared to enrolled CFA negative mothers (Fig 3B). But no significant difference (p = 0.14) in Tregs cell expression was observed among mothers of four different subgroups belonging to CFA positive group. Whereas significantly high Tregs cell expression was observed between these four subgroups of mothers compared to CFA–ve group mothers (M+Ch-vs M-Ch-: p<0.001,M+Ch+ vs M-Ch-: p = 0.0008,M-Ch+ vs M-Ch-:p = 0.0001, M-Ch-vs M-Ch-:P = 0.01).Children born to four sub-groups of CFA positive mothers showed significant (p = 0.01) difference among themselves. Further, children born to these four subgroups of mothers had higher levels of Tregs expression than children born toM-Ch-of CFA -ve mother ((M+Ch-vsM-Ch–:p<0.0001,M+Ch+ vs M-Ch-:p = 0.0008,M-Ch+ vs M-Ch-: p = 0.0003, M-Ch-Vs M-Ch-:p = 0.02). On the other hand one of the children born to CFA–ve mother has acquired filarial infection during follow-up.with 0.15% of Tregs expression.
We have quantitatively assessed the level of IL-10, the hallmark cytokine for regulatory response, in plasma of cord blood as well as children born to infected and uninfected mothers to evaluate role differentiated T helper cell subsets in filarial infection. At the time of enrollment level of IL-10 was significantly higher in mother as well as cord blood of CFA positive mothers as compared to cord and mother of CFA -ve group (mother: p<0.0001, cord: p<0.0001) as shown in Fig 4A.Further a decreasing trend in level of IL-10 has been marked in children compared to cord (p < 0.001 for CFA+ve and p = 0.007 for CFA-ve group). Similarly during follow up significantly higher level of IL-10 was observed in CFA +ve mother as well as their children in comparison to CFA–ve mothers and their children (p<0.0001for mothers, p<0.0001 for children). However when the comparisons were made between the four subgroups of mothers as well as children belonging to the CFA +ve mothers, no significant difference was observed in IL-10 level (p = 0.07 for mothers, p = 0.5 for children) among them (Fig 4B).But IL-10 level in the subgroup of enrolled CFA positive mothers was significantly higher compared to enrolled CFA negative mothers during follow up (M+Ch- vs M-Ch–: p = 0.002,M+Ch+ vs. M-Ch- p:0.002,M-Ch+ vs. M-Ch-: P = 0.013, M-Ch-vs. M-Ch-: P = 0.017). More than that IL-10 level was significantly higher in children born to all four sub-groups of CFA positive mothers than born to CFA negative mother as evident in Fig 4B (M+Ch- vs M-Ch–: p = 0.001, M+Ch+ vs M-Ch-: p = 0.002, M-Ch+ vs M-Ch-: p = 0.002, M-Ch- vs M-Ch- p = 0.03). One child born to CFA–ve mothers acquired infection and having IL-10 level of 9 pg/ml.
To find out the effect of Tregs cells on IL-10 secretion in infected and uninfected mother as well as their children, a correlation was made between percentage of CD4+CD25hi cell expressions and IL-10 level during the follow up. As shown in Fig 5A and 5B, no significant correlation was observed between Treg and IL-10 of enrolled CFA+ve (p = 0.38, r2 = 0.029) and CFA–ve mother (p = 0.91, r2 = 0.012) during follow up. However when we differentiate the CFA+ve and CFA-ve group of enrolled CFA+ve mother during follow-up, a highly significant positive correlation was observed among the CFA +ve mothers (p = 0.0008, r2 = 0.518; Fig 5C) in contrast to CFA-ve mothers (p = 0.47, r2 = 0.066, Fig 5D). From Fig 6A, it is evident that a significant positive correlation (p< 0.0001, r2 = 0.6987) exists between IL-10 level and Tregs in children of infected mothers. In contrast no correlation was marked between IL-10 level and Tregs in children born to CFA negative mothers (Fig 6B, p = 0.8541, r2 = 0.001).Moreover there was no difference in Treg and IL-10 correlation between CFA+ve and CFA-ve children born to enrolled CFA+ve mothers (CFA+ve: p = 0.01, r2 = 0.445; CFA-ve: p<0.001, r2 = 0.7732) as shown in Fig 6C and 6D).
The current study reveals that maternal W bancrofti infection during pregnancy up- regulates the production of Tregs and IL10 in offspring from infancy to early childhood and children born to infected mothers are at greater risk of acquiring filarial infection than children born to uninfected mothers. Further, evaluation of cord blood response and their correlation with infection status of children born to infected mothers suggests that in-utero sensitization rather than transplacental transfer of filarial antigen leads to increased susceptibility to filarial infection after birth.
Presence of parasitic infections during pregnancy is known to influence the immune system of an unborn child directly, through transfer of parasites or antigens across the placenta. As a consequence the neonates born to the infected mother become more susceptible to infection [2, 13, 14]. We have observed filarial antigen in 21.4% (6/28) of the cord blood of infected mothers but 42.8% of children born to them have acquired filarial infection which is double the figure of antigenemia of cord blood. This finding is supported by the previous work which showed that prenatal filarial specific immune tolerance as a consequence of active maternal filariasis increased the risk of infection during the first 7 years after birth [15]. Further, out of the 6 CFA positive cord only two children have acquired infection during follow up indicating no significant difference in acquiring infection between children born with CFA positive and CFA negative cord (from CFA positive mother).Therefore, it is speculated that early priming due to in-utero exposure rather than the transplacental transfer of filarial antigen is playing the key role towards disease susceptibility. Conversely, children born to infected mother but have not acquired infection during the childhood even though living in same endemic area may be due to heterogeneity in exposure to infective larvae, co-infection with malaria and geo-helminths that could bias T cell cytokine response and differences in genetic makeup [15].
Initially it was thought that filarial infections profoundly suppress the T-cell-proliferative and IFN-γ responses because of Th2 bias. Though these infections undoubtedly elicit Th2 cells, but recent studies show expansion of regulatory T-cell population while maintaining the hyporesponsive state in filariasis [16]. Epidemiological evidence suggests that in-utero sensitization results down-regulated responses among the offspring, on encountering the homologous antigen which may be due to bias in the fetal and neonatal immune response towards the development T regulatory cell. Moreover filaria-associated Tregs has been demonstrated to modulate the T and B cell proliferation and polarized cytokine production by effector T cells in microfilaraemics [7, 14].In the present study, significantly high level of IL-10 and Tregs cell from infancy to early childhood signifies their role towards the disease susceptibility. Recently some other studies have revealed that modulated/ regulated T cell responses associated with patent filarial infection reflects expansion of Tregs that include both Tregs induced in peripheral circulation and the thymus-derived Tregs [11, 17]. Further filarial infection during pregnancy leads to an expansion of functionally active regulatory T cells that keep Th1 and Th17 in check [18]. In the present study significantly high expression of Tregs in children born to infected mothers is at par with the findings of others and the high level of CD4+CD25hi cells in cord blood of infected mothers supports our hypothesis that regulation of Tregs cells start from the time of early priming during pregnancy. Contribution of such regulatory network towards hyporesponsiveness has been well documented during an in vitro study in newborns of filaria infected mothers [19]. Further, studies have shown that in utero stimulation with helminth-derived antigens divert fetal immunity towards Th2 responses and/or lead to anergy or tolerance [20, 21]. Since Treg cells produce regulatory cytokine IL-10 that modulates the entire repertoire (Th1/Th2/Th17) of CD4+ effector cell responses indiscriminately in filariasis that limits the Th1 response [22,23],we have evaluated the level of IL-10 in two groups of children to draw the functional relationship with Tregs. In our earlier study, cord blood from filarial infected mother exhibited decreased production of IFN-γ (Th1) response and increased production of IL-10 (Th2) indicating that immune responses have already been skewed towards Th2 type of response at the time of birth [24]. In addition the high level of T- regulatory cells and increased production of IL-10 in cord blood of infected mothers could down regulate inflammatory responses and create a susceptible environment for the parasite to grow. Similar to our observation, in a cross sectional study conducted in Kenya by Malhotra and others have found that maternal filarial infection increases childhood susceptibility to W. bancrofti and skews filaria-specific immunity toward a Th2-type cytokine response [25]. In the present study significantly high level of IL-10 in children born to infected mothers in comparison to children born to uninfected mother emphasizes that in utero sensitization down regulate the immune response in children since the time of birth. Though some of the children born to infected mother are free from infection during follow up yet they maintain high level of Treg and IL-10 which is in agreement with the previous work that shows that helminth-specific T cell immunity acquired in utero is maintained until at least 10 to14 months of age in the absence of infection [26]. In case of CFA-ve group of mothers who are still CFA-ve within the follow-up groups could be then classified as endemic normal as there is no record of infection prior to the survey. The low level of IL-10 in this group suggests that endemic normals have specific immune profile preventing filarial infection. Though the focus on this immunomodulation during helminth infections has been on IL-10, yet contributions of natural T regulatory cells (nTregs) appear to be significant [27, 28] in this context of our study.
A recent study in India has shown that frequencies of regulatory T cell markers were higher in asymptomatic microfilaremics and/or circulating filarial antigen positive subjects than in patients with chronic pathology. It also suggests a more prominent regulatory role of IL-10 producing Tregs [7, 17]. Though in the present study it was not possible to measure the IL-10 producing Tregs still a strong association of Tregs and IL-10 was observed in children born to filarial infected mother during their early childhood. It was also observed that expression of Treg and production of IL-10 in CFA+ve and CFA-ve children born to CFA+ve mothers do not differ from each other. This indicates that in utero priming determines the Treg and IL-10 production independent of acquisition of infection in later life. These findings supports the notion that immunologic memory established by priming of prenatal T cells with antigens that pregnant women encounters the infection that persists from gestation to childhood. This might be the cause of high incidence of infection among the younger age children (2–4 years old) in this cohort as observed by others [3, 15, 25, 26]. From this we can speculate that increased level of Tregs and high production of IL-10 initiates a cascade of hyporesponsive mechanism in children from the time of birth that down regulates the inflammatory responses and lead to a Th2 type of response so as to make them susceptible for parasite survival and ultimately determines the disease outcome in children.
The obvious limitation of our study is small sample size corresponding to both children born to infected and uninfected mothers. Albeit by drawing correlation we can interpret that Tregs in offspring from filarial infected mothers influence the IL-10 production as described in adults. When analyzing regulatory T cells, the measurement of transcription factor FoxP3, CD49b and LAG-3 markers for Treg and Tr1 cell population and intracellular FACS antibodies such as IL-10 in IL-10-producing Tregs were not possible due to poor resource which might have been useful to analyze the functional relationships between their number and mechanism of action.
In conclusion we can state that maternal filarial infection during pregnancy increases the susceptibility of children to infection by immune priming through expression of Tregs as well as regulatory cytokine IL-10. The high incidence of infection among the younger children even after 10 rounds (2014) of MDA in this area is due to high rate of Mf among pregnant women during 2009. While the cause of high Mf rate among the pregnant women might be due to low compliance because of social customs or back to back pregnancy. Hence the present findings relates to a greater impact on mass treatment programs aimed at elimination of transmission of W bancofti infection. To prevent the prenatal immune priming and tolerance supervised therapy can be introduced at the child bearing age of the women, so that they can be free from infection by the time of pregnancy and, thus, decrease the risk of infection during childhood. Implementation of such strategy will help the programme in achieving the target of global elimination of LF by 2020.
|
10.1371/journal.pbio.1000236 | Identification and Functional Characterization of N-Terminally Acetylated Proteins in Drosophila melanogaster | Protein modifications play a major role for most biological processes in living organisms. Amino-terminal acetylation of proteins is a common modification found throughout the tree of life: the N-terminus of a nascent polypeptide chain becomes co-translationally acetylated, often after the removal of the initiating methionine residue. While the enzymes and protein complexes involved in these processes have been extensively studied, only little is known about the biological function of such N-terminal modification events. To identify common principles of N-terminal acetylation, we analyzed the amino-terminal peptides from proteins extracted from Drosophila Kc167 cells. We detected more than 1,200 mature protein N-termini and could show that N-terminal acetylation occurs in insects with a similar frequency as in humans. As the sole true determinant for N-terminal acetylation we could extract the (X)PX rule that indicates the prevention of acetylation under all circumstances. We could show that this rule can be used to genetically engineer a protein to study the biological relevance of the presence or absence of an acetyl group, thereby generating a generic assay to probe the functional importance of N-terminal acetylation. We applied the assay by expressing mutated proteins as transgenes in cell lines and in flies. Here, we present a straightforward strategy to systematically study the functional relevance of N-terminal acetylations in cells and whole organisms. Since the (X)PX rule seems to be of general validity in lower as well as higher eukaryotes, we propose that it can be used to study the function of N-terminal acetylation in all species.
| Widely hailed as the workhorses of the cell, proteins participate in virtually every process within a living organism. How well they perform these diverse tasks depends on successful passage through the intricate course of protein production, from transcription of the protein-encoded DNA template to processing and folding of the nascent amino acid chain. Some of the processing steps—including enzymatic cleavage or the attachment of chemical modifications—take place during protein synthesis, while others occur afterward. One modification that takes place during protein synthesis is the attachment of an acetyl group at the tip (N-terminus) of proteins. Although N-terminal acetylation is found throughout the tree of life and the machinery and mechanisms responsible for this modification are quite well characterized, little is known about how it affects protein function. We analyzed the acetylation state of proteins in the fruit fly Drosophila melanogaster and show that this modification occurs at a lower frequency in flies than in man but at a much higher frequency than in yeast. Based on our dataset we developed a generic method that can analyze the biological relevance of N-terminal protein acetylation in any organism.
| To attain full functionality and/or to reach their final cellular localization, many proteins undergo obligatory modification or processing. During this maturation process, proteins are concurrently properly folded, proteolytically processed, and enzymatically modified. Some of these processes occur co-translationally, i.e. during protein synthesis, while others take place after protein synthesis has been completed. Acetylation of protein N-terminal α-amino groups takes place during protein synthesis [1]. This very common and irreversible modification of proteins often combines two consecutive events [2],[3]. In the first step, the N-terminal methionine (also referred to as initiator methionine [iMet]) is removed from the nascent polypeptide chain by methionine aminopeptidases. This event is not obligatory in protein biosynthesis and has been shown to take place only if the second amino acid is small and uncharged [4],[5]. Larger amino acids at this position prevent removal of iMet by steric hindrance [6]. In the second step, the acetylation of the amino-terminus is catalyzed by N-terminal acetyl transferases (NATs), a class of enzymes conserved in pro- and eukaryotes [7]–[11]. In eukaryotes both processes usually take place co-translationally on the nascent polypeptide chain and appear to be completed when 25–50 residues extrude from the ribosome, as revealed by in vitro studies [12],[13]. This indicates that the N-terminal region of a protein defines its acetylation status. Although previous work could show sequence specificities of the different NAT complexes, for some proteins acetylation does not take place even if the appropriate amino acid sequences are present, suggesting that additional yet unknown amino acid sequence patterns or other determinants like the secondary structure of the protein's N-terminus may play a role [10].
An estimated 60%–90% of the cytosolic proteins are acetylated at their N-terminus [3],[14], however the biological relevance of N-terminal acetylation has been determined only for a few proteins. This was in most cases achieved either through the analysis of mutants of NAT complex components [7], in vitro modification [5], or through mutants for single proteins [15]. Small GTPases such as Arl3p or Arl8 for instance require amino-terminal acetylation for their recruitment to Golgi membranes and lysosomes [15],[16]. In other cases, the acetylated N-terminus promotes protein-protein interactions as has been shown to be important for the binding of F-actin and tropomyosin and the maintenance of the resulting higher order structure [17],[18]. These examples clearly demonstrate that N-terminal acetylation promotes a variety of biological functions that cannot be predicted from the primary amino acid sequence. Therefore, there is a need for a method to generate and express—in cells and organisms—proteins that differ in N-terminal acetylation to investigate functional consequences of the presence or absence of an N-terminal acetyl group.
N-terminal acetylation has been identified in various organisms [10],[19]. A detailed analysis of NAT substrate specificity, sequence requirements, and conservation of substrate specificity for acetylation were only recently documented for yeast and human [11]. Datasets for invertebrates are not available and it has been suggested that acetylations in invertebrates appear to be rare [10]. Here we present an extensive compilation of mature protein N-termini of Drosophila melanogaster that was obtained by shotgun proteomics as well as the enrichment of N-terminal peptides by COFRADIC [20]. We show that amino-terminal acetylation is a common event in Drosophila and that the sequence requirements (amino acids) that promote iMet cleavage and N-terminal acetylation are similar to those in other eukaryotes. Moreover, our dataset enabled us to detect the use of 124 previously unknown alternative translation initiation sites and/or splice variants. A Pfam analysis [21] revealed that a protein's acetylation state in some cases strongly correlates with the presence of certain functional protein domains.
Finally, in contrast to earlier studies that were limited to the identification of amino acid determinants that promoted or inhibited N- terminal acetylation, in this study we could identify a definite determinant, i.e. a proline at position one or two of a nascent protein that prevents N-terminal acetylation under all circumstances. We refer to this finding as (X)PX rule. We have applied this rule to genetically modify a protein such that the biological relevance of N-terminal acetylation could be studied in cell lines and in flies. Since the (X)PX motif seems to be conserved among organisms we propose that by applying the (X)PX rule in similar ways in other species, the function of N-terminal acetylation can now be generically studied.
To enrich for N-terminal peptides, proteins from a membrane, cytoplasmic, and nuclear fraction of Drosophila Kc167 cells, respectively, were subjected to combined fractional diagonal chromatography (COFRADIC) [11],[20],[22]. In COFRADIC, free primary amino groups of proteins (i.e. α-N-termini and ε-amines from lysine residues) need to be chemically acetylated on the protein level. To further distinguish naturally acetylated and non-acetylated protein N-termini, protein amines were blocked by trideutero-acetylation, which leaves a mass tag of 3 Dalton on each free primary amino group [11],[23]. The fractions enriched for N-terminal peptides were then analyzed by mass spectrometry. We identified 835 N-terminal peptides (peptides starting at the residue 1 or 2 of the predicted sequence; Figure 1A, Table S1) among a total 4,203 distinct peptides (19.5%) identified from 8,402 fragment ion spectra. This corresponds to roughly 8.7% of the protein N-termini detectable by mass spectrometry (see Figure S1 for calculations). The actual coverage reached has to be considered much higher since only a subset of all annotated proteins will be expressed in exponentially growing Kc cells. Furthermore, a dataset consisting of 382 N-terminal peptides (Figure 1A) identified by a classical shotgun proteomics approach on Kc cells, that is not using COFRADIC, was additionally considered in subsequent analyses (retrieved from [24]). A comparison of the two datasets revealed that COFRADIC enriched for N-terminal peptides by a factor of roughly 10. In total the two datasets identified 1,102 protein N-termini.
Besides the confirmation of these 1,102 distinct annotated protein N-termini, we expected to find alternative start sites in these two datasets, i.e. peptides with an amino-terminus that starts at position 3 or later of the predicted polypeptide chain and by convention are considered to be internal peptides. However, some of these supposedly internal peptides start with a Met and are semi-tryptic. Others start with a small and uncharged residue, are preceded by a Met in the predicted protein sequence that is missing in the identified peptide, hence indicating an iMet removal as found for a classical protein N-terminus. Our dataset contains 124 distinct peptides that fulfill above criteria (Figure 1B, Table S2) and that we consider to represent alternative translation initiation sites or un-annotated splice variants. To further verify this, we analyzed the sequence context of the AUG that served as putative alternative start codon with respect to its Cavener sequence (C_A/G_A_A/C_AUG; Kozak sequence for insects, the initial Kozak sequence being CC_A/G_C_C AUG_G) [25],[26]. In addition, we analyzed whether the AUG used is the first AUG of that particular exon. A frequency analysis of the residues in the AUG context revealed that the presence of the Cavener sequence could be confirmed for the entire N-terminal dataset (C_A_A_A_AUG; Figure 2A) as well as the putative alternative start sites (C_A_A_C_AUG; Figure 2B). Notably, we detected a change in sequence preference at position −1 (from A to C), which fully complies with the Cavener consensus sequence [25],[26]. It is important to note that these consensus patterns are derived from aggregate frequencies of nucleotides 5′ to the AUG used for translation initiation. If however the Cavener sequence is analyzed for the proposed alternative initiation sites of a single gene model, the Cavener sequences of the used AUG may deviate from the consensus sequence. Cavener and Ray have already recognized this phenomenon and described the consensus as a “strictly statistical term” whereas the optimal context for each individual AUG is defined as a “functional term” [26]. Nevertheless, in about 61% in of the cases the AUG is flanked by an adequate or a strong Cavener sequence, indicating that they represent true alternative start sites (Text S1).
Although essential for genome annotation, computer-based prediction of protein N-termini and alternative translation initiation sites remains a difficult task [27]. In that respect, our dataset not only allowed us to confirm many of the predicted translation sites in the fly but also to identify novel alternative translation initiation sites. In combination, we identified 1,226 amino termini for Drosophila Kc (Figure 1C, Table S1), which have been used for all subsequent analyses.
We next analyzed these 1,226 N-termini with respect to N-terminal acetylation (Figure 1C, Table S1). We observed that in the majority of cases (63%) the iMet is removed and that aminopeptidase cleavage follows the same rules as determined for other organisms [28]. About 71% of the N-terminal peptides are acetylated. Of these 61% have the iMet removed, whereas for free N-termini almost 68% showed iMet removal. A comparison of the present data with the respective data from yeast and man [11] shows that N-terminal acetylation occurs in insects with a similar frequency as in humans. Moreover, the acetylation frequency with respect to certain residues seems to have shifted during evolution (Table 1). For instance, whereas most N-acetylated protein-termini in yeast begin with Ser and rarely with Ala, Drosophila has a high percentage of acetylated proteins that start with Ser or Ala, whereas Ala is the most commonly acetylated N-terminus in man. Finally, the acetylation state of a protein's N-terminus appears in most cases fixed in Drosophila cells as in human HeLa cells but is rare or often incomplete in yeast. Only 57 proteins were identified with both, either a free or an acetylated N-terminus (5% of total in Drosophila, 8% in human, and 45% in yeast [11]; Table S3). Thereof, 48 N-termini exhibited the same iMet cleavage and thus had identical amino acid sequences (Figure 1C), whereas nine showed alternative iMet cleavage.
To assess whether particular protein functions or functional domains are preferentially associated with the N-terminal acetylation state, a Gene Ontology analysis on a reduced set of GO categories (referred to as GO Slim) on all three levels, namely Cellular Component, Molecular Function, and Biological Process, was performed [29]. The results of this analysis are shown in Table S4 as well as in Figure S2A–S2C. Despite the fact that some categories show a statistically significant (p<0.05) over- or underrepresentation of either acetylated or free N-termini, the overall spread of the distributions of acetylated versus non-acetylated gene models does not allow one to make a clear correlation of protein function with a certain GO category or a group of GO categories. Specifically, none of the detected associations with GO categories was strong enough to predict the acetylation state of a protein.
To determine whether proteins that share a specific functional domain also share a common N-terminus (i.e., an acetylated or free amino terminus), a Pfam analysis was performed (see Materials and Methods for details). Pfam is a specialized database that stores protein family classifications and protein domain data and allows one to find relationships between functional domains and any other protein property of interest or classify a so far unknown protein into a protein family [21]. Because N-terminal acetylation is a co-translational process completed after the first part of a protein has been synthesized [12],[13], Pfam domains that start within the first 60 amino acids of a protein were considered. In contrast to the GO analysis presented above, some Pfam domains show a strong association with the acetylation status of certain protein N-termini (Tables 2 and S5A). For example, for the Importin-beta_N-terminal domain (IBN_N, PF03810.11), the N-termini of six out of 15 proteins predicted to contain such a domain have been identified (p<0.048). In all six cases the N-terminus was found to be acetylated. The lack of an N-terminal acetylation appears to correlate with a few selected domains comprising for instance the Tubulin/FtsZ_family,_GTPase_domain (PF00091.17, six out of 14 found: p<0.00025) or the Ubiquitin_family (PF00240.15, seven out of 20 found).
The functional relevance of any of the correlations we identified in this Pfam analysis is currently unknown. Assuming that the function of a particular domain requires an exclusively acetylated or free N-terminus, two possibilities arise: (i) the acetylation state of the N-terminus is determined by the first amino acid residues of the primary amino acid sequence and/or (ii) the domain contains a so far unknown motif that prevents or promotes acetylation. An example for the first is the association of the IBN_N domain with an acetylated N-terminus: all proteins containing such a domain have amino acids at their N-terminus that promote acetylation. The IBN_N domain itself however starts at position 24 or later, which implies that the conservation of the acetylation state is most likely linked with function. We cannot rule out that unknown motifs within the IBN_N domain exist that support N-terminal acetylation.
For the Tubulin/FtsZ_family,_GTPase_domain and the Ubiquitin_family domain that are linked with a non-acetylated N-terminus, the latter assumption is true: Proteins carrying either domain do not necessarily have amino acid residues that prevent N-terminal acetylation. Other, unknown determinants, likely to reside within the domains themselves, may exist that eventually prevent an acetylation of these protein N-termini.
Our initial Pfam analysis was restricted to domains starting within the first 60 residues of the N-terminal protein sequence. To see whether the correlation of any Pfam domain with the N-terminal acetylation remains, we extended the Pfam analysis to all domains irrespective of their location in the protein (Table S5B). For most of these domains, this association with a free or an exclusively acetylated N-terminus remains.
In the previous section we described the correlation between the N-terminal acetylation state and certain Pfam domains present in a protein. We next asked whether a similar correlation could be detected on the level of the primary amino acid sequence, i.e. within the first two amino acids of a protein's N-terminus. As already shown in Table 1, Ala, Ser, and Thr residues at the mature N-terminus as well as Met-Asp and Met-Glu are acetylated with a high frequency in flies and man (Table 1). On the other hand, proteins with a Val, Gly, or Pro residue at the mature N-terminus or a lysine or arginine as the penultimate residue tend to have a free N-terminus (Tables 1 and S1). This has already been shown for yeast and man [10],[11]. However, it has never been claimed that the presence of one of these amino acid residues (or any other) at the amino terminus is sufficient to unequivocally define a protein's acetylation state. The comparison of our dataset with the ones available for yeast and man (Table 1) clearly shows that a proline residue at the first or second position of the mature protein N-terminus always and in all species analyzed so far prevents amino-terminal modification by the acetylation machinery and thus seems to represent a generic inhibitory signal. Hence, we have formulated a simple rule: a protein with the sequence X1-Pro-X3 or Pro-X2 at its very amino-terminus remains unacetylated under all circumstances (X1 being Met or any small amino acid that allows iMet removal by Met aminopeptidases, X2 and X3 being any amino acid; Figure 3A–3D). This means that proteins, which undergo the partial removal of the initiator Met (iMet) in proteins with the sequence Met-Pro-X3, as has been reported by Boissel and others [5], remain not acetylated (Figure 3B, 3C). Similarly, from a protein having the sequence Met-Sur-Pro (Sur being a small and uncharged amino acid residue), the iMet will be removed and the processed amino-terminus will remain unacetylated (Figure 2D). We would like to emphasize that in this context a Pro at position 2 of the mature amino-terminus overrules N-terminal acetylation even in the presence of promoting amino acids such as Ser or Ala at position 1 (Tables S1). To the above described inhibitory potential of a Pro, we refer to as (X)PX rule.
To unequivocally confirm the (X)PX rule we quantified Pro residue containing protein N-termini by Selective-Reaction-Monitoring (SRM) in total lysates from Kc-cells (for details see Materials and Methods, Figure S3) [30]. SRM enables one to specifically target and quantify peptides of interest in complex mixtures and has shown to be more sensitive and selective than classical tandem-mass spectrometry experiments [31]–[33]. In contrast to conventional mass spectrometry approaches, SRM not only allows for the detection of peptides and peptide modifications but also for their absence. SRM measurements on a set of 17 N-termini following the (X)PX rule were carried out. In our measurements we included the acetylated and non-acetylated form of each peptide and also generated transitions for the iMet either to be cleaved or not (Tables S6 and S7). These targeted SRM measurements revealed that all N-termini are non-acetylated. The acetylated isoforms remained undetectable. Taken together, these findings confirm that a Pro at position 1 or 2 efficiently prevents the acetylation of a protein N-terminus.
In order to test the (X)PX rule in an in vitro situation we decided to either introduce or replace an inhibitory Pro into selected proteins with a conserved amino terminus and to measure the consequences of these alterations on N-terminal acetylation, similarly to the experiments reported by Boissel and colleagues [5]. First, in our dataset we identified the amino terminal most peptide (Ace-ADPLSLLR) of Hyrax/Parafibromin (Hyx, CG11990) as being acetylated after iMet cleavage. The tumor suppressor Hyx is a component of the Polymerase-Associated Factor 1 (PAF1) complex and has recently been found to be required for nuclear transduction of the Wnt/Wg signal in Drosophila [34]. As a second test protein, we chose to investigate the Cyclin-dependent kinase subunit 85A (Cks85A, CG9790) of Drosophila [35]. Cks85A has an important role in mitotic progression. The protein follows the (X)PX rule with a proline at position 2 of the primary sequence and the iMet cleaved upon translation (PADQIQYSEK, Table S1). In our datasets, we always found the protein to be non-acetylated.
In order to challenge the (X)PX rule, i.e. to either create or abolish an N-terminal acetylation, the cDNAs of hyrax and Cks85A were modified as follows: (i) in the Drosophila hyx cDNA the codon for the secondary Ala was replaced by a Pro (Figure 4A). Both the wild-type (wt) as well as the mutated cDNA were C-terminally HA-tagged, which allowed for the isolation of the respective proteins via immunoprecipitation; (ii) similarly, we replaced the codon for the secondary Pro in the Drosophila cks85A cDNA by either a Ser or an Ala. For both amino acids we found a strong promoting effect on N-terminal acetylation (Table 1). All constructs are driven by a ubiquitous tubulin-1α promoter. In the following we will refer to the different constructs as Hyx-A2P-HA, Hyx-Wt-HA, Cks-P2A-HA, Cks-P2S-HA, and Cks-Wt-HA, respectively. To test the (X)PX rule in vitro, Drosophila Schneider S2 cells were transiently transfected with one of the above cDNAs. The tagged proteins were isolated via immunoprecipitation trypsinized and the N-terminal peptides were subjected to mass spectrometry analysis via SRM (Figure 4B).
As expected, the N-terminus of Hyx-Wt-HA was found to be acetylated whereas the amino terminus of Hyx-A2P-HA was unmodified. Neither acetylated Hyx-A2P-HA nor unacetylated Hyx-Wt-HA was detectable. In addition, we observed a complete iMet removal from the Hyx-Wt-HA N-terminus and we could detect the unacetylated peptide MPDPLSLLR via SRM, indicating an incomplete iMet removal from the Hyx-A2P-HA isoform (Table S7). Quantitative analysis revealed that the iMet cleavage was omitted for approximately 9% of the proteins (Figure 4B), confirming the results of Boissel and others that have observed iMet retention in 20% of the cases [5]. This observation also demonstrates the inhibitory potential of a Pro residue at the penultimate position of the mature protein N-terminus ((X)PX-rule).
For both kinase mutants, Cks-P2A-HA and Cks-P2S-HA, we found the N-terminus to be acetylated in immunoprecipitation experiments (Table S7). For Cks-P2S-HA (but not Cks-P2A-HA) we also detected the acetylated peptide MSADQIQYSEK indicating an incomplete iMet removal (9.4%). When over-expressing the wt form of the kinase (Cks-Wt-HA) via a tubulin promoter (four attempts in two different cell lines), the cells stopped growing, and the protein could not be successfully measured after immunoprecipitation, neither by LC-MS/MS nor by SRM. This might be due to interference of this form of the protein with a proper cell cycle progression. Wt Cks85A is known to be essential for progression in mitosis [35]. Since the mutated Cks proteins could be well detected, this suggests that an unacetylated N-terminus of Cks seems to be relevant for its proper function.
In conclusion, we could show that the (X)PX rule robustly predicts the acetylation state of proteins in vitro. Furthermore, our experiments show that a single amino acid change is sufficient to explore the function of an N-terminal acetylation of any protein of interest.
To test the functional relevance of amino-terminal protein acetylation of Hyx in vivo, transgenic flies were generated using either tub>a2p-HA or tub>wt-HA, respectively. The hyx transgenes were integrated at the identical, pre-defined chromosomal locus (51D) on the second chromosome, making use of the site-specific phiC31-mediated integration system [36]. Transgene integration at the identical genomic locus guarantees the same protein expression levels. In our experimental context it is important to note that the Ala to Pro exchange in Hyx-A2P-HA alters the translation initiation context at position +4 and thus expression levels might differ between the two transgenes. In Drosophila, however, the +4 position has been shown not to be relevant for translation initiation efficiency [25],[26].
Flies harboring one copy of either the tub>wt-HA or the tub>a2p-HA construct rescued the lethality of hyx transheterozygous mutants (as had been shown for untagged hyx before [34]). To determine the expression levels of the transgenes, tub>wt-HA/CyO; hyx2/TM6B or the tub>a2p-HA/CyO; hyx2/TM6B flies (carrying one copy of the transgene and only one endogenous hyx wt gene) were collected. Total fly lysates were subjected to Western blot analysis without any prefractionation of the samples and revealed identical expression levels for the transgenes (Figure 5A). This confirms the predictions of Cavener and Ray [26] and due to equal expression levels allowed us to directly compare the biological activity of the two proteins in vivo.
To determine the acetylation state of the respective N-termini, Hyx-Wt-HA and Hyx-A2P-HA were immunoprecipitated from total fly lysates of either of the transgenic lines and analyzed by SRM, whereby transitions specific for (M)ADPLSLLR or (M)PDPLSLLR were measured. These data showed that the N-terminus of Hyx-Wt-HA was exclusively acetylated whereas the amino terminus of Hyx-A2P-HA always remained unmodified as was shown in the in vitro experiments.
To test the relevance of the N-terminal acetylation for the biological activity of Hyx, the transgenes were crossed into a hyx homozygous mutant background [34]. Both Hyx-WT-HA as well as Hyx-A2P-HA fully rescued mutant hyx allele combinations (hyx1/hyx2) to adulthood at 18°C as well as 25°C, without obvious phenotypic defects and at the expected Mendelian ratios (Table S8).
However, at 29°C animals carrying a single copy of the Hyx-A2P-HA transgene showed a ∼50% reduction in rescue capability (p>6.3521E-06) compared to the Hyx-Wt-HA flies that rescued at the expected rates. To assess whether the observed reduction in rescue capability might be correlated with changes in protein localization, an α-HA antibody staining of 3rd instar wing imaginal discs was performed. Figure 5B shows confocal images of Hyx-Wt-HA and Hyx-A2P-HA of stained wing discs from 3rd instar larvae reared at 25°C and 29°C, respectively. At both temperatures, a similarly strong nuclear staining was observed for Hyx-Wt-HA and Hyx-A2P-HA and no difference in localization could be detected.
To test whether the protein levels have changed at 29°C and might contribute to the difference in rescue capability a Western blot analysis of flies reared at 29°C was performed (same setup as described above). Indeed the Western blot revealed a 32% reduction of Hyx-A2P-HA protein as compared to the Hyx-Wt-HA control (Figure 5A). Therefore, it might be possible that at least in part the reduced protein levels of Hyx-A2P-HA at 29°C are the cause for the lower numbers of rescued flies. Mechanistically, the reduced protein levels may be explained in two ways: (i) one possibility is that protein expression is reduced due to the exchange of the codon for Ala to the one for Pro. This mutation leads to an exchange from G to A at position +4. Although Cavener and colleagues state that the +4 position is not relevant for translation initiation in Drosophila [25], we cannot rule out effects on the translation machinery under these conditions (i.e., 29°C). (ii) As a second possibility, protein half life might be affected. It has previously been reported that N-terminal acetylation may contribute to protein stability [3]. If this was the case for Hyx-A2P-HA, we assume that the reduced protein amount in A2P flies reared at 29°C is not due to a change of the N-terminal residue from an Ala to a Pro but rather directly related to the changed modification state of the protein's N-terminus. This hypothesis is supported by the so-called N-end rule, which relates the in vivo half life of a protein to the identity of its N-terminal residue. According to this rule both a Pro and an Ala (as well as Ser, Thr, Gly, Val, and Met) as N-terminal residues confer protein stability in all species [37].
In summary, our experiments show that the (X)PX rule allows one to generate transgenes in order to study the biological relevance of the presence or the absence of an N-terminal acetyl group in vivo.
Although the cellular machinery that is required to acetylate a nascent protein N-terminus is also conserved in invertebrates, it had been suggested that acetylation was a rare event in this animal subphylum since only a few modified N-termini had been reported [10]. In this study we present a comprehensive compilation of mature protein N-termini from Drosophila melanogaster Kc 167 cells and show that (i) N-terminal acetylation in fact is also a common protein modification in insects and (ii) follows the same rules (in terms of sequence requirements and iMet cleavage) and (iii) occurs with a lower frequency than in man but at with a much higher frequency than in yeast. Moreover, we observed that the acetylation at the alpha amines of certain residues seem to have shifted in frequency during evolution such that the acetylation of an N-terminus in insects is in some cases in strong agreement with the patterns found in man, while for others it resembles more the distributions and frequencies found in yeast. These differences in acetylation frequency, the partial acetylation of proteins, as well as the lack of comprehensive datasets from different animal kingdoms may explain why prediction of the acetylation state from a primary protein sequence is still not yet clear-cut [11].
In fact, for some N-termini we still lack the understanding of what determines their acetylation state (especially for Gly, Val, or Thr as the first or second residue in the mature protein). Also some proteins carrying an acetylation promoting Ser and Ala at position 2 remain unacetylated. In Drosophila, 30 of a total of 289 identified N-termini that expose a Ser residue at their mature N-terminus are not acetylated (11 of which are only partially acetylated, i.e. appear acetylated and non-acetylated after iMet cleavage) (Table S9). In two cases, an inhibitory Pro residue fulfilling the (X)PX rule prevents the acetylation in spite of a Ser in the first position. One protein CG2679-PB, for which we have defined a putative alternative start site (putative alternative iMet at position 36 in the predicted sequence; Text S1, Table S2), has a signal sequence (SignalP 3.0, http://www.cbs.dtu.dk/services/SignalP/; [38]) predicted to be cleaved between residue 25 and 26. For the remainder (as well as for the Val, Gly, and Thr) the only plausible explanation is that there must be other features within the first 50 amino acids that prevent N-terminal acetylation during protein synthesis. Polevoda and Sherman suggested that the secondary structure of a protein's N-terminus may play a role [10]. We have analyzed the predicted secondary structure of all identified proteins (residues 1–50, unpublished data) but could not identify any correlation of structure and acetylation state. Consequently the expected determinants (such as complex sequence patterns within the first amino acids) remain to be identified through sophisticated computational analysis for which this dataset may act as an ideal starting point.
If no additional determinants (such as patterns or structural motifs) exist steric hindrance could be responsible for the preservation of a free N-terminus despite the presence of promoting amino acid sequences. During protein synthesis, cofactors like chaperones could bind to the nascent polypeptide chain and prevent proper function of the acetylation machinery: for instance it is known that tubulins undergo a sequence of folding steps catalyzed by chaperones, which are commonly assumed to take place after translation is completed [39],[40]. Our data show that tubulins always appear to have a free N-terminus despite the occasional presence of amino acids that promote its acetylation. In this context it may therefore well be that chaperones could already associate with the transcription machinery, thereby preventing the acetylation of tubulin by steric hindrance.
We have observed that in Drosophila the acetylation is fixed in most cases. This is similar to the case in human and in contrast to yeast where partial acetylation is common [11]. Possible explanations for this discrepancy between the organisms could be that yeast cells (but not fly or human cells) grow too fast, making acetyl-CoA poorly available as substrate, or the degradation of the unacetylated proteins cannot cope with division rates and thus the proteins with a free N-terminus accumulate in the cells. Nevertheless, in certain cases a partial acetylation of proteins may be of relevance for yeast as well as for higher eukaryotes. To asses this relevance such proteins could be genetically engineered according to the (X)PX rule, shifting the equilibrium of the mutant proteins towards a fixed acetylated or free N-terminus.
N-terminal acetylation has been shown to be absolutely necessary for the proper localization, activity, or stability of various proteins. The current view is that the presence rather than the absence of an acetylation is important for protein function. Our Pfam analysis challenges this view as it also shows a clear correlation for a non-acetylated N-terminus and certain protein domains. Additional evidence for the perspective that the absence of an acetylation is also important for proper protein function has been demonstrated in the case of human hemoglobin. The exchange of Val to Ala at the second position in the human hemoglobin beta chain reverts the usually free N-terminus into an acetylated one. Patients carrying this so-called “Raleigh” mutation suffer from thalassaemia due to a reduced affinity of the mutant hemoglobin to oxygen [41], underlining the importance of the free amino terminus for proper function. Finally, our own transgene analysis on Cks85A strongly indicates that the preservation of the protein's non-acetylated N-terminus is relevant for its proper function. In this context it is important to note that no pattern exists that guarantees an N-terminal acetylation, although Ser and Ala as second residues usually ascertain a modification. For the prevention of acetylation the opposite is true: based on the data we formulated the (X)PX rule that states that a Pro residue at the primary or secondary position of the mature protein N-terminus prevents acetylation and assures a free amino terminus under all circumstances. Since the majority of protein N-termini are acetylated in eukaryotic organisms investigated so far, and due to the presence of an absolute inhibitory rather than a promoting signal, we conclude that the acetylation of protein N-termini is the default state and, where necessary, is prevented by specific motifs. Along this line it has been suggested that some N-termini are acetylated just because of the presence of a promoting pattern but that acetylation has no functional relevance [3]. One can easily imagine such a case: a Ser at position 2 in the primary sequence has been evolutionary conserved because its phosphorylation is relevant for its biological activity. Upon protein synthesis this Ser by default also promotes iMet cleavage and the subsequent acetylation of the N-terminus. The conservation of the Ser, however, occurred due to the need for Ser phosphorylation and not N-terminal acetylation. In such a scenario one could apply the (X)PX rule to discriminate whether acetylation and phosphorylation or phosphorylation alone would contribute to protein function: the introduction of an inhibitory Pro after the Ser residue would allow iMet cleavage as well as the phosphorylation of the Ser but would prevent the acetylation of the N-terminus. Likewise, the Ser to Ala (or comparable) amino-acid exchange would preserve the acetylation (at least in insects and human) but not the phosphorylation of the protein, allowing one to assess the contribution of the N-terminal acetylation to protein function. A relevant example would be the rat RNA polymerase subunit 6 (RPB6), which is phosphorylated by Casein KinaseII (CkII) at a Ser at residue 2. It has, however, not been reported whether the acetylation state of this protein is relevant for its function [42].
Unlike a phosphorylation, which is “dynamic” and may be rapidly added or removed from a protein upon stimulation of a cell, N-terminal acetylation has been shown to be a co-translational, irreversible, and thus a “static” modification [3]. However, we and others have evidence that acetylation of N-termini does not only occur co-translationally [14]: we have identified 511 internal, acetylated peptides that reside within the annotated protein sequence (Table S10). These N-termini do not comply with the suggested N-terminal acetylation rules and thus are suggested to occur post-translationally. They require a to-date unknown acetylase activity as well as internal proteolytic cleavage at a specific site of the protein. It is tempting to speculate that these post-translational acetylation events may also occur in a dynamic fashion, allowing fast responses to various stimuli.
It remains to be noted that acetylation of N-terminal Pro residues has been reported in yeast [43]–[45]. The experimental evidence for an N-terminally acetylated Pro residue is solely based on shifts of protein spots observed in comparative 2D gel experiments of NAT-mutant versus wt strains. A direct measurement of acetylation has not been performed in this context [45]. In contrast, in two large-scale studies N-terminal Pro acetylation has only been observed if an internal Pro residue is exposed at the N-terminus after post-translational processing of the protein (see Tables S1 and S10). From this we conclude that the (X)-P-X rule, which is formulated for the first three residues of the primary amino acid sequence, is robust and that possible exceptions to this rule will be extremely rare and most likely attached post-translationally by a yet undescribed protein activity.
Although amino-terminal acetylation has been studied for more than 30 years, some aspects, foremost the relevance of its presence or absence for individual protein function, remain unclear for most of the cases. With this work we have defined tools that will help to better understand the elusive mechanisms as well as to explore the function of protein N-terminal acetylation in all organisms in a protein-specific manner. Complementary studies using for instance mutations in the relevant NAT enzymes will help to eventually draw firm conclusions about the function of the N-terminal acetylation.
To generate expression vectors for the hyrax and Cks85A transgenes, the respective DNAs were amplified by PCR. The forward primers contained a 5′ Kozak consensus (CGCCACC) and the appropriate base exchanges to generate the desired wt or mutant forms.
PCR fragments were inserted into the pOP-118 vector to create cDNAs driven by a ubiquitous tubulin-1α promoter and tagged with a 3xHA. For in vivo experiments the expression cassettes were cloned into an attB integration backbone [36].
Drosophila melanogaster Kc 167 and S2 cells were cultivated in Schneider's Drosophila medium (Gibco/Invitrogen) supplemented with 10% heat-inactivated fetal-calf serum, penicillin (100 U/ml), and streptomycin (100 µg/ml) at 25°C. Cells were split 1∶4 (v/v) every 3–4 d when they reached confluency.
For immunoprecipitations 75 cm2 flasks of S2 cells were transfected with 2 µg of the respective tub-[cDNA]-HA plasmid via Effectene Transfection Reagent (Qiagen). Cells were collected 48 h after transfection and lysed with lysis buffer (20 mM sodium phosphate-buffer, 200 mM NaCl, 0.5% NP40, supplemented just before use with 1 mM DTT, and Complete™ protease inhibitor cocktail (Roche)). Cell lysates were incubated for 3 h with monoclonal anti-HA agarose beads (clone HA-7, Sigma). Finally, beads were washed three times with lysis buffer, twice with lysis buffer without detergent, and bound proteins were eluted with 0.2 M glycine (pH 2.5).
For in vivo assays, the following tester lines were used: transgenic vectors were injected into Drosophila (yw) that contained the second-chromosomal attP landing site at map position 51D [36]. Potential founders were individually outcrossed against the balancer stock yw, hs-flp; Sp/CyO; TM6b/MRKS. Their white+ progeny was then individually backcrossed to the balancer background to establish clonal stocks. The resulting founder stocks were verified for the correct transgene by PCR. Successfully established transgenic insertions were then crossed into hyx mutant backgrounds, as described previously [34].
Wing discs were isolated, fixed, and incubated as previously described [46]. Antibodies used for the staining were mouse HA11 (1∶1000, BAbCO) and a goat secondary antibody coupled to Alexa594 (1∶500, Molecular Probes).
For total protein isolation adult flies were grinded in liquid nitrogen and the homogenate was resuspended in a 4 times volume of lysis buffer (20 mM TrisHCl pH 7.5, 150 mM NaCl, 0.2% NP40, 10% glycerol supplemented just before use with 1 mM DTT, and Complete™ protease inhibitor cocktail (Roche)). Protein samples were run on a 10% SDS-PAGE at 140 V for 90 min and subsequently transferred to a Hybond-P PVDF Membrane (Amersham). After the transfer, the membrane was rinsed in PBST, blocked with 5% nonfat dried milk in PBST and then incubated with either monoclonal mouse anti-alpha tubulin DM1A (1∶2000, Sigma) or mouse anti-HA.11 (1∶1000, BAbCO). As secondary antibody, peroxidase-conjugated goat anti-mouse IgG (1∶10000, Jackson) was applied. Signals were detected via ECL Plus (Amersham). Intensities of bands were quantified using the histogram option (average gray value) of ImageJ (NIH).
Total protein lysates or proteins isolated via immunoprecipitation were reduced with 5 mM Tris(2-carboxyethyl)phosphine hydrochloride (TCEP) and treated with 10 mM iodoacetamide to modify cysteine residues. Tryptic digestion was carried out overnight using 5 µg trypsin per sample. Samples were purified by reverse phase C-18 chromatography (Sep-PacK, Waters). For mass spectrometry analysis samples were resuspended in buffer A (5% acetonitrile, 0.2% formic acid).
SRM was performed on a triple quadrupole mass spectrometer (TSQ Quantum Ultra EMR, Thermo Fisher Scientific) operated with Xcalibur 2.0.7 (Thermo Fisher Scientific). The instrument was coupled to an Eksigent nano-LC system. Samples were automatically injected into a 10-µl sample loop and loaded onto an analytical column (9 cm length×75 µm (internal diameter) packed in-house with Magic C18 AQ beads 5 µm, 100 Å (Microm)). Peptide mixtures were delivered to the analytical column at a flow rate of 500 nl/min of buffer A (5% acetonitrile, 0.2% formic acid) for 18 min and then eluted using a gradient of acetonitrile (10%–35%; 0.36%/min) in 0.2% formic acid at a flow rate of 250 nl/min.
SRM measurements were carried out with a Q1 resolution of 0.4 and a Q3 resolution of 0.7 m/z half-maximum peak width. Scan speed was set to 0.020 ms per scan event. SRM transitions specific for proteotypic peptides were generated using the SRM Workflow (software from Thermo Fisher Scientific). For each precursor transitions were calculated with precursor charges 2+ or 2+ and 3+. At least four y-ions with m/z> precursor were monitored. Collision energies (CE) were calculated according to the following formulas: CE = 0.034×m/z+3.314 (2+) and CE = 0.044×m/z+3.314 (3+). SRM traces were evaluated via the SRM Workflow. For each peptide the co-elution of all transitions was confirmed. Peptides with non-coeluting elution profiles or bad resolution or signal to noise ratios were not considered for further analysis. iMet cleavage was quantified by comparing the ion current profile of identical y-product ions.
SRM-triggered MS/MS experiments were performed to validate the identity of the peptides for the relevant SRM traces. MS/MS spectra were acquired with a Q3 resolution at 0.7 m/z half-maximum peak width. The scan range was automatically determined by the instrument using the assigned charge state of the precursor ion (2+). MS/MS spectra were assigned to peptide sequences using the Mascot software version 2.2 (Matrix Science).
The data for Kc167 cells from the previously released proteome catalog for Drosophila melanogaster (retrieved from [24]) were searched using the Mascot search algorithm for finding N-terminal peptides. The search criteria were set as follows: Variable modifications were set to Acetyl (N-term), Deamidation (NQ), Carbamidomethyl (C, or where required ICAT-C, ICAT-C:13C(9)), Oxidation (M), Mass values monoisotopic, Peptide mass tolerance 3 Da, Fragment mass tolerance ±0.8 Da, Maximum missed cleavages 2, Instrument type Esi Trap. All peptide assignments where the Mascot ion score is greater than the homology score and where the Mascot expect score is smaller than 0.05 were considered as good hits for further analysis.
All peptides called as present by Mascot in the SAX, SCX, and the published datasets [47] were remapped against the Berkeley Drosophila melanogaster protein database (BDGP) release 3.2 for the purpose of extracting all the possible splice variants and alternative proteins belonging to a specific peptide. Furthermore, the proteins obtained in this way were then classified according to their information content into five different classes [48]. This information content considers the relationship between the protein sequences and gene models and minimizes protein inference errors due to ambiguous protein sequences.
The N-termini were extracted from the overall list of peptides. Two types of N-termini are distinguished: (i) N-termini that start in position 1 or 2 (after methionine cleavage) in at least one protein of all the possible alternatives for the respective peptide and (ii) potential alternative methionine starting sites, i.e. semi-tryptic peptides that start with methionine, or where the precedent amino acid is a methionine. This list was manually cured, excluding fully tryptic peptides that likely represent internal peptides (the peptides identified in the COFRADIC approach may be preceded by a Lys residue since they are modified and thus no longer cleaved by trypsin). Furthermore, possible redundancies with respect to N-terminus were reduced, i.e. cases where one N-terminus is a prefix of another N-terminus with the exception of partially acetylated N-termini. The entire list of N-termini is presented in Table S1.
Peptide sequences were mapped onto the genomic sequence in order to extract the DNA context information of the iMet (or of the starting ATG triplet). For this purpose, for each gene model of BDGP3.2 the protein coding sequence, the respective transcript, as well as the exon containing the potential alternative iMet were extracted and the Cavener and Kozak context was analyzed. A sequence (−4 to −1) for which up to 1 nucleotide was identical to the Cavener or Kozak consensus sequence was labeled as weak, 2–3 perfect nucleotide matches were labeled as adequate, and identical sequences were labeled as strong (see Text S1).
Weblogos were created using the library weblogo (http://code.google.com/p/weblogo/) in Python in order to represent the relative frequency and information content of different nucleotides at varying positions around the Kozak/Cavener sequence and bar plots were created using R (www.r-project.org).
Pfam (www.sanger.ac.uk/Software/Pfam/) was used to identify functional domains in the experimentally identified protein sequences. For this purpose, the set of 16,743 distinct protein sequences of BDGP3.2 were searched against the Pfam database of Hidden Markov Models (release July 23, 2008, 10,340 protein family models) using the HMMER software package (hmmer2.3.2, http://hmmer.janelia.org/).
The Pfam results were compared then with the results of acetylated seen protein sequences for any over- or underrepresentation and Fisher's exact test with multiple testing correction was performed for selected Pfam categories.
To assess any possible association between the acetylation status of a group of proteins and different GO terms, all possible gene models were extracted from Table 1 and were submitted to a GO Slim analysis (http://go.princeton.edu/cgi-bin/GOTermMapper). This tool uses the map2slim.pl script (Chris Mungall, BDGP) to bin the submitted gene lists to a static set of broader, high-level parent generic GO-slim terms.
The GO analysis was performed for “Biological Process,” “Cellular Component,” and “Molecular Function,” and Fisher's exact test, with multiple testing correction, was performed for each GO term to provide a p value for the respective over- or underrepresentation.
8 mg of the Kc167 cell cytosolic fraction was first dissolved in 1 ml of 6 M guanidinium hydrochloride in 50 mM phosphate buffer (pH 7.5), after which the guanidinium hydrochloride concentration was lowered to 4 M by adding 0.5 ml of 50 mM phosphate buffer (pH 7.5). Proteins were reduced and alkylated by adding TCEP and iodoacetamide to final concentrations of 50 mM and 100 mM, respectively (reaction was allowed for 60 min at 30°C). Then, excess reagents were removed by desalting the protein mixture over a 10 ml Zeba Spin Desalting Column (Pierce, Thermo Fisher Scientific, Erembodegem, Belgium). Proteins were collected in 1 ml of 2 M guanidinium hydrochloride in 50 mM phosphate buffer (pH 8). Free primary amines (protein α-N-termini and ε-amines of lysines) were acetylated using an N-hydroxysuccinimide ester of trideutero-acetate (prepared according to [49]). Here, 5.8 mg (about 36 µmol) of this NHS-ester was dissolved in 20 µl DMSO and then added to the protein sample. Acetylation was allowed for 90 min at 30°C, after which O-acetylation of Ser, Thr, and Tyr side-chains was reversed by adding 10 µl of 50% of hydroxylamine to the protein solution and incubation for 20 min at room temperature.
This modified protein mixture was desalted as indicated above, but now eluted in 50 mM of NH4HCO3 (pH 8, freshly prepared). The protein mixture was boiled for 5 min, placed on ice for another 5 min, and then overnight digested by adding 40 µg of sequencing-grade modified trypsin (Promega Corporation, Madison, WI, USA). Small particulate matter was removed by centrifugation for 10 min at 10,000 g. Half of the digest (500 µl) was loaded on a strong anion exchange (SAX) column (SAX-Zirchrom 2.1 mm I.D.×150 mm length; 3 µm particles, ZirChrom Separations, Anoka, MN, USA) after which a binary solvent gradient was applied to fractionate peptides. Solvent A consisted of 10 mM Tris-HCl (pH 8) in 25% acetonitrile and solvent B was 1 M NaCl in 10 mM Tris-HCl (pH 8) and 25% acetonitrile. Under a constant flow of 70 µl/min, the following gradient was applied: 0% solvent B (0–50 min), 10% solvent B (55 min), 50% solvent B (100 min), washing with 50% solvent B until 105 min, 100% solvent B (107 min), followed by a 5 min wash with solvent B, 100% solvent A (114 min), and re-equilibration with solvent A (170 min). SAX-separated peptides were collected in 2 min wide fractions between 10 and 130 min and were finally pooled into six fractions in such a way that each fraction contained similar amounts of total peptide material (judged from the UV absorbance trace).
It was previously demonstrated that strong cation exchange (SCX) chromatography at acidic pH can be used to enrich N-terminal peptides from whole proteome tryptic digests [14],[22]. These peptides are in fact less well retained by the SCX column as other peptides and are thereby segregated from the bulk of non-N-terminal peptides. One disadvantage, however, is that N-terminal peptides containing histidine will still be retained by the SCX column and are thus largely absent from the final lists of reported identifications [22].
Here, SCX was used to enrich N-terminal peptides from digests of the cytoplasmic, nuclear, and membrane fraction of Kc 167 cells prior to COFRADIC. Full experimental details can be found in [22], and in brief, protein preparations were first reduced and S-alkylated with iodoacetamide, then all free primary amines were blocked by trideutero-acetylation, and finally proteins were digested into peptides using trypsin. This peptide mixture was then applied on a SCX column at pH 3 and blocked N-terminal peptides were found enriched in the non-binding fraction. These peptides were then further enriched by N-terminal COFRADIC (see below).
All six SAX fractions and the SCX-enriched proteome digests of the Kc167 cell cytosolic, microsomal, and nuclear fractions were then further analyzed by the N-terminal COFRADIC technology [20] as described in [22]. Such isolated peptides were then further analyzed by LC-MS/MS analysis on either a Bruker Esquire HCT ion trap (SAX peptides, ion trap operated as described in [50]) or an LTQ linear iontrap operated as described in [47] or an Agilent XCT-Ultra IT mass spectrometer equipped with Agilent's Chip Cube (SCX peptides, ion trap operated as described in [51]). Following analysis, raw MS/MS-spectra were converted to mgf files, which were used to identify the corresponding peptides using the MASCOT database search tool [52], in-house installed. The following parameters were set. Carbamidomethylation of Cys and trideutero-acetylation of Lys were set as fixed modifications. Acetylation or trideutero-acetylation of peptide N-termini, carbamylation of Lys (only for SAX-separated peptides), deamidation of Asn and Gln, oxidation of Met (sulfoxide), N-terminal pyrocarbamidomethyl Cys, and N-terminal pyroglutamic acid were all set as variable modifications. The mass tolerances for both the precursor ion and fragment ions were set to ±0.5 Da. The enzyme was set at endoproteinase Arg-C and cleavage between Arg-Pro and one missed cleavage were allowed. MASCOT's instrument setting was ESI-TRAP. The BDGP 3.2 database was searched as well truncated versions of this database that anticipate for protein processing events and were previously found to increase the number of identifications. These truncated versions were made using the DBToolkit algorithm [53]. The raw DAT result files of Mascot were then queried using in-house developed software tools. Only MS/MS-spectra receiving a score exceeding Mascot's identity threshold score at the 95% confidence level were kept. Such identified peptides were then automatically stored in a MySQL relational database (see http://genesis.ugent.be/ms_lims/) in which links were made to their MS/MS-spectra and precursor proteins.
|
10.1371/journal.pntd.0007403 | PCR-RFLP analyses of Leishmania species causing cutaneous and mucocutaneous leishmaniasis revealed distribution of genetically complex strains with hybrid and mito-nuclear discordance in Ecuador | PCR-Restriction Fragment Length Polymorphism (RFLP) analyses targeting multiple nuclear genes were established for the simple and practical identification of Leishmania species without using expensive equipment. This method was applied to 92 clinical samples collected at 33 sites in 14 provinces of Ecuador, which have been identified at the species level by the kinetoplast cytochrome b (cyt b) gene sequence analysis, and the results obtained by the two analyses were compared. Although most results corresponded between the two analyses, PCR-RFLP analyses revealed distribution of hybrid strains between Leishmania (Viannia) guyanensis and L. (V.) braziliensis and between L. (V.) guyanensis and L. (V.) panamensis, of which the latter was firstly identified in Ecuador. Moreover, unexpected parasite strains having the kinetoplast cyt b gene of L. (V.) braziliensis and nuclear genes of L. (V.) guyanensis, L. (V.) panamensis, or a hybrid between L. (V.) guyanensis and L. (V.) panamensis were identified. This is the first report of the distribution of a protozoan parasite having mismatches between kinetoplast and nuclear genes, known as mito-nuclear discordance. The result demonstrated that genetically complex Leishmania strains are present in Ecuador. Since genetic exchanges such as hybrid formation were suggested to cause higher pathogenicity in Leishmania and may be transmitted by more species of sand flies, further country-wide epidemiological studies on clinical symptoms, as well as transmissible vectors, will be necessary.
| Leishmaniasis caused by intracellular protozoa of the genus Leishmania is a neglected tropical disease widely distributing worldwide, especially in tropical and subtropical areas. Approximately 20 species are known to be pathogenic to humans, of which eight species have been recorded as causative agents of cutaneous and mucocutaneous leishmaniases in Ecuador. Since infecting species are the major determinant of clinical outcomes, identification at the species level is important for the treatment and prognosis. The parasite species have been identified conventionally by multilocus enzyme electrophoresis (MLEE) and recently by genetic analysis such as sequencing and genotyping. In the present study, PCR-Restriction Fragment Length Polymorphism (RFLP) targeting multiple nuclear genes was employed, and the results were compared with those obtained by kinetoplast cytochrome b (cyt b) gene sequence analysis, which is widely applied to species identification. Although most results corresponded between the two analyses, PCR-RFLP revealed presence of unexpected genetically complex Leishmania strains having characteristics of hybrid and mito-nuclear discordance. Since hybrid strains of Leishmania were suggested to increase disease severity and may be transmitted by a wider range of sand fly species, careful epidemiological research, including clinical courses and vector research, will be needed.
| Leishmaniasis, caused by protozoan parasites of the genus Leishmania, is a neglected tropical disease widely distributed worldwide, especially in tropical and subtropical areas, affecting at least 12 million people in 96 countries [1]. Approximately 20 Leishmania species belonging to the subgenera Leishmania (Leishmania), Leishmania (Viannia) and Leishmania (Mundinia) are pathogenic to humans [1, 2]. Since infected parasite species is known to be the major determinant of clinical outcomes in leishmaniasis [1], identification of the causative parasite is important for appropriate treatment and prognosis.
Leishmania species have been classified conventionally by multilocus enzyme electrophoresis (MLEE) [3, 4]. Genetic analysis of kinetoplast and nuclear targets, such as cytochrome b (cyt b), cysteine protease (cpb), heat shock protein 70 (hsp70) genes and the internal transcribed spacer (ITS) regions of ribosomal RNA, has commonly been used for species identification due to its sensitivity, simplicity and reliability [5–13]. In addition, a simple PCR-Restriction Fragment Length Polymorphism (RFLP), which does not require costly equipment, was developed for species identification, and the ITS region and hsp70 gene are widely applied to epidemiological studies [11, 14–19].
In Ecuador, leishmaniasis is endemic in Pacific coast, Andean highland, and Amazonian areas, and eight species, Leishmania (Leishmania) mexicana, L. (L.) amazonensis, L. (L.) major-like, L. (Viannia) guyanensis, L. (V.) panamensis, L. (V.) braziliensis, L. (V.) naiffi, and L. (V.) lainsoni, have been recorded as causative agents of cutaneous leishmaniasis (CL) and mucocutaneous leishmaniasis (MCL) [8, 20, 21]. Of these, distribution of L. (L.) amazonensis and L. (L.) major-like have been reported to be localized, and infections by them have not been reported recently [8, 21]. Infection by L. (V.) guyanensis together with its closely-related species, L. (V.) panamensis, has been identified from CL patients in Pacific coast areas by MLEE [21–24]; however, our recent cyt b gene analysis revealed a wide range distribution of L. (V.) guyanensis, without detecting any L. (V.) panamensis in these areas [8]. These results suggest that endemic species may change, or the reported results may be caused by the discordance between the MLEE analysis and kinetoplast cyt b gene analysis employed for species identification. Recently, a countrywide epidemiological study was carried out based on the cyt b sequence analysis and it identified L. (V.) guyanensis and L. (V.) braziliensis widely in Pacific coast and Amazonian areas and L. (L.) mexicana in Andean high lands as current major causative species in Ecuador [8]. Additionally, L. (V.) naiffi and L. (V.) lainsoni were recently recorded in Amazonian areas [8, 20, 25].
In this study, a simple and practical method for the identification of Leishmania species in Ecuador was established on the basis of PCR-RFLP analyses targeting mannose phosphate isomerase (mpi) and 6-phosphogluconate dehydrogenase (6pgd) genes, and the result was compared with that obtained by the cyt b gene sequence analysis. This study demonstrated the presence of genetically complex Leishmania strains in Ecuador, and strongly suggested the importance of applying multiple target approaches to enhance the reliability of species identification and to characterize more detailed genetic properties of the parasite.
Frozen stocks of 24 parasite strains of five Leishmania species [L. (V.) guyanensis, L. (V.) panamensis, L. (V.) braziliensis, L. (L.) major-like, L. (L.) mexicana] that were isolated from CL patients in Ecuador and identified at the species level by MLEE [22–24] (Table 1) were spotted on an FTA Classic Card (Whatman, Newton Center, MA) and subjected to sequence analysis. Three strains of L. (V.) naiffi identified by cyt b gene analysis [25, 26] were also utilized (Table 1).
Most of the clinical samples employed in this study were collected from patients suspected of CL in the previous study [8, 20], and each 3 samples newly obtained from Provinces of Manabi and Santo Domingo de los Tsachilas, all of which were identified as L. (V.) guyanensis by the cyt b gene analysis, were included in this study. Leishmania parasites were identified on the basis of cyt b sequence analysis [8, 20]. The samples were collected at 33 sites in 14 provinces of Ecuador (S1 Fig). Residual tissue materials were spotted onto an FTA Classic Card, after taking scraped margin samples of active lesions for routine diagnosis. Two-mm-diameter disks of FTA card were punched out from each filter paper, washed three times with an FTA Purification Reagent (Whatman), and subjected to PCR amplification.
PCR primers for amplification of cyt b, hsp70, mannose phosphate isomerase (mpi) and 6-phosphogluconate dehydrogenase (6pgd) gene fragments were designed based on the sequence regions conserved among species (Table 2). PCR amplification with a pair of outer primers was performed with 30 cycles of denaturation (95°C, 1 min), annealing (55°C, 1 min) and polymerization (72°C, 2 min) using Ampdirect Plus reagent (Shimadzu Biotech, Tsukuba, Japan). Each 0.5-μl portion of the PCR product was reamplified with inner primers under the same condition described above. The products were cloned into the pGEM-T Easy Vector System (Promega, Madison, WI) and sequences were determined on both strands by the dideoxy chain termination method using a BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Foster City, CA). Primers for amplification of a partial sequence of the kinetoplast cytochrome oxidase subunit II-NADH dehydrogenase subunit I region (COII-ND1) were also designed based on the sequences conserved among species (Table 2). The COII-ND1 sequences were determined on both strands by direct sequencing with inner primers, L.COII-2S and L.COII-2R. Restriction enzyme mapping was performed in silico by using BioEdit Sequence Alignment Editor to obtain species-specific RFLP patterns.
Clinical samples spotted on FTA cards, in which parasites were identified by cyt b gene analysis in a previous study, were subjected to PCR-RFLP analysis. PCR amplifications targeting mpi and 6pgd were performed as described above using a high fidelity DNA polymerase, KOD plus (Toyobo, Osaka, Japan). The PCR products were digested by restriction enzymes HaeIII, HapI, and BstXI for the mpi gene and Bsp1286I and HinfI for the 6pgd gene, and resulting restriction fragment patterns were analyzed by 2% agarose gel electrophoresis. GeneRuler 100 bp Plus DNA Ladder (Thermo Fisher Scientific, Waltham, MA) was used as a DNA size marker. The gel was stained with GelRed Nucleic Acid Gel Stain (Biotium, Hayward, CA), and DNA fragments were visualized with UV transilluminator.
Differentiation between L. (V.) guyanensis and L. (V.) panamensis was performed by restriction enzyme-digestion of the hsp70 gene fragment [27]. Briefly, the hsp70 gene fragment was amplified by a nested PCR using sets of outer primers (L.HSP-Ty1S and L.HSP-OR) and inner primers (L.HSP-Ty2S and L.HSP-IR2) (Table 2). The amplicons were digested with a restriction enzyme, BccI, and resulting fragment patterns were analyzed by 3% agarose gel electrophoresis.
Clinical samples were collected by local physicians and well-trained laboratory technicians of health centers of the Ministry of Health, Ecuador. For routine parasitological diagnosis, scratching smear samples of skin lesions were taken from suspected leishmaniasis patients at health centers. In this study, only residual tissue materials were collected after the routine procedure to minimize the burden on patients. Signed consent was obtained from the adult subjects and from the children’s parents or guardians, prior to the diagnostic procedures at each health center of the Ministry, providing information on the process of diagnosis and Leishmania species analysis, following the guidelines of the Ethics Committee of the Ministry. The subjects studied were volunteers in routine diagnosis/screening and treatment programs promoted by the Ministry. All routine laboratory examinations were carried out free of charge, and treatment with specific drug, meglumine antimoniate (Glucantime) was also offered free of charge at each health center. The study was approved by the ethics committee of the Graduate School of Veterinary Medicine, Hokkaido University (approval number: vet26-4) and Jichi Medical University (approval number: 17–080) [8].
Leishmania cyt b, hsp70, mpi and 6pgd partial gene sequences were amplified from 27 strains of 6 species isolated in Ecuador. Sequences of these fragments showed high degrees of homology (88–100%, 82–100%, 83–100% and 94–100% in cyt b, mpi, 6pgd and hsp70 genes, respectively) with corresponding leishmanial genes registered in GenBank. The restriction enzyme mapping was performed in silico to see if species-specific enzyme sites could be found in cyt b, mpi, 6pgd and hsp70 gene fragments obtained in this study. Species-specific RFLP patterns could not be obtained for the cyt b gene because of intraspecies genetic variations through the sequences. On the hsp70 gene, restriction enzymes to differentiate Leishmania species were found; however, RFLP patterns including several smaller fragments (< 300 bp) were similar among species. Therefore, it seems difficult to identify the species based on RFLP patterns of hsp70 using agarose gel electrophoresis in some cases because of the resolution. On the other hand, restriction enzyme sites that can differentiate Leishmania species in Ecuador were identified in mpi and 6pgd genes, except for two very closely-related species, L. (V.) guyanensis and L. (V.) panamensis. Different RFLP patterns were obtained in L. (V.) guyanensis/L. (V.) panamensis, L. (V.) braziliensis/L. (V.) naiffi, L. (L.) major-like and L. (L.) mexicana for digested mpi gene fragments with a restriction enzyme HaeIII (Fig 1A). Although an RFLP polymorphism was observed in one (strain PT27) of seven L. (L.) mexicana strains, it did not affect species identification (Table 3). L. (V.) braziliensis and L. (V.) naiffi, showing the same RFLP patterns as HaeIII digestion, were differentiated by HpaI digestion (Table 3, Fig 1B). Although L. (V.) lainsoni, a recently reported species in the Ecuadorian Amazon [20], showed the same RFLP patterns as L. (V.) guyanensis/L. (V.) panamensis when digested with HaeIII and HpaI, BstXI-digestion successfully differentiated it from L. (V.) guyanensis/L. (V.) panamensis, as reported in Peruvian strains (S2 Fig) [28].
Digestion of the 6pgd gene with Bsp1286I resulted in distinct gene fragment patterns of L. (V.) guyanensis/L. (V.) panamensis, L. (V.) braziliensis, L. (V.) naiffi, L. (L.) major-like and L. (L.) mexicana; however, the patterns between L. (V.) guyanensis/L. (V.) panamensis and L. (V.) naiffi were similar and difficult to discriminate because of only about a 50 bp difference in a fragment of approximately 1 kbp (Fig 2A). The two species were successfully differentiated by digesting with HinfI (Fig 2B).
Although L. (V.) guyanensis and L. (V.) panamensis were not discriminated by PCR-RFLP of mpi and 6pgd genes, PCR-RFLP of the hsp70 gene with a restriction enzyme, BccI, successfully differentiated the two species as reported previously (Fig 3) [27].
PCR-RFLP analyses of mpi gene with restriction enzymes, HaeIII and HpaI, and 6pgd gene with Bsp1286I and HinfI were applied to 92 clinical samples collected at 33 sites in 14 provinces of Ecuador. PCR-RFLP analysis of the hsp70 gene with a restriction enzyme, BccI, was used for differentiation between L. (V.) guyanensis and L. (V.) panamensis. The results obtained by PCR-RFLP analyses were compared with those obtained by the cyt b gene sequence analysis. The results of the species identification obtained by the two nuclear genes always agreed with each other. The identification by PCR-RFLP analyses completely matched with that obtained by the cyt b gene sequence analysis in all of L. (V.) naiffi (2 samples) and L. (L.) mexicana (3 samples) (Table 4). Of the 73 samples identified as L. (V.) guyanensis by cyt b gene analysis, 72 samples were identified as L. (V.) guyanensis by PCR-RFLP analyses, whereas one sample from a Pacific coast area showed a hybrid pattern of L. (V.) guyanensis and L. (V.) panamensis based on the PCR-RFLP of the hsp70 gene (Figs 3 and 4). The sequence of the hsp70 gene fragment was analyzed by direct sequencing, and a single nucleotide polymorphism was confirmed, showing “C” in L. (V.) guyanensis but “T” in L. (V.) panamensis, whereas a sample having a hybrid RFLP pattern had both “C” and “T” peaks at the corresponding position (S3 Fig), indicating the presence of a hybrid strain of L. (V.) guyanensis and L. (V.) panamensis in Ecuador. On the other hand, of the 14 samples identified as L. (V.) braziliensis by cyt b gene analysis, only 6 samples were identified as L. (V.) braziliensis by RFLP analyses (Table 4). In the other 8 samples identified as L. (V.) braziliensis by the cyt b gene analysis, three samples showed hybrid patterns in PCR-RFLP analyses of both the mpi and 6pgd genes (Fig 5A and 5B). The sequences of mpi and 6pgd gene fragments were analyzed by direct sequencing, and a single nucleotide polymorphism was confirmed, showing “C” in L. (V.) guyanensis but “T” in L. (V.) braziliensis of the mpi gene, and “T” in L. (V.) guyanensis but “C” in L. (V.) braziliensis of the 6pgd gene. On the other hand, the mpi and 6pgd genes from the three samples with hybrid RFLP patterns had both “C” and “T” peaks at the corresponding position (S4 Fig). From these results, the parasite species of these three samples were identified as a hybrid of L. (V.) braziliensis and L. (V.) guyanensis (Table 4, Fig 4). In the remaining 5 samples identified as L. (V.) braziliensis by sequence analysis of the cyt b gene, PCR-RFLP analyses showed that one sample from a Pacific coast area was L. (V.) guyanensis, three samples from the northern Pacific coast and Amazonian areas were L. (V.) panamensis, and one sample from a northern Pacific coast area had a hybrid pattern of L. (V.) guyanensis and L. (V.) panamensis (Table 4, Fig 4). The sequence analyses of mpi, 6pgd, and hsp70 gene fragments corresponded to PCR-RFLP analyses, indicating the presence of a mismatch between kinetoplast and nuclear genes, known as mito-nuclear discordance, in Leishmania distributing in Ecuador (Table 4, Fig 4). To further confirm the mito-nuclear discordance, partial sequences of the COII-ND1 region were analyzed as another target of kinetoplast genes in samples showing a mismatch between kinetoplast cyt b gene and nuclear mpi, 6pgd and hsp70 genes. The sequences were compared to each two corresponding sequences obtained from L. (V.) braziliensis and L. (V.) guyanensis in this study since this region has not been well-analyzed in subgenus Viannia species. The sequences from parasites with mito-nuclear discordance showed 98.9–99.1% and 98.5–98.9% identities with those of L. (V.) braziliensis and L. (V.) guyanensis, respectively (accession numbers: LC475135-LC475142). When partial COII gene sequences in the obtained COII-ND1 region sequences were analyzed on the GenBank database, the sequences from parasites with mito-nuclear discordance showed 99.5% and 98.9% identities with those of L. (V.) braziliensis and L. (V.) guyanensis, respectively. This result strongly suggested that the kinetoplast genes of these parasites originated from L. (V.) braziliensis, corresponding to the result of cyt b gene analysis.
In the present study, PCR-RFLP analyses were employed for the identification of Leishmania species distributing in Ecuador in order to develop a simple and practical way for species identification independent of expensive equipment such as a genetic analyzer. As a result, mpi and 6pgd genes, for which encoding enzymes have been widely used as the gold standard of species identification, were identified as suitable targets for this purpose in the tested samples. The results obtained by the PCR-RFLP analyses of multiple nuclear targets were compared to those of cyt b gene sequence analysis [7, 8, 29–36]. Although most results corresponded between the two analyses, PCR-RFLP revealed distribution of hybrid and mito-nuclear discordant Leishmania strains, which could not be identified only by cyt b gene sequence analysis. The results indicated that Leishmania strains distributing in Ecuador are genetically more complex than previously thought.
PCR-RFLP analysis has been employed for species identification of Leishmania species, and its utility is widely accepted [34]. The rRNA internal transcribed spacer 1 (ITS-1) region and hsp70 gene are mostly used as suitable target genes, of which the former is applied mainly in the Old World [6, 11, 12, 14, 17, 19, 27, 34, 37–41]. Although the hsp70 gene is one of the most valuable genetic markers for PCR-RFLP-based species identification, intraspecific polymorphism of RFLP patterns and very similar RFLP profiles among species, which affect species identification, have been reported in some Leishmania species [42]. In this study, other nuclear genes, mpi and 6pgd genes, for which encoding enzymes have been used for MLEE, were shown to be alternative useful targets for classification by PCR-RFLP analysis. Of these, the mpi gene was reported to be the only genetic marker that can distinguish two very closely-related species, L. (V.) braziliensis and L. (V.) peruviana [7, 43, 44]. In addition, a recent study demonstrated that PCR-RFLP of the shorter mpi gene fragment (approximately 500 bp) can differentiate 4 Leishmania species [L. (V.)braziliensis, L. (V.) peruviana, L. (V.) guyanensis, and L. (V.) lainsoni] and a hybrid of L. (V.) braziliensis and L. (V.) peruviana circulating in the Department of Huanuco, Peru [28]. In the present study, PCR-RFLP analyses of longer mpi and 6pgd gene fragments (>1000bp) were successfully established and applied to 92 clinical samples in Ecuador. Although a polymorphic RFLP pattern, which does not affect the identification, was detected in the mpi of one L. (L.) mexicana strain, the variant RFLP pattern was not detected in the present clinical samples identified as L. (L.) mexicana. Further sample analyses from different areas and different countries will be important to confirm the utility of this analysis, although polymorphic RFLP profiles may be detectable in these genes. Since polymorphism was also reported in the hsp70 gene of several Leishmania species [42], PCR-RFLP analyses of multiple target genes, rather than single nuclear or kinetoplast genes, will result in more accurate species identification and disclose more detailed genetic characteristics of the parasite.
Several samples showing hybrid RFLP patterns were identified as hybrid strains rather than mixed infection of different Leishmania species. It is due to the following reasons: 1) It is little or no chance to be infected by more than one parasite in a cutaneous lesion because the lesion is typically developed at the site bitten by a sand fly transmitting specific Leishmania species, 2) Even if mixed infection occurs, either parasite becomes dominant in the lesion, resulting in the presence of dominant allele by the genetic analysis. However, both alleles were comparably amplified as observed in the PCR-RFLP analysis, which is indicative of a putative hybrid strain. In addition, similar results were obtained on electrograms of the direct sequencing, showing comparable fluorescence intensities of polymorphic nucleotides derived from both species. 3) The presence of hybrid strain has been reported in the same area as described below [45]. Isolation of putative hybrid strains as a culture is necessary for further detailed characterization of these parasites.
Although multiple PCR-RFLP and cyt b sequence analyses showed corresponding results in most clinical samples, the present study revealed the distribution of several unexpected strains in Ecuador, including hybrid and mito-nuclear discordance strains. Since hybrid strains cannot be identified by the cyt b gene analysis after molecular cloning, this is another advantage of identifying parasite species by PCR-RFLP. Distribution of a hybrid strain of L. (V.) guyanensis/panamensis complex and L. (V.) braziliensis was reported in Zumba, a province of Zamora-Chinchipe in a southern part of Ecuador by using MLEE and random amplified polymorphic DNA (RAPD) [45]. The present study confirmed the presence of the hybrid strain in Zumba, and also in another area in the same province, Palanda. In addition, a hybrid of L. (V.) guyanensis and L. (V.) panamensis was detected in northern Pacific areas of Ecuador. This is the first report of the presence of a hybrid strain of L. (V.) guyanensis and L. (V.) panamensis in Ecuador. L. (V.) guyanensis and its closely related L. (V.) panamensis have been reported to be endemic in northern Pacific areas of Ecuador by MLEE; however, only L. (V.) guyanensis was identified in the same areas by cyt b gene analysis in recent studies [8, 21, 46]. The present study confirmed that L. (V.) guyanensis is dominantly present in these areas, suggesting that endemic species may change, or that there may be discordance between MLEE and genetic analysis. However, the identification of a hybrid of L. (V.) guyanensis and L. (V.) panamensis as a minor population suggests that parental L. (V.) panamensis may still be present in some of these areas. Another unexpected finding was identification of mito-nuclear discordant strains of Leishmania species in northern Pacific and Amazonian areas. Interestingly, mito-nuclear discordant strains were identified only in the species identified as L. (V.) braziliensis by cyt b gene analysis. This finding supports a recent study using cyt b gene analysis reporting increasing cases of L. (V.) braziliensis infection in Pacific coast areas when compared to previous studies using enzymatic MLEE analysis [8]. The hybrid strain of L. (V.) braziliensis and L. (V.) peruviana was suggested to increase disease severity when compared to parental species in an animal model [47]. Therefore, careful investigation is needed to clarify the presence of hybrid strains, including mito-nuclear discordance, and their effects on clinical courses. In addition, hybrid strains may increase the range of transmissible sand fly species if they have a potential to be transmitted by both vector species of parental parasites. Continuous vector research is important in these endemic areas, as well as parasitological and clinical studies. Further, basic parasitological research on how genetic exchange and mito-nuclear discordance occur among Leishmania species would be another interesting subject [48–51]. Mito-nuclear discordance is reported in various animals such as mammals, birds, reptiles, amphibians, fish and insects, and is inferred to result from various processes: 1) adaptive introgression of mitochondrial DNA, 2) demographic disparities, 3) sex-biased asymmetries, 4) hybrid zone movement, 5) an intracellular bacteria, Wolbachia infection in insects, and 6) human actions [52]. It provides deeper insights into the phylogenetic relationship, population structure, and evolutionary signature of these animals. Mito-nuclear discordance is also reported in helminth parasites: trematodes Schistosoma turkestanicum between populations [53], and cestodes Taenia solium between lineages [54], and between T. saginata and T. asiatica [55–57]. This is the first report of mito-nuclear discordance in protozoan parasites. Mito-nuclear discordance is speculated to be resulted from the similar process as hybridization of nuclear genes in protozoa. Further study is needed to disclose the mechanism of mito-nuclear discordance formation in protozoa. In addition, association of mito-nuclear discordance with the pathogenicity and vector competency of the parasites is important issues to be clarified. In this study, we established a novel PCR-RFLP-based genotyping approach to identify Leishmania species in Ecuador. Although the present PCR-RFLP analyses was shown to be practical for identification of Leishmania species in Ecuador, further study focusing on other Leishmania species and clinical samples from different countries will be needed to enhance the utility of this approach. PCR-RFLP analyses of clinical samples and subsequent comparison with kinetoplast cyt b sequence analysis revealed the distribution of genetically complex Leishmania strains having genetic characteristics of hybrid and mito-nuclear discordance. Although intraspecies genetic variation observed in the cyt b gene resulted in this gene as an unsuitable target for RFLP analysis, there is no doubt about the utility of cyt b gene sequence analysis for species identification and phylogenetic analysis since distinct interspecies genetic diversity of this gene overcomes the disadvantage of the intraspecies variation. However, the present study points to the importance of applying multiple target approaches as the combination of cyt b and the PCR-RFLP assays presented here, enhancing the reliability of species identification and characterization of genetic properties including hybrid and mito-nuclear discordance. Further studies are needed to reveal the parasitological characteristics of hybrid and mito-nuclear discordance, clinical outcomes caused by these parasites, and the range of vector species of these parasites. In addition, studies on mito-nuclear discordance in Leishmania and other protozoa may provide further insights into the mechanism of genetic exchanges of these parasites.
|
10.1371/journal.ppat.1003493 | The Viral Chemokine MCK-2 of Murine Cytomegalovirus Promotes Infection as Part of a gH/gL/MCK-2 Complex | Human cytomegalovirus (HCMV) forms two gH/gL glycoprotein complexes, gH/gL/gO and gH/gL/pUL(128,130,131A), which determine the tropism, the entry pathways and the mode of spread of the virus. For murine cytomegalovirus (MCMV), which serves as a model for HCMV, a gH/gL/gO complex functionally homologous to the HCMV gH/gL/gO complex has been described. Knock-out of MCMV gO does impair, but not abolish, virus spread indicating that also MCMV might form an alternative gH/gL complex. Here, we show that the MCMV CC chemokine MCK-2 forms a complex with the glycoprotein gH, a complex which is incorporated into the virion. We could additionally show that mutants lacking both, gO and MCK-2 are not able to produce infectious virus. Trans-complementation of these double mutants with either gO or MCK-2 showed that both proteins can promote infection of host cells, although through different entry pathways. MCK-2 has been extensively studied in vivo by others. It has been shown to be involved in attracting cells for virus dissemination and in regulating antiviral host responses. We now show that MCK-2, by forming a complex with gH, strongly promotes infection of macrophages in vitro and in vivo. Thus, MCK-2 may play a dual role in MCMV infection, as a chemokine regulating the host response and attracting specific target cells and as part of a glycoprotein complex promoting entry into cells crucial for virus dissemination.
| Several human herpesviruses form alternative gH/gL complexes which determine the tropism for different cell types. For murine cytomegalovirus (MCMV), a gH/gL/gO complex has recently been characterized. Here, we present the identification and characterization of an alternative gH/gL/MCK-2 complex which promotes MCMV spread and is important for efficient infection of macrophages in vitro and in vivo. Association of the MCMV CC chemokine MCK-2 with a glycoprotein complex promoting virus entry is a novel function for the well-characterized MCK-2. Virus mutants lacking MCK-2 have been shown to exhibit a reduced capacity to attract leukocytes and a disregulated T cell control of the MCMV infection in vivo. These defects can be attributed to the chemokine function of MCK-2. Yet, the observation that MCK-2 knock-out mutants additionally are impaired in infecting leukocytes in vivo is consistent with our new finding that MCK-2 forms a glycoprotein complex promoting entry into monocytic cells. gH/gL complexes associating with multifunctional proteins add a new level of complexity to the interpretation of infection phenotypes of the respective knock-out herpesviruses.
| Herpesviruses enter their host cells either by fusion of the viral envelope with the plasma membrane or with membranes of endocytotic vesicles. The fusion process is promoted by a concerted action of the conserved viral glycoproteins gB, gH, and gL [1] of which gH and gL consistently form a tight heterodimer [2], [3]. These three glycoproteins can promote receptor recognition and subsequent fusion as has been shown for the entry of Epstein-Barr virus (EBV) into epithelial cells [1]. Often, gB and gH/gL are not sufficient to promote receptor recognition. For instance, entry may depend on further envelope glycoproteins, as has been shown for gD of Herpes simplex virus [1], or on gH/gL forming tight complexes with additional viral proteins, as has for example been shown for the gH/gL/gp42 complex of EBV [4]–[6], the gH/gL/Q1/Q2 complex of HHV6 [7], or the gH/gL/pUL(128,130,131A) complex of HCMV [8]–[10].
For HCMV, two gH/gL complexes have been identified. In vitro, formation of gH/gL/gO ensures efficient production of infectious supernatant virus and promotes entry into a restricted set of cells by fusion at the plasma membrane [11], [12]. In the absence of gO, HCMV spreads in a cell-associated manner [11]. A restriction of cell tropism for mutants lacking gO has not been observed. The second complex, gH/gL/pUL(128,130,131A) promotes entry into a broad range of HCMV host cells including endothelial, epithelial, and dendritic cells [13], [14] by using endocytotic pathways [15]–[17]. Data published recently strongly suggest that gH/gL/gO and gH/gL/pUL(128,130,131A) promote virus entry through distinct cellular receptors [18], [19]. Depending on the HCMV strain analyzed, gO has been found to be incorporated into the virion or not [20]–[22]. The UL128, UL130 and UL131A gene products have consistently been shown to be incorporated into the virion [8], [9], [20], [21], [23], [24]. Their precise functions in the entry process have not yet been determined. It is also not known what the exact functions of gH/gL/gO and gH/gL/pUL(128,130,131A) are in the infection of humans. In a recent publication, we could show that the gH/gL complexes of HCMV are distributed to distinct virus populations which consequently differ in their cell tropism. In vitro, host cells like fibroblasts and endothelial cells either released or retained the population promoting infection of endothelial cells. We have proposed that, by determining the target cells of their virus progeny, host cells may route infection in vivo [23].
Infection of mice with MCMV serves as an animal model for the HCMV infection. We have recently identified the m74 ORF of MCMV as a functional homolog of HCMV gO [15]. Although the MCMV genome does not contain sequence homologs for HCMV UL128, UL130, and UL131A, the relative positions of the MCMV m130, m131/129, and m133 ORFs within the MCMV genome are comparable to the positions of the UL128, UL130, and UL131A ORFs in the HCMV genome. The m130 gene product has not been characterized. Deletion of m133 has been shown to result in reduced virus growth in salivary glands in vivo [25], [26]. The ORFs of m131 and m129 are fused by a splicing event which results in a protein product designated MCK-2 [27], [28]. The m131-derived part of MCK-2 contains, like the UL128 protein of HCMV, a CC (ß) chemokine domain. Besides that, MCK-2 does not show further sequence homologies to the UL128 gene product. MCK-2 and synthetic peptides of the m131 ORF or the complete MCK-2 have been shown to attract monocytes confirming its predicted chemokine activity [29], [30]. When mice are infected with MCMV mutants lacking MCK-2 the most apparent phenotype is a reduced virus production in salivary glands [28], [31], [32]. MCK-2 knock-out mutants are impaired in recruiting leukocytes which might serve as vehicles for virus dissemination [29], [30], [32]. Some populations of the attracted leukocytes have been shown to control virus specific CD8+ T cell immunity [33]. Yet, these populations differ from the myelomonocytic cells which are infected at the site of virus entry [30], [33]. Notably, MCK-2 knock-out viruses have additionally been shown to exhibit a 10-fold lower capacity to infect attracted myelomonocytic leukocytes [30].
Here, we report a completely new role for MCK-2, namely, as part of a gH/gL/MCK-2 complex promoting entry into macrophages. This offers an explanation for the hitherto unexplained low infection capacities of MCK-2 knock-out viruses for leukocytes in vivo [30]. The gH/gL/MCK-2 complex can complement the function of the gH/gL/gO complex of MCMV with respect to virus spread in vitro and strongly increases the efficiency of MCMV in infecting macrophages in vitro and in vivo. We propose that MCK-2 might have a dual role in infection, one as a chemokine attracting cells regulating the host immune response or attracting MCMV target cells and one in infecting viral target cells promoting subsequent virus dissemination.
HCMV and MCMV lacking gO both show the same spread phenotype in vitro, namely, strongly reduced titers of infectious virus in supernatants of infected cells and a focal spread pattern. For HCMV, we could show that the residual focal spread of mutants lacking gO is dependent on the alternative gH/gL/pUL(128,130,131A) complex [15]. To find out whether MCMV also forms an alternative gH/gL complex, we infected cells with bacterial artificial chromosome (BAC)-derived wildtype MCMV and precipitated gH-associated proteins from extracts of virus released into the supernatant by using an antibody specific for MCMV gH [34]. The precipitates were separated on SDS-polyacrylamide gels, proteins extracted from gel slices and then analyzed by liquid chromatography-tandem mass spectrometry. The obtained peptides were compared to MCMV gene translations. One prominent hit was a LLCLVR peptide which matches the C-terminus of the m131 ORF which together with the m129 ORF forms the MCMV MCK-2 protein (data not shown).
On Western blots, MCK-2 appears as multiple glycosylated forms running between 30 and 45 kDa ([27] and (Fig. 1A)). When we prepared extracts of supernatant virus, MCK-2 ran at a slightly higher molecular weight than MCK-2 from extracts of infected cells (Fig. 1A). This points towards a differentially modified protein. A similar pattern has been shown for MCK-2 secreted from transfected cells [27]. MCK-2 could also be detected in extracts of gradient purified virus which strongly suggests that it is incorporated into virions (Fig. 1B). Under non-reducing conditions, MCK-2 migrated at a molecular weight of about 180 kDa (Fig. 1B), which argues for MCK-2 forming a tight complex with other viral proteins in virions. As there is no antibody available which recognizes MCMV gH in Western blots, we constructed an MCMV BAC which expresses a C-terminally HA-tagged gH (Fig. S1) which grew like wildtype virus (Fig. S2). gH-HA could easily be detected in extracts of supernatant virus (Fig. 1C). Under reducing conditions it migrated at the expected molecular weight of about 85 kDa [34]. When supernatant virus from cells infected with MCMV-gH-HA was analyzed under non-reducing conditions, two prominent high molecular weight bands, one running slightly above and one running below the 180 kDa marker could be detected for gH-HA. The upper band co-migrated with MCK-2 (Fig. 1C). Whereas an anti-gH antibody could precipitate all gH bands visible in extracts of supernatant virus, an anti-MCK-2 antibody specifically precipitated the band co-migrating with MCK-2 (Fig. 1C). This band very likely represents a gH/gL/MCK-2 complex. The prominent gH-HA positive band below 180 kDa could represent a gH/gL complex consisting of gH-HA and the 274 amino acid long gL. We could also show that the upper band represents a complex containing gH and MCK-2 by using an MCMV-gH-HA mutant which carries a disrupted MCK-2 ORF (MCMV-gH-HA/129stop) (Fig. S1). This mutant does not express MCK-2 (Fig. 1D, upper panel) and lacks the upper gH-HA band under non-reducing conditions (Fig. 1D, right lower panel). The protein extracts of the gH-HA/m129stop mutant were at least five times more concentrated than the extracts of the gH-HA virus which can be seen from the strength of the gH-HA band under reducing conditions and the lower gH-HA band under non-reducing conditions (Fig. 1D, lower panel). Thus, it could be excluded that the gH/MCK-2 band had escaped detection. To show that the anti-MCK-2 antibody specifically co-precipitates gH and to confirm the co-precipitation of MCK-2 which we had found by mass spectrometry analysis of proteins precipitated with an anti-gH antibody, we performed the reverse co-immunoprecipitation using an anti-MCK-2 antibody to precipitate MCK-2 associated proteins. An extract was prepared from supernatant virus of cells infected with MCMV-gH-HA and aliquoted. Anti-gH and anti-HA antibodies readily precipitated gH-HA from this extract (Fig. 1E). The anti-MCK-2 antibody clearly co-precipitated gH-HA, whereas a mouse control IgG antibody did not (Fig. 1E).
To show that gH/gL/gO and gH/gL/MCK-2 are alternative complexes, we used an MCMV BAC expressing HA-tagged gO. This BAC was generated by adding an HA-tag to the 3′ end of the m74 ORF and by introducing a duplication of the overlapping C-terminus of the m73 ORF (gN) to preserve the function of gN (gO-HA, Fig. S1). MCMV-gO-HA grew like wildtype virus in fibroblasts (data not shown). We have previously shown that HA-tagged gO expressed in the virus context forms a complex of more than 200 kDa which can be precipitated with anti-gH and anti-HA antibodies [15]. To show that this complex is different from the complex formed by gH and MCK-2, we compared extracts from supernatant virus from MCMV-gH-HA and MCMV-gO-HA. gO-HA could easily be detected in extracts of supernatant virus (Fig. 2A). Under reducing conditions it migrated at the expected molecular weight of about 70 kDa [15]. Under non-reducing conditions a weak band representing the gH/gL/gO complex could only be detected after a very long exposure of the Western blots (Fig. 2A). In Figure 2A extracts of supernatant virus from cells infected with either MCMV-gH-HA or MCMV-gO-HA are depicted side by side and stained for the HA-tag. To show the position of the gH/gL/gO complex more clearly, an anti-HA immunoprecipitation from extracts of infected cells was included [15]. The comparison of extracts from MCMV-gO-HA and MCMV-gH-HA shows that the gO and MCK-2 complexes clearly have different molecular weights. Additionally, anti-MCK-2 which had co-precipitated gH from lysates of supernatant virus (Fig. 1 E), did not co-precipitate gO-HA from extraxts of infected cells, whereas anti-gH antibodies clearly co-precipitated gO-HA (Fig. 2B). These findings strongly support that gH/gL/gO and gH/gL/MCK-2 are indeed distinct complexes. In total cell extracts (data not shown) and in extracts of supernatant virus infected with the gH-HA virus, the anti-HA antibody could not detect a complex corresponding to the gH/gL/gO complex (Fig. 1C and Fig. 2A), a failure which might be due to a loss of the accessibility of the HA-tag of gH when the gH/gL/gO complex is formed.
To analyze the role of MCK-2 within a glycoprotein complex promoting entry, we constructed BAC-derived MCMV mutants in which the m131/129 reading frame was disrupted by stop cassettes and which do not express MCK-2 (Fig. S1). MCK-2 knock-out mutants previously constructed by others have been extensively studied in vivo and shown to exhibit defects in recruitment of leukocytes, in virus dissemination, and in growth in salivary glands [28], [30], [32]. We have recently shown that an MCK-2 mutant of the MCMV strain Smith which was cloned as a BAC also showed a reduced growth in salivary glands in vivo [31]. None of these mutants has been reported to show attenuation in vitro [28], [29], [31]. We could confirm this for two clones of the 131stop mutant (Fig. 3). Multistep growth curves even exhibited a marginal growth advantage for the MCK-2 knock-out mutants at days 4 and 5 after infection, yet, even three independent growth curves could not show that these differences were statistically significant.
For HCMV, it has been shown that the inability to form the alternative gH/gL/pUL(128,130,131A) complex abolishes the tropism for cells like endothelial and epithelial cells. Thus, we infected primary (MEF) and immortalized fibroblasts (NIH3T3), endothelial cells (MHEC-5T), and epithelial cells (TCMK-1) with wildtype virus and 131stop mutants and compared the infection capacities by staining the cells for expression of the immediate early 1 (IE1) protein of MCMV. The numbers of infected MEF cells were set to 100% and numbers of infected NIH3T3, MHEC-5T, and TCMK-1 were expressed as percent of MEF infection (Fig. 4A). No significant differences in infection capacities for fibroblasts or endothelial cells could be detected when wildtype and 131stop MCMV were compared. Only infection of TCMK-1 epithelial cells was slightly but significantly enhanced. As staining for expression of IE1 reflects successful entry, but not the ability to replicate in certain cell types, we also tested virus production of wildtype and 131stop viruses in these cell types but could not detect any differences (data not shown).
When MCK-2 knock-out mutants were analyzed in vivo, reduced capacities to infect myelomonocytic leukocytes were observed when compared to wildtype infections [30]. To find out whether this is also observed when closely related cells like macrophages are infected in vitro, we infected macrophage cell lines like ANA-1 or J774, primary bone marrow derived macrophages (BMDM), or macrophages directly from peritoneal exudates (PEC/M). For the latter, infection was studied for cells in the macrophage-enriched gate of peritoneal exudate cells (PEC) from untreated BALB/c mice ([35] and Fig. S3). ANA-1 cells were infected with wildtype virus, two clones of the 131stop mutant, and a gH-HA/129stop mutant (Fig. 4B). J774 cells were infected with wildtype virus, a 131stop mutant, and a pSM3fr BAC-derived virus which carries a stop mutation in m129 and shows the typical reduction of growth in salivary glands after in vivo infection [31] (Fig. 4B). BMDM and PEC were infected with wildtype virus and a 131stop mutant (Fig. 4C). For all macrophages tested, all mutants unable to express an intact MCK-2 showed a strongly and significantly reduced capacity to infect macrophages (Fig. 4B and C). To exclude that the differences in infection capacities are due to soluble MCK-2 produced by cells infected with wildtype virus and not to the presence of a gH/gL/MCK-2 complex promoting infection, virus pelleted from supernatants of infected cells and purified by centrifugation through sucrose cushions was used.
When mice were infected with the 131stop mutant, a reduced virus production in salivary glands was observed (Fig. S4) as described for vpSM3fr [31] and for other MCK-2 mutants [28], [32]. To study macrophage infection in vivo, we infected adult BALB/c mice with wildtype and 131stopD MCMV and analyzed F4/80- and CD11b-double-positive macrophages from the peritoneal cavity 6 hours post infection. We observed a more than 50% reduction of the percentage of MCMV-infected macrophages when mice were infected with the 131stopD mutant (Fig. 5A). A significant reduction of normalized numbers of MCMV-infected macrophages could also be observed when immunocompromised BALB/c mice were infected via the footpad. Here, liver tissue sections were stained for F4/80 and MCMV IE1 protein 10 days after infection. Infected F4/80+IE1+ liver macrophages were counted and the numbers normalized to the numbers of all F4/80+ macrophages and all IE1+ cells present in the same tissue sections to take account of differences in the overall levels of infection and macrophage recruitment (Fig. 5B and C). Thus, very consistently, infection of immortalized macrophages, of primary bone marrow-derived macrophages, of macrophages ex vivo, and of macrophages in vivo were impaired when infected with MCMV lacking MCK-2.
MCK-2 knock-out mutants only show very subtle phenotypes when their growth behavior is studied in vitro (Fig. 3 and 4). To evaluate the mechanism how MCK-2 controls infection of cells, it would be of advantage to analyze effects on a strong infection phenotype. For HCMV, it has been shown that knock-out of both, gO and pUL(128,130,131A) abolishes the capacity of the virus to infect cells [11]. Assuming that also MCMV either uses gH/gL/gO or gH/gL/MCK-2 for promoting entry into cells, knock-out of both proteins should also abolish its capacity to infect cells. To test this, we constructed double mutants lacking gO (Δm74) and additionally carrying stop cassettes either in the m129 ORF (129stop) or the m131 ORF (131stop). Reconstitution of both double mutants resulted in infected cells from which infection could barely spread (data not shown). Release of infectious virus into the supernatants could never be detected (data not shown). Yet, infectious supernatant virus carrying double mutations could readily be produced by virus reconstitution in NIH3T3 cells expressing gO (NIH3T3-gO) or MCK-2 (NIH3T3-MCK-2) (data not shown). Infection of these trans-complementing cell lines with the double mutants did not result in detectable levels of recombination between the mutated loci and the wildtype m74 or m131/129 ORFs of the trans-complementing cells (data not shown). We reconstituted the Δm74/131stop double mutant in NIH3T3-gO cells and used this virus to infect NIH3T3, NIH3T3-MCK2, or NIH3T3-gO cells. Double mutant virus produced in NIH3T3-gO cells should be gO-positive and MCK-2-negative and, after infection of new cells, virus progeny will be gO- and MCK-2-negative. If the new target cells are expressing gO then virus progeny will be gO-positive and MCK-2-negative. If the target cells are expressing MCK-2, virus progeny will be gO-negative and MCK-2-positive. Thus, we could study spread of a virus with an identical genetic backbone, but a different protein complementation. We either infected cells at a very low m.o.i. to study spread (Fig. 6A), or infection was enhanced by a centrifugation step to initially infect about 10% of cells to study virus production (Fig. 6B). Spread of the double mutant in NIH3T3 cells and, thus, in the absence of MCK-2 and gO was highly restricted (Fig. 6A, upper panel). Release of infectious virus, which was tested by titration of supernatants on NIH3T3-gO cells, could not be observed (Fig. 6B). Spread of the double mutant in cultures of NIH3T3-MCK-2 cells was predominantly focal (Fig. 6A, middle panel), and production of infectious supernatant virus was reduced when compared to the production by the double mutant growing in NIH3T3-gO cells (Fig. 6B). Thus, complementation of the MCK-2 defect of the double mutant resulted in a growth pattern comparable to the growth pattern observed for Δm74 or m74stop mutants [15]. In NIH3T3-gO cells the double mutant readily spread (Fig. 6A, lower panel) and produced infectious virus like wildtype virus (Fig. 6B). Thus, both, gO and MCK-2 could restore the spread deficiency of the double mutant, and for the first time we could show that MCK-2 is indeed promoting virus spread. MCK-2 restored not only efficient focal spread but also production of infectious virus. When supernatants where tested for DNAse-resistant viral DNA by real-time PCR, which should be an equivalent for DNA in virus particles, we found that independent of production of infectious virus, comparable numbers of viral DNA copies were released into the supernatants of NIH3T3, NIH3T3-MCK-2, and NIH3T3-gO cells (Fig. 6C). DNA copy numbers in the cell culture supernatants were identical at time points 6 and 24 hours after infection and reflect leftovers of the input supernatant (Fig. 6C). After 48 hours, the first round of replication was completed which was reflected by an increase in DNA copy numbers. In supernatants from infected NIH3T3 cells, the copy numbers were higher than in supernatants from trans-complementing cells. Very likely, this indicates that particles produced by NIH3T3 cells are not infectious, cannot enter new cells, and are accumulated, whereas particles from trans-complementing cells are infectious and infect new cells. At 96 h after infection, only supernatants from NIH3T3-gO and NIH3T3-MCK-2 cells which support efficient virus spread showed a further increase in DNA copy numbers mirroring a second round of infection (Fig. 6C). In summary, the experiments with the double mutant showed that even in the absence of gO and MCK-2, virus particles are produced and released, but they are not infectious. If the mutant is trans-complemented either with MCK-2 or gO, comparable numbers of virus particles are produced, but the infection capacities for fibroblasts seem to be lower when they are trans-complemented with MCK-2.
If gH/gL/MCK-2 is the alternative complex to gH/gL/gO with respect to promoting infection of host cells, antibodies directed against the MCK-2 complex should inhibit infection with MCMV lacking the gH/gL/gO complex but not infection with MCMV expressing gH/gL/gO. To study this, virus preparations were preincubated with a rabbit antiserum specific for MCK-2 or with a control rabbit antiserum. Then, cells were infected with these virus-antibody mixtures and infected cells were detected by staining the cells for expression of MCMV IE1. Numbers of infected cells were expressed as percent of infected cells obtained with mock-treated virus. Infection of MEF and ANA-1 cells with a Δm74 mutant (Fig. S1), could be strongly and specifically inhibited when virus was preincubated with the anti-MCK-2 antiserum, whereas infection with a 131stop mutant which expresses gO, but lacks MCK-2, could not be inhibited (Fig. 7A). Thus, in the absence of gO, infection is MCK-2-dependent. If MEF cells were infected with the Δm74 mutant trans-complemented in NIH3T3-gO cells, the inhibition by anti-MCK-2 antibodies was abrogated although not completely (Fig. 7A). This partial abrogation indicates that infection with the trans-complemented Δm74 mutant depends on gO and also on MCK-2.
We have shown before that infection of fibroblasts with MCMV lacking gO, but not with MCMV expressing gO, is energy- and pH-dependent [15]. To find out whether MCK-2 is promoting an energy- and pH-dependent entry pathway, we infected fibroblasts with a Δm74/m129stop mutant trans-complemented with gO or MCK-2 in the presence of inhibitors of ATP depletion or inhibitors of endosome acidification like bafilomycin A1 and NH4Cl [15]. Infection with MCK-2-complemented Δm74/m129stop MCMV was inhibited by all three inhibitors and inhibition was significantly different from inhibition of gO-complemented Δm74/m129stop MCMV (Fig. 7B). Bafilomyin A1 even increased infection of gO-complemented Δm74/m129stop MCMV. The inhibitor studies clearly indicate that MCK-2 promotes an energy- and pH-dependent entry pathway which is different from entry promoted in the presence of gO.
gH/gL complexes of herpesviruses have been extensively studied over the past years. The major function attributed to gH/gL associated proteins is receptor recognition. For HCMV, two gH/gL complexes, gH/gL/gO and gH/gL/pUL(128,130,131A) have been identified. gH/gL/gO determines entry into a restricted set of cell types and ensures efficient production of infectious supernatant virus in vitro [21]. The gH/gL/pUL(128,130,131A) complex determines the broad cell tropism characteristic for HCMV, very likely by recognizing a receptor found on many different cell types.
Infection of the mouse with MCMV has been shown to be a model for the HCMV infection in many, although not all, aspects [36]–[38]. We have recently characterized a functionally homologous gH/gL/gO complex of MCMV [15]. As MCMV mutants lacking gO can still infect cells and spread in cell culture, it was obvious that MCMV may also form a second gH/gL complex.
The role of the chemokine homolog MCK-2 of MCMV has been studied in vivo by using viruses in which the MCK-2 gene was deleted. Reduced salivary gland titers and reduced numbers of infected peripheral blood leukocytes have been attributed to the missing chemokine function of MCK-2. The observed phenotypes were explained by a role of MCK-2 in attracting myelomonocytic leukocytes to the site of infection which are then infected and promote dissemination and finally efficient infection of salivary glands [28]–[30], [32]. Recently, it has been shown that MCK-2 also attracts inflammatory monocytes which down-modulate antiviral CD8+ T cell responses [33]. Yet, these monocytes are not targets of infection. Additionally, it has been described that MCK-2 knock-out mutants not only recruit less myelomonocytic leukocytes to the site of infection but are also highly impaired in infecting them [30]. This pointed to an additional protein function of MCK-2 which drives infection efficiencies. However, this putative function has never been addressed.
Here, we propose a new function of MCK-2 which could explain the reduced infection efficiencies described above. We could show that MCK-2 forms a complex with gH which is incorporated into virions. It is known from crystal structures of other herpesviruses that gH and gL usually form tight heterodimers [2], [3], thus, the high molecular weight complex of gH and MCK-2 very likely is a gH/gL/MCK2 complex. In SDS-polyacrylamide gels, the complex showed a different size than the gH/gL/gO complex and anti-MCK-2 antibodies did not co-precipitate HA-tagged gO from extracts of cells infected with a virus expressing gO-HA indicating that the gH/gL/MCK-2 complex indeed is an alternative complex to gH/gL/gO.
It is difficult to study how MCK-2 promotes infection of cells in vitro, as spread and virus production of MCK-2 knock-out mutants are not drastically affected. Therefore, we used MCMV mutants lacking gO or double mutants lacking both, gO and MCK-2 to evaluate the contribution of MCK-2 to infection. Infection of cells with gO knock-out mutants could be blocked with anti-MCK-2 antibodies which demonstrated that gH/gL/MCK-2 can act as an alternative mediator of virus spread when gH/gL/gO is not formed. Trans-complementation of Δm74/131stop double mutants with MCK-2 showed that MCK-2 promotes mainly focal spread. Supernatants of cells infected with this virus only showed low titers of infectious virus, although high numbers of virus particles were released. This suggests that virions complemented with MCK-2, but lacking gO, are less efficient in infecting cells. In contrast to gO, MCK-2 promoted entry through a pH- and energy-dependent entry pathway as has been observed for MCMV mutants lacking gO. It is noteworthy in this context that in contrast to HCMV, where double mutants lacking gO and pUL(128,130,131A) are lethal [11], the MCMV double mutant can be reconstituted and spread in cell culture, although to a very limited degree and without producing free infectious virus. We do not know whether this residual spread occurs only by direct cell-to-cell transmission. It will have to be determined in the future whether MCK-2 and gO are directly involved in the entry process or whether they just promote infection as cofactors rendering target cells more susceptible for infection.
Whether gH/gL/MCK-2 is a tripartite complex or can associate with additional proteins is currently not known. Potential candidates would be the m130 and m133 genes which neighbor the m131/129 ORF. In an analysis of the MCMV transcriptome, we found that the putative m130 ORF which lies on the opposite strand and overlaps with m131/129 is not transcribed (data not shown). This is in line with data from Saederup et al. [32] who showed that interruption of the m130 ORF does not affect the phenotype of an m131/129 deletion mutant. It is intriguing that mutants lacking the m133 gene show, like MCK-2 mutants, reduced titers in salivary glands of infected mice [25], [26]. We could not detect peptides derived from the m133 ORF by mass spectrometry of anti-gH precipitates (data not shown). As this failure is not an absolute criterion to exclude that the m133 gene product is part of a gH/gL/MCK-2 complex, we deleted m133 and additionally m74. In contrast to MCK-2stop/Δm74 double mutants, the 133stop/Δm74 double mutant grew like a Δm74 mutant (data not shown).
We observed a slight growth advantage for MCK-2 mutants in fibroblasts with respect to production of supernatant virus which was not detected before [28], [31]. Interestingly, this finding is reminiscent of what was observed for UL131A mutants of HCMV [9], [39], and it might explain why isolates of MCMV do, just as isolates of HCMV, loose their capacity to form the second gH/gL complex during passage in fibroblasts [31], [40].
HCMV, which cannot form a gH/gL/pUL(128,130,131A) complex, completely loses its broad cell tropism in vitro, including its tropism for monocytes and macrophages, but can still infect fibroblasts like wildtype virus [41], [42]. This is a strong phenotype and it implies that for HCMV, infection of most cells types depends on the gH/gL/pUL(128,130,131A) complex. In contrast, deletion of MCK-2 was associated with a more restricted phenotype in vitro, namely, the loss of its capacity to efficiently infect macrophages. Additionally, an increased capacity to infect TCMK-1 epithelial cells was observed. This implies that the MCMV gH/gL/MCK-2 complex rather modulates infection capacities. Whether these differences reflect completely different roles for the gH/gL/MCK-2 complex of MCMV and the gH/gL/pUL(128,130,131A) complex of HCMV or are due to different in vitro culture systems is not known. Comparable to HCMV, rhesus CMV (RhCMV) lacking its gH/gL/pUL(128,130,131) complex shows reduced infection capacities for endothelial and epithelial cells [43], [44] but not for fibroblasts. Guinea pig CMV lacking its gH/gL/GP(129,131,133) complex loses its capacity to efficiently infect both, endothelial cells and fibroblasts [45].
Currently, it is not clear how these in vitro phenotypes translate to the in vivo infection. All CMV mutants, which lack the gH/gL complex containing a chemokine homolog, share one phenotype in vivo, namely, the loss of their capacity to efficiently establish infection in salivary glands [28], [32], [45]–[47]. We found that infection of mice with MCK-2 knock-out mutants results in reduced numbers of infected macrophages due to an impaired capacity of the mutants to infect the macrophages. How and whether this defect contributes to MCK-2 knock-out phenotypes like reduced viral titers in the salivary gland or elevated CD8+ T cell responses is currently not clear. It is also not known whether it is true for other cytomegaloviruses. Reduced virus replication of RhCMV mutants lacking gH/gL/pUL(128,130,131) is not restricted to salivary glands [47]. Yet, all in vivo studies performed so far used a RhCMV mutant which not only lacked a functional gH/gL/pUL(128,130,131) complex, but also additional viral genes coding for alpha chemokine-like proteins. When infection of different cell types in skin biopsies was tested for this RhCMV mutant a strong reduction in infection of endothelial cells and a slight, but not significant reduction in the numbers of infected macrophages was observed [48].
Both, MCMV gH/gL/MCK-2 and HCMV gH/gL/pUL(128,130,131A) contain potentially functional CC chemokines. Recombinant UL128 protein can interfere with the chemokine responsiveness of monocytes [49], and also isolated MCK-2 can act as a chemokine [29], [30]. The r129 gene product of rat CMV which is homologous to HCMV UL128 has also been shown to induce migration of lymphocytes as a recombinant protein [50]. At the moment it is not known whether MCK-2 and UL128 promote infection as gH/gL complex constituents and exert their chemokine functions only as free proteins or whether both functions can be complex-associated. Co-immunoprecipitation of gH and MCK-2 was only possible using an antibody recognizing the m131 ORF, but not the m129 derived protein part which indicates that the latter is involved in complex formation, whereas the part containing the CC chemokine domain is accessible. Thus, complex formation might still allow chemokine function of MCK-2. This also raises the question whether the chemokine function of MCK-2 and entry promoted by MCK-2 are both transmitted by the same cellular receptor. It will be of particular interest to find out whether it is possible to make MCK-2 mutants which are active chemokines, but no longer promote infection in the absence of gO or vice versa and to study them in vivo.
Primary mouse embryonal fibroblasts from BALB/c mice (MEF), NIH3T3 cells (ATCC: CRL-1658), the endothelial cell line MHEC5-T [51], the epithelial cell line TCMK-1 (ATCC: CCL-139), the macrophage cell line J774 (ATCC: TIB-67), and peritoneal exudates cells (PEC) from BALB/c mice were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum. The macrophage cell line ANA-1 [52] was maintained in RPMI medium supplemented with 10% fetal calf serum.
BMDM were prepared from BALB/c mice. Femurs and tibias were removed and cleaned, and bone marrow was flushed through with DMEM supplemented with 10% FCS, 2 mM L-glutamine, 100 U/mL penicillin, 100 µg/mL streptomycin and 50 µM 2-mercaptoethanol. To remove stromal cells, bone marrow cell suspensions were first seeded in 10 cm tissue culture dishes for four hours. Then, non-adherent cells were collected, resuspended in complete medium additionally containing 20 ng/ml murine recombinant M-CSF (Peprotech), and cultivated for 7 days in 10 cm tissue culture dishes. During this time, non-adherent cells were removed daily and half of the medium was replaced by fresh, M-CSF containing medium. At day 7, cells were harvested and used for FACS analysis and subsequent experiments. More than 95% of the cells generated by this method stained positive for the macrophage marker F4/80 (data not shown).
As wildtype MCMV, a BAC-derived virus (pSM3fr-MCK-2fl) cloned from MCMV strain Smith was used [31]. pSM3fr BAC-derived virus was used as an additional m129 stop mutant [31], [53]. For infection experiments, supernatants from infected cells showing complete cytopathic effect (CPE) and precleared at 3,500× g were used. For production of supernatant virus for protein analysis, NIH3T3 cells were infected at an m.o.i. of 0.1. Media were collected when a full CPE was observed, cleared at 6,000× g for 10 min and then pelleted for 4 h at 20,000× g. Virus stocks for analysis of macrophage infection efficiencies were prepared as described recently [31]. Virus titers were determined by a TCID50 assay performed in 96 well plates on MEF or on NIH3T3-gO.
Monoclonal mouse anti-MCK2 antibodies 5A5 and 2H9, rabbit anti-MCK2 antiserum WU1073 [27], and rabbit anti-pUL131A antiserum [9] have been described before. HA-tagged proteins were detected with rat anti-HA antibody (3F10, Roche Diagnostics). Mouse macrophages were stained with rat anti-F4/80 antibody (BM8, BioLegend). Mouse anti-MCMV gH (8D122A) was kindly provided by Lambert Loh, University of Saskatchewan, Canada. Mouse anti-MCMV immediate early protein 1 (IE1) antibody (Croma101) was kindly provided by Stipan Jonjic, University of Rijeka, Croatia.
An NIH3T3 cell line stably expressing gO has been described before [15]. For NIH3T3 cells stably expressing MCK-2, the complete m131/129 ORF was amplified by PCR from a pCR3-MCK-2 expression vector and cloned in a modified pEPi-luc vector [54] following the same strategy as used for pEPi-gO [15]. The resulting plasmid pEPi-MCK-2 was transfected into NIH3T3 cells using Fugene (Promega), and MCK-2 expressing cell clones isolated by limiting dilution under blasticidin S selection (10 µg/ml, Invivogen). MCK-2 expression was tested by staining cell extracts in the Western blot using an anti-MCK-2 antibody.
For indirect immunofluorescence, adherent cells were fixed in 50% acetone-50% methanol and stained using anti-IE1 antibody and Fluor488-coupled goat anti-mouse antibody (Invitrogen). For counterstaining of cell nuclei, cells were incubated in PBS containing 5 µg/ml Hoechst 333258 (Invitrogen). For intracellular FACS staining, cells were detached with 0.5 mM Na-EDTA, fixed with 1% paraformaldehyde for 10 min and then stained in PBS containing 0.3% Saponin and 1% BSA using the antibodies described above. Cells were washed with PBS containing 0.03% Saponin. After staining, cells were resuspended in 1% paraformaldehyde and analyzed on a FACSCalibur using CellQuest software (BD Biosciences).
Cells or virus pellets were lysed in RIPA buffer (50 mM Tris (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1% NP-40, 0.1% SDS, 0.5% deoxycholate). Lysates were precleared with Sepharose G beads (GE Healthcare) and then, beads with antibody bound were added to the precleared lysates and coincubated for 4 h at 4°C. The beads were washed, proteins released in reducing sample buffer (0.13 M Tris-HCl (pH 6.8), 6% SDS, 10% α-thioglycerol) or in non-reducing sample buffer without α-thioglycerol and subjected to SDS-PAGE, followed by either Western blot analysis or LC-MS/MS.
For preparation of peptides for LC-MS/MS, gel slices were chopped from the SDS-PAGE, treated with water and ammonium bicarbonate, and afterwards dehydrated using acetonitrile. Samples were reduced in DTT buffer (10 mM DTT, 40 mM ammonium bicarbonate) for 1 h and then alkylated with iodoacetamide buffer (55 mM iodoacetamide, 40 mM ammonium bicarbonate) for another 30 min in the dark. After washing in 40 mM ammonium bicarbonate, gel slices were dehydrated again in acetonitrile and soaked in 40 mM ammonium bicarbonate containing sequencing grade modified trypsin (Promega). Samples were incubated overnight at 37°C and resulting peptides were extracted by 5% formic acid, dried in a SpeedVac concentrator, resuspended in 15 µl 0.1% formic acid and analyzed in a nano-ESI-LC-MS/MS. Here, each sample was first separated on a C18 reversed phase column via a linear acetonitrile gradient (UltiMate 3000 system, Dionex) and column (75 µm i.d. ×15 cm, packed with C18 PepMap, 3 µm, 100 Å; LC Packings), before MS and MS/MS spectra were acquired on an Orbitrap mass spectrometer (Thermo Scientific). Recorded spectra were analyzed via Mascot Software (Matrix Science) using an MCMV protein database.
Markerless BAC mutagenesis was performed to introduce stop cassettes in the m131/129 ORF, to delete 532 bp at the N-terminus of the m74 ORF, to introduce a C-terminal HA-tag to the M75 ORF and to introduce a C-terminal HA-tag to the m74 ORF in the pSM3fr-MCK-2fl BAC as described previously [11], [55]. A schematic presentation of the pSM3fr-MCK-2fl mutants is depicted in Figure S1. For the pSM3fr-m129stop BACmid (virus: 129stop), the primers m129stop-for (5′- GTACCGTTCCCGACCCAGGTGATCTCACAGACACACTCTATCCAGTTTTCGGCTAGTTAACTAGCCAGGATGACGACGATAAGTAGGG-3′) and m129stop-rev (5′-AATCGCCACGCATCACGGTGGGCAAGTACCCCTACGAGGTGAAGGACGGTGGCTAGTTAACTAGCCGAAAACTGGATAGAGTGTGTCAACCAATTAACCAATTCTGATTAG-3′) were used. For the pSM3fr-m131stop BACmid (virus: 131stop), the primers m131stop-for (5′-TGACCAGACACAAGAGTCTGTCCGACCACCAGGCCCGCTTAGCGCACACCGGCTAGTTAACTAGCCAGGATGACGACGATAAGTAGGG-3′) and m131stop-rev (5′-AACACTTCGTGCGGACGAGAGGTGGTTTTCACTACCTTCTCTGGGATGAGGGCTAGTTAACTAGCCGGTGTGCGCTAAGCGGGCCTCAACCAATTAACCAATTCTGATTAG-3′) were used. For the pSM3fr3-Δm74 BACmid (virus: Δm74), the primers deltam74-for (5′-TTTAAAATATTTGGCGGTGATGTTACTTTTCGGGGTGATGAGGTCTCTCCAGGATGACGACGATAAGTAGGG-3′) and deltam74-rev (5′-AGAGCCGCGATTAATGTCCGCTGTATTCAACGCGGAGATCAGCCCTCCCGGGAGAGACCTCATCACCCCGAAAAGTAACATCACCGCCAAATATTTTAAACAACCAATTAACCAATTCTGATTAG-3′) were used. For the pSM3fr-M75-HA BACmid (virus: gH-HA), the primers M75HA-for (5′-TAGCGATCCTCATGGCGCTAGGGCTGTACCGGCTGTGCCGGCAAAAAAGATACCCATACGACGTCCCAGACTACGCTAGGATGACGACGATAAGTAGGG-3′) and M75HA-rev (5′-GACGCAATAAAGAATCTTTTCTTTCTTCATTCACCTCGCGTGTGTCCTTACTAAGCGTAGTCTGGGACGTCGTATGGGTACCGACACGGCCGTTTTTTCTCAACCAATTAACCAATTCTGATTAG-3′) were used. For the pSM3fr-m74-HA BACmid (virus: gO-HA), the primers m74HAfor (5′-AGAAACCACAACAACACGTACCGTCTCTGCCCCACAAAAGGCGCACCGGCTCAATATCCTTTAGCCGTGTCTACCCATACGACGTCCCAGACTACGCTAGGATGACGACGATAAGTAGGG-3′) and m74HA-rev (5′-GGCACTGGTGTTACAAGGCCTTCACCTCAGACACGGCTAAAGGATATTGACTAAGCGTAGTCTGGGACGTCGTATGGGTAGACACGGCTAAAGGATATTGAGCCGGTGCGCCTTTTGTGGGCAACCAATTAACCAATTCTGATTAG-3′) were used. This BAC also has a duplication of 18 C-terminal base pairs of m73 which overlapped with the C-terminus of m74. The sequences of the stop cassettes and the HA-tags in all primers are highlighted. Deletions and insertions of stop cassettes or HA-tags were controlled by restriction pattern analysis and subsequent sequencing. BACs were reconstituted to virus by transfection of BAC DNA into MEF using Superfect transfection reagent (Qiagen) according to the manufacturer's instructions. Transfected cells were propagated until viral plaques appeared, and supernatants from these cultures were used for further propagation.
Virus particles were purified from supernatants of MCMV-infected cells by Nycodenz-gradient purification [56]. Briefly, supernatants were cleared at 6,000× g for 10 min to remove cell debris, and then virions were pelleted by centrifugation at 20,000× g for 4 h. The resulting pellet was resuspended in VS-buffer (0.05 M Tris, 0.012 M KCl, 0.005 M EDTA (pH 7.8)) and free DNA removed by overnight treatment with 625 U/ml Benzonase (Novagen) at 4°C. Then, the suspension was loaded onto a continuous 10–40% Nycodenz (Axis-Shield) density gradient and separated at 20,000× g for 105 min at 4°C, and the band corresponding to virus particles was collected.
100 µl supernatant from infected cells was pretreated with 75 U Benzonase for 20 min at RT to remove free DNA, and then DNA was extracted using the DNeasy blood and tissue kit (Qiagen). 1/20th of the extracted DNA was used for real-time PCR which was performed on a Light Cycler (Roche Molecular Biochemicals) as described recently [57]. Primers used were specific for the MCMV M54 gene [15]. Viral DNA copy numbers/ml were calculated by comparing the amplification to standard curves using pSM3fr-LBR BAC DNA.
For energy depletion, cells were preincubated in energy depletion medium (glucose-free DMEM with 2% bovine serum albumin, 50 mM 2-deoxy-D-glucose, 0.1% sodium azide) for 1 h followed by coincubation with virus for 90 min in the presence of energy depletion medium. Virions that had not penetrated were inactivated by washing the cells two times with PBS pH 3.0. For inhibition of pH-dependent endocytosis, cells were pretreated with medium containing NH4Cl or bafilomycin A1 (Sigma) for 1 h at 37°C. Infection (90 min) and further incubations were all performed in the presence of the respective inhibitors. For all inhibitions, infection was monitored by staining cells for IE1 expression three hours after removing supernatant virus.
Female BALB/c mice were housed and bred under specified-pathogen-free conditions at the Central Animal Facility of the Medical Faculty, University of Rijeka, in accordance with the guidelines contained in the International Guiding Principles for Biomedical Research Involving Animals. The approval of animal protocols has been obtained from the authorised Ethics Committee of the Croatian Ministry of Agriculture, Veterinary Department (Class: UP/I-322-01/13-01/31; No.: 525-10/0255-13-2). The animal care authorisation for the Central Animal Facility of the Medical Faculty, University of Rijeka has been issued by the Croatian Ministry of Agriculture, Veterinary Department (authorisation number: HR-POK-004). Eight- to 12-week-old mice were used in all experiments. The mice were infected intraperitoneally (i.p.) with 5×105 PFU of wildtype or 131stopD in a volume of 500 µL. PEC collection: Mice were sacrificed 6 h p.i. and PEC were collected by washing the peritoneal cavity with 10 ml cold PBS. Erythrocytes were lysed, cells counted and 1×106 cells stained for surface markers with the following antibodies: anti-F4/80-APC (BioLegend, BM8), anti-CD11c-PE (eBioscience, N418), anti-CD19-PerCP-Cy5.5 (eBioscience, eBio1D3), anti-CD11b-PECy7 (eBioscience, M1/70). Cells were then fixed using Cytofix/Cytoperm solution (BD) and Perm/Wash (BD) was used to dilute Abs for IC staining as well for washing. Cells were first incubated with CROMA229 (anti-m06) antibody and then with FITC-labeled rat anti-mouse IgG1 mAb (BD, A85-1). F4/80+CD11b+ macrophages were gated according to a recently published strategy [58] and analyzed for m06 expression. Flow cytometry was performed on FACSAria (BD Bioscience; San Jose, CA), and data were analyzed using the FlowJo software (Tree Star).
Female BALB/c mice were immunocompromised by total-body γ-irradiation with a dose of 6.5 Gy and infected in the left hind footpad with 105 PFU of the indicated viruses. Mice were bred and housed under specified-pathogen-free conditions in the Central Laboratory Animal Facility (CLAF) at the University Medical Center of the Johannes Gutenberg-University, Mainz. Animal experiments were approved according to German federal law, permission numbers 23 177-07 and G10-1-052. Two-color immunohistochemical analysis (IHC) was performed on liver tissue sections at day 10 after infection. Macrophages were labeled specifically with a rat mAb directed against antigen F4/80 (Ly71; clone BM8, Acris antibodies). Black staining was achieved by using biotin-conjugated polyclonal anti-rat Ig (BD) and the peroxidase-coupled avidin-biotin complex (Vectastain Elite ABC kit, Vector Laboratories) with DAB as substrate and ammonium nickelsulfate hexahydrate for color enhancement. Infected cells were then labeled specifically with murine mAb CROMA 101, directed against viral protein IE1, and stained red with goat polyclonal alkaline phosphate-conjugated anti mouse IgG (AbD Serotec) and a fuchsin substrate-chromogen kit (Dako-Cytomation). Light blue counterstaining was performed with hematoxylin.
|
10.1371/journal.pntd.0000730 | Antiangiogenic and Antitumor Effects of Trypanosoma cruzi Calreticulin | In Latin America, 18 million people are infected with Trypanosoma cruzi, the agent of Chagas' disease, with the greatest economic burden. Vertebrate calreticulins (CRT) are multifunctional, intra- and extracellular proteins. In the endoplasmic reticulum (ER) they bind calcium and act as chaperones. Since human CRT (HuCRT) is antiangiogenic and suppresses tumor growth, the presence of these functions in the parasite orthologue may have consequences in the host/parasite interaction. Previously, we have cloned and expressed T. cruzi calreticulin (TcCRT) and shown that TcCRT, translocated from the ER to the area of trypomastigote flagellum emergence, promotes infectivity, inactivates the complement system and inhibits angiogenesis in the chorioallantoid chicken egg membrane. Most likely, derived from these properties, TcCRT displays in vivo inhibitory effects against an experimental mammary tumor.
TcCRT (or its N-terminal vasostatin-like domain, N-TcCRT) a) Abrogates capillary growth in the ex vivo rat aortic ring assay, b) Inhibits capillary morphogenesis in a human umbilical vein endothelial cell (HUVEC) assay, c) Inhibits migration and proliferation of HUVECs and the human endothelial cell line Eahy926. In these assays TcCRT was more effective, in molar terms, than HuCRT: d) In confocal microscopy, live HUVECs and EAhy926 cells, are recognized by FITC-TcCRT, followed by its internalization and accumulation around the host cell nuclei, a phenomenon that is abrogated by Fucoidin, a specific scavenger receptor ligand and, e) Inhibits in vivo the growth of the murine mammary TA3 MTXR tumor cell line.
We describe herein antiangiogenic and antitumor properties of a parasite chaperone molecule, specifically TcCRT. Perhaps, by virtue of its capacity to inhibit angiogenesis (and the complement system), TcCRT is anti-inflammatory, thus impairing the antiparasite immune response. The TcCRT antiangiogenic effect could also explain, at least partially, the in vivo antitumor effects reported herein and the reports proposing antitumor properties for T. cruzi infection.
| In Latin America, 18 million people are infected with Trypanosoma cruzi, a protozoan that causes Chagas' disease. Vertebrate calreticulins (CRTs) are multifunctional, intra- and extracellular calcium binding, chaperone proteins. Since human CRT (HuCRT) inhibits capillary growth (angiogenesis) and suppresses tumor growth, the presence of these functions in T. cruzi CRT (TcCRT) may have interesting consequences in the host/parasite interactions. Previously, we have cloned and expressed TcCRT and shown that, when translocated from the endoplasmic reticulum to the area of trypomastigote flagellum emergence, it promotes infectivity, inactivates the complement system, an innate defense arm and inhibits angiogenesis in the chorioallantoid chicken egg membrane. TcCRT inhibits angiogenesis, since it interferes with endothelial cell multiplication, migration and capillary morphogenesis in vitro, as well as angiogenesis in rat aortic rings. The parasite molecule also displays important antitumor effects. In these activities, TcCRT is more effective than the human counterpart. Perhaps, by virtue of its capacity to inhibit angiogenesis, TcCRT is anti-inflammatory, thus impairing the antiparasite immune response. The TcCRT antiangiogenic effect could also explain, at least partially, the in vivo antitumor effects reported herein and the reports proposing antitumor properties for T. cruzi infection.
| Chagas′ disease affects 16 million people in South America, with 14.000 deaths per year and 0.7 million disability-adjusted life-years [1]. T. cruzi has a variety of molecules that modulate several effector arms of the immune system [2], calreticulin (TcCRT) being one of them [3]. TcCRT, first isolated in our laboratory [4], [5], is highly homologous with human calreticulin (HuCRT) [6], an exceedingly pleiotropic chaperone molecule [7]. In spite of its primary endoplasmic reticulum (ER) location, TcCRT is also expressed on the cell membrane [3].
Based on their capacity to bind laminin [8] and to inhibit endothelial cell proliferation, both HuCRT and its N-terminal fragment, vasostatin or N-TcCRT, display antiangiogenic properties in vitro and in vivo [9], [10]. These HuCRT properties are paralleled by inhibitory activities on several tumor models [11]–[13]. Identifying these properties in TcCRT may define important aspects of the host/parasite interaction.
We have recently reported that TcCRT is strongly antiangiogenic in the chorioallantoid membrane in chicken eggs (CAM assay) [14]. Since angiogenesis modulators behave differently across species [15], we verified this effect in different experimental set ups in mammals, Homo sapiens sapiens included. Thus, TcCRT and its vasostatin-like domain, inhibit angiogenesis in the ex vivo rat aortic ring assay. It also affects key cellular angiogenic parameters in human endothelial cell cultures, such as proliferation, chemotaxis and cell morphogenesis into tubular-like structures in Matrigel. These results correlate with TcCRT binding and internalization in these cells. Perhaps, the TcCRT antiangiogenic (and anti-complement) properties result in anti inflammatory outcomes, thus inhibiting the host antiparasite immune response. Also, at least a partial explanation for those reports [16], [17] proposing anti-tumor effects for trypanosome infection is herein provided. Although anti-tumor effects have been reported for several decades now, for a variety of infections with other microbial agents [18], [19], pathogen molecules mediating those statistically based tumor resistances, have been poorly defined. In synthesis, here we describe that a parasite chaperone molecule, most likely by interacting with endothelial cells, and inhibiting angiogenesis, interferes with tumor growth.
Human umbilical vein endothelial cells (HUVECs) were isolated [20], following informed patient's written consent (University of Chile Hospital Bioethics Committee).
The human endothelial EAhy926 cell line (kindly provided by Dr. Gareth Owen, Pontifical Catholic University, Chile), was maintained in Iscove's Modified Dulbecco's Medium (IMDM, Invitrogen, USA) with 10% fetal bovine serum (FBS, Invitrogen, USA) and 100 units/ml penicillin/streptomycin (Sigma, USA). HUVECs were 80% pure by flow cytometry and immunofluorescence using anti CD31 monoclonal antibodies (Sigma, USA) as a marker. The cells were cultured in M199 medium (Sigma, USA), with 20% FBS, 2 mM glutamine (Invitrogen, USA), 100 units/ml penicillin/streptomycin, 100 µg/ml endothelial cell growth supplement (ECGS) (BD Biosciences, USA), and 10 µg/ml heparin (Sigma, USA) in gelatin-coated flasks.
TcCRT, its R-domain (R-TcCRT) and HuCRT were obtained from E. coli [3], [21]. N-TcCRT (amino acids 20–193, GenBank accession no. AF162779) was amplified by PCR using Tli DNA polymerase (Promega, USA). Primers were: (5′-GGAATTCCACGGTGTACTTCCACGAG-3′) and (5′- CTCGAGCCAGTCTTCTTCGAGCTG-3′).
N-TcCRT DNA was ligated into the EcoRI and XhoI sites of the pET-28b (+) plasmid (Novagen, UK). Competent E. coli TOP10F′ bacteria were transformed, plated and selected with 50 µg/ml ampicillin. E. coli BL21 (DE3)pLysS was transformed with the plasmid and grown in the presence of 34 µg/ml chloramphenicol with 50 µg/ml kanamycin. After adding isopropyl β-D-thiogalactoside and 3 h incubation, the cells were sonicated, centrifuged, and the supernatants filtered. The recombinant proteins were purified using His Bind resin (Novagen, UK), eluted with buffer containing 1 M imidazole, and dialyzed against 2 mM Tris-HCl and 150 mM NaCl, pH 7.4. Both, TcCRT and N-TcCRT were tested for endotoxin by the Limulus Amebocyte Lysate Kinetic-QCL assay (BioWhittaker, USA) and contained <5 EU/10 mg protein.
The R-TcCRT domain (aa 136–281) was expressed and purified as previously described [3].
This ex vivo angiogenesis assay [22], was performed with slight modifications. Six week old Sprague-Dawley rats, from our Animal Facility were used in this experiment. Briefly, the animals were sacrificed by CO2 inhalation, their thoracic aortas dissected and sliced into 1 mm thick rings. Two or three rings per well were placed on a 24-well plate and embedded in 100 µl Matrigel (BD Biosciences, USA), followed by 30 min incubation. Wells were overlaid with 300 µl of FBS-supplemented M199 medium with 100 µg/ml ECGS and phosphate buffered saline (PBS) or several TcCRT concentrations. The rings were incubated for 7 days and visualized under phase contrast in a Nikon Eclipse E400 microscope. Fields were photographed and the length of capillaries measured using Adobe Photoshop software. For each experiment and in sextuplicate, 3 capillaries (shortest, medium and longest) per ring were measured. The average length was considered as 100%. The statistical validation of these experiments was defined by the Student's t-test.
24-microwell plates were filled with 300 µl Matrigel/well and polymerized for 1 h at 37°C. 70×103 HUVECs/well were suspended in FBS-supplemented M199 medium, with 100 µg/ml ECGS and several TcCRT, N-TcCRT, lypopolisaccharide (LPS), HuCRT or R-TcCRT concentrations. The cells were layered on the gel. After 6 h incubation, morphogenesis was assessed by phase contrast microscopy and images were imported into the Adobe Photoshop program. Tubular capillary-like structures were quantified by manual counting in 40× fields, in quadruplicates, as previously described [23]. Data were analyzed by one way ANOVA. Values are reported as means ± SEM. Comparison of means was performed by the Bonferroni method.
With HUVECs, the assays were performed in Boyden chambers, while Transwell chambers (Costar, USA) were used with EAhy926 cells [24]. HUVECs were pretreated for 24 h with PBS, LPS, or variable TcCRT concentrations in FBS-supplemented M199 medium. EAhy926 cells were pretreated with IMDM containing several TcCRT concentrations. 7.5×104 HUVECs or 5×104 EAhy926 cells/chamber were washed, resuspended in serum-free medium, and placed in the upper compartment, with or without TcCRT or LPS. Supernatants from NIH3T3 cells (for HUVECs) or 10% FBS (for EAhy926) were used as chemo attractants in the lower chamber. After 6 h (HUVECs) or 16 h (EAhy926) incubation, the cells on the upper filter surface were removed, and those on the lower surface, fixed and stained. Filters were photographed with CCD optics and a digital analysis system (Image ProPlus, Media Cybernetics, Silver Spring, MD) and nine fields per filter were counted (HUVECs). EAhy926 cell migration was measured by densitometry analysis at 595 nm. All experiments were performed in triplicates. Data were analyzed by one way ANOVA. Values are reported as means ± SEM. Comparison of means was performed by the Bonferroni method.
These assays were quantified using MTT (3-[4,5-dimethylthiazol-2-yl]2,5-diphenyltetrazoliumbromide, Calbiochem, USA) or crystal violet reagents. Briefly, in the MTT assay, 2,500 HUVECs/well were seeded in sestuplicate in 96-well plate and growth, in the presence of various TcCRT, N-TcCRT or HuCRT concentrations, was assessed at 24-h periods over 4 days. Then, MTT was added, incubated for 4.5 h, solubilized in DMSO and the absorbance was read at 550 nm. The same assay was performed with 2,000 VERO cells, as a negative control showing that recombinant TcCRT did not affect the in vitro growth of an unrelated cell line. Data were analyzed by one way ANOVA, followed by the Bonferroni test. Values are reported as means ± SEM. In the crystal violet assay, the same number of HUVECs were seeded in gelatin-coated wells and treated with R-TcCRT at different concentrations. The number of viable cells was measured over time with the crystal violet reagent, following standard procedures.
TcCRT was labeled with the FluoReporter FITC Protein Labeling Kit (Molecular Probes, USA). HUVECs or EAhy926 cells were incubated with 1 µM TcCRT, FITC-TcCRT or FITC-TcCRT plus 10 µM unlabelled TcCRT, for 1 h. After washing, the cells were fixed with 4% paraformaldehyde, for 15 min at room temperature, washed and mounted in 50% glycerol, containing 4′-6-diamidino-2-phenylindole (DAPI). Slides were visualized in a Nikon Eclipse E400 epifluorescence microscope. Protein uptake was detected by incubating the cells for 30 min, in medium containing 1 µM FITC-TcCRT, alone or in the presence of 25 µg/ml fucoidin (Sigma, USA). Images were collected using the LSM510 Software system attached to a Zeiss (Oberkochen, Germany) LSM510meta confocal microscope.
The TcCRT and HuCRT effects on in vivo growth of the TA3 MTXR murine mammary tumor cell line was assessed in 2 independent experiments, performed 6 months apart, in adult female A/J mice. Four animals were used in the first experiment and 6 in the second one. In both experiments, the animals were inoculated s.c., every other day, with 50 µg TcCRT or HuCRT or solvent, during 25 days. At day 0, the animals were challenged with 5×105 tumor cells. Tumor size was determined with a digital caliper (Mitutoyo Corp, Japan), in a double blind procedure, as previously described [25]. The experiments were validated by using the Wilcoxon Signed Rank test (GraphPad Prism 4). P values≤0.05 were considered as statistically significant.
Six week old New Zealand rats and adult (20–25 g) female A/J mice were obtained from our Central Animal Facility. Experiments were performed in compliance with the “Guide for the Care and Use of Laboratory Animals”, National Research Council, Washington DC, USA, 2002. All procedures with these animals were approved by the local Bioethics Committee (Bioethics Committee, Faculty of Medicine, University of Chile). Surgeries and sacrifices were performed by the Animal Facility Veterinary Surgeons.
Two representative experiments are shown in Figure 1, A–B. Micro vessels are observed after culturing the aortic rings for 1 week (Figure 1A, control). Incubation with 1 µM TcCRT mediated complete capillary growth abrogation (Figure 1A, TcCRT). A dose-dependent antiangiogenic effect is observed (Figure 1B), until reaching complete capillary growth arrest. In Figure 1C, quantification of this TcCRT inhibitory capacity is shown. At concentrations of 0.1 and 1.0 µM, about 50% and 100% inhibition is respectively observed. In separate experiments, the vasostatin like N-TcCRT also inhibits angiogenesis in this ex vivo experimental model (data not shown).
A set of representative experiments is shown in Figure 2. In a 5-hour culture, control non-treated HUVECs generated a typical cell network (Figure 2A). Although strong inhibitory effects were observed with 1 µM HuCRT (Figure 2B), when N-TcCRT (Figure 2C) and TcCRT (Figure 2D) were compared at equal molarities with HuCRT, the effects of the parasite–derived molecules were clearly stronger than those of the human counterpart. Figure 2E shows the quantification of these assays. The TcCRT inhibitory effect was dose-dependent down to 0.1 µM (data not shown), while R-TcCRT did not affect capillary morphogenesis (Figure 2F–H).
HUVECs migration, as a response to the strong angiogenic factors present in NIH/3T3 cell conditioned media, was inhibited in a dose-dependent manner by TcCRT. LPS, at concentrations similar to those present in the TcCRT 1 µM preparation, showed no detectable effects (Figure 3A). Treatment with TcCRT also significantly inhibited migration of Eahy926 cells in response to FBS, over the same dose range (Figure 3B).
Figure 4 summarize these experiments. TcCRT inhibited endothelial cell proliferation in a dose-dependent manner, when they were stimulated with ECGS (Figure 4A). Maximum inhibition (60%) was observed with 1 µM TcCRT, at 96 hours (Figure 4B). A similar activity was also observed when TcCRT or N-TcCRT were added to HUVECs stimulated with basic fibroblast growth factor (bFGF) (Figure 4C). R-TcCRT, up to 1 µM, had no significant effects on HUVECs proliferation (Figure 4D). TcCRT did not affect VERO cell proliferation (Figure 4E), used as negative control.
Although both HuCRT [8] and TcCRT bind to laminin, only the former interferes with the adhesion of endothelial cells to this molecule (data not shown). Therefore, the TcCRT antiangiogenic effect may be explained by other mechanisms, such as direct interaction with endothelial cells. FITC-TcCRT binds to live HUVECs (Figure 5C). This binding is reversed by a molar excess of the unlabeled protein (Figure 5D). Given the similarity between the DAPI and FITC-TcCRT mediated signals in this experiment (Figure 5C, merge), confocal microscopy was used to test if TcCRT was internalized after binding to the cell surface. After 30 min incubation, TcCRT accumulates around the HUVECs nuclei, in punctuate structures (Figure 5E), a phenomenon also observed in EAhy926 endothelial cells (data not shown). In order to better substantiate the TcCRT internalization by endothelial cells, an enlargement of a representative cell is shown (extreme right panel in Figure 5E). TcCRT internalization seems to be receptor-dependent, since fucoidin, a specific scavenger receptor ligand [26], [27], abrogated TcCRT uptake (Figure 5F).
The TcCRT and HuCRT effects on the in vivo growth of the TA3 MTXR murine A/J mammary tumor cell line was assessed in adult mice, in two independent experiments, performed 6 months apart (Figure 6A–B). Under the experimental conditions used, only the parasite chaperone molecule displayed significant (p = 0.0078) inhibitory effects on this tumor cell line, in both cases (Figure 6A–B). In one experiment (Figure 6A), TcCRT displayed a stronger antitumor effect, than the human orthologue (p = 0.0078 vs p = 0.1094). In the second experiment, HuCRT also had an effect (Figure 6B, p = 0.0078). However, again TcCRT had a stronger antitumor effect than HuCRT (p = 0.0078) (Figure 6B).
We have shown that TcCRT strongly inhibits capillary growth in the CAM in vivo assay [14]. Since angiogenesis modulators behave differently, not only across species, but also depending on the assay used [15], we studied the TcCRT antiangiogenic properties in the rat, a natural T. cruzi host. The ex vivo rat aortic ring assay provides a model closer to the physiologic in vivo situation, since endothelial cells are in a quiescent state, in a natural histological environment. In this assay, TcCRT completely abrogates capillary growth, in a dose-dependent manner (Figure 1). Capillary morphogenesis in Matrigel is a valid in vitro correlate of in vivo angiogenesis. As shown in Figure 2, when TcCRT, N-TcCRT and HuCRT were compared in their capacities to inhibit morphogenesis, only the parasite-derived molecules significantly interfered with this process. The relevant TcCRT aminoacid sequence spans residues 20–191, corresponding to N-TcCRT. R-TcCRT did not affect capillary morphogenesis, in spite of its overlapping with N-TcCRT in aminoacids 136–191.
Chemotaxis is an essential step in capillary morphogenesis and angiogenesis. In HUVECs and Eahy926 cells, migration was inhibited in a dose-dependent manner by TcCRT (Figure 3). Cell migration inhibition by TcCRT may explain (at least partly) its potent effects on in vitro capillary morphogenesis and ex vivo capillary formation. These results agree with those describing the HuCRT capacity to increase cell binding to extracellular matrix, with consequent cell migration inhibition [28], [29].
As shown in Figure 4, TcCRT and N-TcCRT share the HuCRT capacity to specifically inhibit endothelial cell proliferation, a key initial event in angiogenesis [10]. These effects were not observed in a different cell line, like fibroblasts, used as negative controls. In HuCRT, the smallest anti-proliferative fragment spans aa 120–180 [10]. Since, as observed in the morphogenesis assay, R-TcCRT had no significant effect on HUVECs proliferation, relevant residues also map between aa 20–135. TcCRT interferes with pro angiogenic bFGF (Figure 4C), by unknown mechanisms. HuCRT also inhibits the proliferation of endothelial cells from diverse origins, such as FBHE [10], BAECs [30], HUVECs [31] and ECV304 [32], in response to bFGF and VEGF. R-TcCRT did not affect HUVECs proliferation (Figure 4D), nor morphogenesis (Figure 2F–H).
HUVECs proliferation inhibition by TcCRT may imply its involvement in the cell cycle or, alternatively, in cell death induction. TcCRT added at different concentrations to 24, 72 and 96 h HUVECs cultures did not induce apoptosis. Therefore, in the TcCRT-mediated inhibition of cell proliferation, a cytostatic effect, rather than apoptosis induction, may be mediated by the parasite molecule.
Recombinant proteins from E. coli, are normally contaminated with LPS, an antiangiogenic molecule [33]. In all the experiments discussed above, LPS was ineffective at concentrations equivalent to those present in the recombinant TcCRT preparations.
Although both HuCRT and TcCRT bind laminin, only the former interferes with endothelial cell adhesion and, as a consequence, with angiogenesis. Thus, the antiangiogenic TcCRT effects could be explained by other mechanisms, such as direct TcCRT interaction with endothelial cells. Alternatively, TcCRT could be internalized and fulfill other functions in the intracellular compartments. We now show that TcCRT binds to endothelial cells, followed by internalization. The transduction pathways involved are unknown. SREC-I (scavenger receptor expressed by endothelial cell-I) could be involved in these phenomena. HuCRT binds SREC-I, is endocytosed, and delivers associated peptides for cross presentation via MHC- I [34], [27], a fact compatible with our observations on the fucoidin (a specific SREC-I ligand [26], [27]) capacity to inhibit TcCRT internalization by HUVECs (Figure 5E). Besides being an endocytic receptor, SREC-I is an interesting candidate for signal transduction. Its intracellular domain comprises almost half of the molecule, surprisingly large among known scavenger receptors. It also contains several potential phosphorylation consensus sites [35], [36]. These results are compatible with the possibility that TcCRT internalization is a requisite to mediate its antiangiogenic effects on endothelial cells. Whether TcCRT interferes with the endothelial cell cytoskeleton, is unknown.
Perhaps, the parasite ability to inhibit angiogenesis interferes with immune/inflammatory responses against this aggressor. On the other hand, the role of angiogenesis in solid tumor progression has been long established in a variety of experimental models [37]. For six decades now, several reports have proposed a possible growth inhibitory effects that several T. cruzi strains may have on multiple transplanted and spontaneous tumors, in animals and humans [16], [17], [38]. The induction of specific immune anti-tumoral responses [39] and/or the secretion of “toxic substances” by the parasite [16], [40] were invoked to explain these effects, but no experimental evidences have been provided. Maybe, TcCRT, by interacting with endothelial cells and preventing neoangiogenesis, interferes in tumor growth and metastasis. For these reasons we tested the TcCRT and Hu-CRT capacity to inhibit in vivo the growth of a murine mammary tumor (TA3 MTXR). Only TcCRT displayed significant anti-tumor effects in both experiments. Moreover, the parasite molecule displayed stronger effects than HuCRT. Although maximum efforts were made to perform the experiments under similar conditions, the tumor growth was different by about 2-fold, in the experiments shown in Figure 6 A–B. The cell line is maintained in our laboratories, as ascites tumor in A/J mice, with weekly passages and the experiments were performed six months apart. Thus, although the conclusions drawn from both experiments are basically the same, we cannot rule out minor variations in handling, site of inoculation or in the cell line itself that could explain the different overall tumor growth observed in both experiments.
While the prevalence of tumor aggressions in wild and domestic T. cruzi hosts has not been assessed, in humans they may reach almost epidemic dimensions (i.e. mammary, prostate, ovary and cervix-uterine cancers, taken altogether). Thus, the TcCRT capacity to delay tumor growth, together with its anti inflammatory properties (derived from its complement inhibition capacity), may represent an evolutionary parasite adaptation, with final increased infectivity.
In synthesis, in this report we show that T. cruzi calreticulin has potent antiangiogenic activities, both on rat arterial (aortic ring assay) and human venous (HUVECs) endothelial cells. These properties map to the N-TcCRT domain in the parasite molecule. TcCRT plays key in vitro antiangiogenic roles, expressed as inhibition of capillary morphogenesis, proliferation and migration of endothelial cells. TcCRT internalization by endothelial cells is perhaps necessary in the antiangiogenic process. These facts, together with those previously reported by us, showing that TcCRT is a potent in vivo inhibitor of angiogenesis in a third vertebrate species (CAM assay), allow us to propose that the TcCRT antiangiogenic effects may be implicated in inflammatory and antineoplastic effects, with benefits for the parasite in its interactions with the vertebrate host. These findings may open interesting possibilities for the development of new antineoplastic strategies, especially if we consider that the parasite molecule displays stronger antiangiogenic and anti-tumor effects than its human counterpart. Biotechnological implications of these findings may be envisaged. Whether the antiangiogenic properties were consolidated, first in the parasite chaperone molecule, and HuCRT conserved some of these properties, as an evolutionary relict or, alternatively, the parasite hijacked this activity from its vertebrate host, remains an open question.
|
10.1371/journal.ppat.1000233 | FimH Adhesin of Type 1 Fimbriae Is a Potent Inducer of Innate Antimicrobial Responses Which Requires TLR4 and Type 1 Interferon Signalling | Components of bacteria have been shown to induce innate antiviral immunity via Toll-like receptors (TLRs). We have recently shown that FimH, the adhesin portion of type 1 fimbria, can induce the innate immune system via TLR4. Here we report that FimH induces potent in vitro and in vivo innate antimicrobial responses. FimH induced an innate antiviral state in murine macrophage and primary MEFs which was correlated with IFN-β production. Moreover, FimH induced the innate antiviral responses in cells from wild type, but not from MyD88−/−, Trif−/−, IFN−α/βR−/− or IRF3−/− mice. Vaginal delivery of FimH, but not LPS, completely protected wild type, but not MyD88−/−, IFN-α/βR−/−, IRF3−/− or TLR4−/− mice from subsequent genital HSV-2 challenge. The FimH-induced innate antiviral immunity correlated with the production of IFN-β, but not IFN-α or IFN-γ. To examine whether FimH plays a role in innate immune induction in the context of a natural infection, the innate immune responses to wild type uropathogenic E. coli (UPEC) and a FimH null mutant were examined in the urinary tract of C57Bl/6 (B6) mice and TLR4-deficient mice. While UPEC expressing FimH induced a robust polymorphonuclear response in B6, but not TLR4−/− mice, mutant bacteria lacking FimH did not. In addition, the presence of TLR4 was essential for innate control of and protection against UPEC. Our results demonstrate that FimH is a potent inducer of innate antimicrobial responses and signals differently, from that of LPS, via TLR4 at mucosal surfaces. Our studies suggest that FimH can potentially be used as an innate microbicide against mucosal pathogens.
| The innate immune system is an evolutionarily conserved defence mechanism that protects the host from infection by microbes such as viruses, bacteria and fungi. Incoming pathogens are recognized by a set of evolutionary conserved receptors, including the Toll-like receptors (TLRs), that can be found on the surface of epithelial cells at the mucosal surface. We recently found that FimH, a specific adhesin located at the tip of type 1 fimbriae in uropathogenic E. coli, binds directly to TLR4. Here, we demonstrate the biological significance of this interaction. In the context of a natural infection, recognition of FimH by TLR4 is important for the host to mount an innate immune response against uropathogenic E. coli. Furthermore, we show that purified FimH protein induces a potent innate antiviral response, both in tissue culture and in animal models. This response is mediated predominantly by the production of type I interferon. Our results suggest that FimH is an excellent candidate for development as a microbicide against pathogen infection.
| The innate immune system plays a crucial role in the early defence against microbial infections [1],[2],[3],[4],[5],[6]. A key aspect of the innate immune response is the synthesis and secretion of type I interferons (IFN) such as IFN-α and IFN-β. The innate immune system detects infections through germ-line encoded pattern recognition receptors [7], such as Toll-like receptors (TLRs). TLRs recognize conserved structures present in large groups of microorganisms, but not found in the host, called pathogen-associated molecular patterns (PAMPs) [8],[9],[10],[11],[12]. Thus far, 10 TLRs have been identified in mice and humans [13],[14],[15],[16], with each receptor recognizing a unique set of PAMPs [17],[18]. Examples of PAMPs include lipopolysaccharide (LPS, a ligand for TLR4), flagellin (a ligand for TLR5), double-stranded RNA (dsRNA, a ligand for TLR3), bacterial CpG DNA (a ligand for TLR9) and profillin (a ligand for TLR11). Upon ligand binding, TLRs initiate intracellular signalling through their cytoplasmic Toll/IL-1 (TIR) domain. These signalling pathways can be divided into common (MyD88-dependent) and specific (MyD88-independent) categories. TLR2, 5, 7–9 and 11 signalling is mainly MyD88-dependent, while TLR3 and 4 signalling is mediated through either MyD88 or Trif.
Recently, we and others have reported that CpG ODN/TLR9 signalling leads to the induction of potent innate protection against herpes simplex virus (HSV-2) infection both in vivo and in vitro [3],[19],[20],[21],[22],[23]. Local intravaginal (IVAG) delivery of CpG ODN or Poly I:C resulted in rapid proliferation and thickening of the vaginal epithelium and induction of a innate antiviral state that did not block virus entry but inhibited viral replication in vaginal epithelial cells. This TLR ligand-induced innate protection correlated with production of IFN-β, but not IFN-α, IFN-γ or TNF-α. Treatment of mice lacking the IFN-α/βR with CpG or Poly I:C did not provide innate antiviral protection against genital HSV-2 challenge compared to control mice. More recently, it has been shown that DC-derived IFNs are crucial for the innate antiviral activity of CpG in the genital tract [24].
Several PAMPs of bacterial origin including LPS, flagellin, peptidoglycan and bacterial DNA can activate the innate immune system via TLRs [18]. FimH, the adhesion portion of type 1 fimbriae produced by most Enterobacteriaceae including uropathogenic E. coli, is a conserved protein involved in bacterial attachment to mucosal epithelial cells [25],[26],[27]. Type I fimbriae have long been implicated in bacterial urinary tract infections in humans [25],[28],[29],[30],[31],[32],[33] and have been the focus of many attempts to generate a vaccine against pathogenic Gram-negative bacteria [26],[29],[34],[35],[36]. We have recently shown that recombinant FimH protein can activate the innate immune system through MyD88 and TLR4 in primary murine cells ([37] and un-published data). Little is known about the antiviral or antibacterial activity of FimH. Here we report that FimH is a potent inducer of innate antimicrobial responses. Our in vitro and in vivo experiments clearly show that FimH-induced innate antiviral immunity is associated with IFN-β production and requires MyD88, Trif, TLR4, IRF-3 and type I IFN signalling.
We have previously reported that bacterial LPS and CpG DNA as well as Poly I:C can induce innate antiviral responses in RAW264.7 cells. More recently, we have shown that FimH can stimulate these cells and induce the production of TNF-α ([37] and un-published data). Here we first examined if FimH-mediated signalling resulted in antiviral activity. RAW264.7 cells treated with FimH had significantly lower HSV-2 titers compared to cells treated with PBS (Fig. 1A&B). This innate protection correlated with the production of IFN-β, but not IFN-α or IFN-γ (Fig. 1C). It is well documented that FimH binds to alpha-D-mannosides [38],[39],[40],[41]. We then examined if the mannose binding domain of FimH is involved in the signaling. Incubation of FimH with D-mannose had no effect on FimH-induced innate antiviral response (Fig. 1D).
Recently we have shown that FimH signalling required TLR4 and MyD88 pathway in murine macrophages [37]. It is well documented that Poly I:C induces strong antiviral responses in mouse embryonic fibroblasts (MEFs) as measured by a standard VSV plaque reduction assay. We first examined whether FimH could induce the production of type 1 IFNs, resulting in an innate antiviral state in B6 MEFs. Interestingly, FimH induced similar levels of IFN-β, but not IFN-α, in B6 MEFs compared to those treated with Poly I:C (Fig. 2A) and provided complete protection against VSV challenge (Fig. 2B& 3). B6 MEFs treated with FimH also showed little or no VSV-GFP replication, as detected by GFP fluorescence, when compared to untreated MEFs (Fig. 3A). We then examined whether FimH-induced innate antiviral responses were MyD88, Trif and/or type I IFN dependent. FimH failed to provide protection against VSV challenge in MEFs deficient in either the MyD88 or Trif adaptors, whereas Poly I:C provided complete protection in MEFs lacking MyD88 and partial protection in MEFs lacking Trif (Fig. 2B, 3B,C). To further investigate whether type 1 IFNs, particularly IFN-β, were involved in FimH-induced innate antiviral immunity, MEFs from IFN-α/βR−/− or IRF-3−/− mice were treated with FimH or Poly I:C and then challenged with VSV. FimH failed to induce an innate antiviral state in the absence of either IRF-3 or type 1 IFN signalling (Fig. 2B, 3D,E). Poly I:C induced only moderate protection at 30 nM or 15 nM, but no significant differences in VSV-GFP fluorescence was observed in MEFs at lower concentrations compared to untreated MEFs (Fig. 2A,B, 3D,E).
We have reported that the mucosal delivery of some TLR ligands/agonists can induce an innate antiviral state and provide complete protection against subsequent IVAG HSV-2 challenge [3],[19],[22]. To examine if local delivery of FimH can provide innate protection against subsequent IVAG HSV-2 challenge, FimH was administered intravaginally to mice and then mice were challenged with lethal doses of HSV-2 24 hours following this treatment. FimH provided 100% protection against IVAG HSV-2 challenge compared to control mice (Fig. 4A). Moreover, HSV-2 virus particles were not present in the vaginal washes from FimH-treated mice compared to control mice (Fig. 4B).
To further verify that the innate antiviral protection in FimH-treated mice was due to the direct effects of FimH protein, but not LPS or other bacterial contaminants, we performed three experiments: 1) mice were treated with 5000 ng of LPS and then challenged with HSV-2; 2) mice were treated with protease-digested FimH (complete digestion was confirmed by gel electrophoresis) and then challenged with HSV-2; 3) mice were treated with either FimH or another component of bacterial pilin (PapG), which were expressed, purified and prepared in the same manner. As shown in Figure 5A neither LPS nor digested FimH protected mice against subsequent challenge with IVAG HSV-2. Mice treated with PapG were also not protected against IVAG HSV-2 challenge (Fig. 5B). However, both FimH and PapG were prepared identically and the preparations had similar levels of LPS. Interestingly, FimH, but not LPS, induced dramatic morphological changes in the genital mucosa, including thickening of the vaginal epithelium and recruitment of polymorphonuclear cells (PMNs) (Fig. 5C).
Our in vitro observations show that FimH signalling requires TLR4, MyD88 and type 1 IFN signalling to induce innate antiviral activity. Thus, we examined whether vaginal delivery of FimH could protect mice lacking these innate factors against genital HSV-2 infection. While FimH provided nearly complete protection in B6 mice, there was no protection against IVAG HSV-2 challenge in FimH-treated MyD88−/− mice (Fig. 6A). We also examined if type I IFNs, particularly IFN-β, were involved in the FimH-induced innate antiviral immunity in vivo. Vaginal delivery of FimH and Poly I:C to IFN-α/βR−/− and IRF-3−/− mice failed to protect them against IVAG HSV-2 challenge compared to control mice (Fig. 6A). This innate protection strongly correlated with the production of IFN-β, but not IFN-α, levels in the vaginal washes (Fig. 6B). To ensure that the ELISA kit could detect naturally produced IFN-α, we used supernatants from BM-DCs treated with Poly I:C or CpG ODN and were able to detect high levels of mIFN-α (Figure S1). Finally, to confirm that FimH signals via TLR4, B6 and TLR4−/− mice were treated with FimH and then challenged with IVAG HSV-2. FimH failed to protect TLR4−/− mice but completely protected B6 mice against genital HSV-2 challenge (Fig. 6C).
FimH plays an important role in the pathogenicity of uropathogenic E. coli (UPEC) [26]. To examine whether FimH plays a role in innate immune induction in the context of a natural infection, we measured PMN leukocyte recruitment to the urinary tract in B6 mice and TLR4-deficient mice following infection with wild type UPEC and a fimH null mutant. In B6 mice, UPEC expressing FimH induced a rapid PMN response whereas mutant bacteria lacking FimH did not (Fig. 7A). This FimH-induced PMN influx required TLR4, since the cellular influx was blocked in TLR4-deficient mice (Fig. 7B). The bacterial load was enumerated in the bladder 24 h after infection. TLR4 was required for control of infection in the bladder, as TLR4−/− mice had ∼1.5-log more wild type bacteria in the bladder at 24 h compared to B6 mice (Fig. 7C), which correlated with a loss of PMN recruitment in the TLR4−/− mice. Deletion of fimH resulted in decreased colonization of the bladder but this decrease was not overcome in a TLR4−/− background, confirming that while FimH is important for UPEC colonization of the bladder [33], FimH-independent signaling through TLR4 is not likely a major contributor to infection control in B6 mice.
We have demonstrated here that FimH can induce a potent innate antiviral state, both in vitro and in vivo. Pre-treatment of MEFs from B6 mice, but not MyD88−/−, Trif−/−, IRF-3−/− or IFN-α/βR−/− mice, with FimH conferred protective antiviral responses. Mucosal delivery of FimH, but not LPS, provided complete protection against IVAG HSV-2 challenge in B6 mice while it failed to provide any protection in MyD88−/− mice. Moreover, the FimH-induced innate antiviral immunity was associated with the induction of IFN-β in the genital tract and required TLR4 and IRF-3. To evaluate the biological significance of FimH in host pathogen interaction, we have examined the host innate response in FimH knockout uropathogenic E.coli in the presence and absence of TLR4. TLR4 was required for the induction of innate immune responses against UPEC.
We have shown that FimH requires TLR4 and MyD88 to activate murine primary macrophages. In addition we have also reported that dsRNA and CpG ODN confer protection against HSV-2, both in vitro and in vivo [3],[19],[22]. Similar to Poly I:C and CpG, the FimH-induced innate antiviral state correlates with IFN-β, but not IFN-α or IFN-γ production. Although we have observed high levels of TNF-α, it is unlikely that the FimH induced innate antiviral activity is mediated via TNF-α. Previous work showed that treatment of RAW264.7 cells with TNF-α cannot protect them against HSV-2 infection [42]. FimH also induces significant levels of NO production. Both IFN-β and NO are able to block HSV-2 replication [1],[2],[43],[44],[45].
We have observed induction of a strong innate antiviral state by FimH that correlated with the production of IFN-β, and required TLR4. It was first important to establish whether, in addition to TLR4, FimH binding to its natural receptor, mannose, is essential for induction of innate antiviral activity. It is well documented that FimH adhesin of uropathogenic E. coli type 1 fimbriae bind to mannose on epithelial cells. Blocking the mannose-binding portion of FimH with D-mannose had no effect on the FimH-induced innate antiviral activity. This suggested that FimH may bind to TLR4 independent of mannose to induce antiviral responses. We were unable to detect any IFN-α from FimH treated RAW264.7 or B6 MEFS by ELISA. This suggested that IFN-β is the key factor in the FimH-induced innate antiviral state. Given the importance of the transcription factor IRF3 in the production of IFN-β in fibroblast and epithelial cells, FimH also failed to induce IFN-β production from IRF-3−/− MEFs and did not protect these cells against VSV. Taken together, these data indicate that FimH signals through IRF3 in the induction of an innate antiviral response.
We and others have shown that mucosal delivery of TLR ligands protect mice against subsequent IVAG HSV-2 challenge [3],[19],[20],[21],[22],[46]. More recently, we have found that the TLR ligand-induced innate antiviral responses against IVAG HSV-2 strongly correlate with the production of IFN-β, but not IFN-α [47]. Our data show that FimH activity of FimH against genital HSV-2 challenge requires MyD88−/−, IRF-3−/− IFN-α/βR−/− and TLR4−/− mice did not provide any protection against subsequent IVAG HSV-2 challenge compared to B6 control. FimH also induced significantly lower levels of IFN-β in MyD88−/− mice while IFN-β was not detectable in IRF-3−/− mice compared to B6 control mice. This clearly suggested that FimH-induced innate antiviral activity against IVAG HSV-2 is mediated via TLR4, MyD88 and type 1 IFNs, particularly IFN-β.
Since we purified recombinant FimH from bacteria, it was essential to confirm that the antiviral activity seen with the purified FimH was not due to LPS contamination and/or other possible minor proteins. Our in vitro and in vivo data clearly showed that the antiviral activity of FimH was not due to contamination with LPS. We have used all possible controls to confirm that FimH was responsible for the innate antiviral responses. First, control samples from bacteria that contained empty vector and processed in exactly the same manner as FimH had no antiviral activity assuring that the antiviral response seen with FimH is not due to low levels of LPS contamination. Second, both enzymatic digestion and heat inactivation of FimH protein significantly abrogated the activity of FimH protein in vitro and in vivo. To also confirm that FimH, but not LPS, is responsible for in vivo innate antiviral responses, we performed several experiments. First; local delivery of LPS or digested/heat-inactivated FimH protein gave no protection against subsequent IVAG HSV-2 challenge in B6 mice compared to treatment with intact FimH protein. Second; local delivery of recombinant PapG protein, another adhesin of fimbriated bacteria which was prepared exactly with the same protocol as FimH, gave no protection against subsequent IVAG HSV-2 challenge in B6 mice compared to FimH. However, PapG protein had the same levels of LPS compared to FimH. In addition, our in vitro experiments clearly showed that low levels of LPS present in our samples cannot provide any protection against viral infections. In addition we have shown that FimH can directly bind TLR4 ([37] and un-published data). Furthermore, while FimH induced dramatic changes in genital mucosa, there was no difference in the histomorphology of the genital mucosa from LPS- or PBS-treated mice. More importantly, our recent data indicates that FimH signals in cells unresponsive to LPS ([37] and un-published data).
It is well known that FimH play an import role in attachment of bacteria to epithelial cells and contributes to pathogenecity of UPEC. Our data show that TLR4 is involved in FimH signalling with epithelial cells. FimH-expressing UPEC were able to induce recruitment of PMNs to the urinary tract of wild type mice, while isogenic bacteria lacking FimH did not. Interestingly, this response is controlled by TLR4 expression and is abolished when we used FimH− UPEC, even in the presence of TLR4. These data are similar to the PMN response seen following infection with a type I fimbriated derivative of non-adhesive E. coli [48]. Because these UPEC strains share expression of another TLR4-activating PAMP (LPS), these data suggest that the dominant innate immune-activating PAMP on uropathogengic E. coli may in fact be FimH. In support of this, deletion of fimH resulted in decreased colonization of the bladder as reported previously [33] but the level of colonization by fimH− UPEC was similar in a B6 and TLR4−/− background. These data suggest that FimH-independent signaling through TLR4 is not likely a major contributor to infection control in B6 mice.
This is the first report to show that FimH has potent innate antiviral activity which also requires TLR4, MyD88, Trif, IRF-3 and type 1 IFNs, particularly IFN-β. So far, TLR ligands such as dsRNA, ssRNA, and CpG DNA have been associated with the induction of innate antiviral immunity. Protein ligands of TLRs have not been associated with the induction of innate antiviral immunity. Here, however, we demonstrate that FimH, but not LPS, mediates innate antiviral activity at the genital mucosa. It is of particular interest that both TLR5 and TLR11 ligands signal via MyD88, whereas FimH protein signals through both MyD88 and Trif, leading to activation of the IRF-3 pathway. Results from this study may provide the basis for a novel mucosal innate microbicide for a vast variety of mucosal viral infections such as HSV-2, HIV-1 or other sexually transmitted infections.
Female C57BL/6, 129SVPasCrl mice, 8–12 weeks old, were purchased from Charles River Laboratory (Quebec, Canada). TLR4−/− mice were purchased from Jackson laboratory (Bar Harbor, USA). Breeding pairs of IFNα/βR−/− were kindly provided by Rolf M. Zinkernagel (Zürich, Switzerland). Breeding pairs of IRF-3−/−, MyD88−/− and Trif−/− mice were kindly provided by Dr. T. Taniguchi (via Dr. T. Moran), Dr. S. Akira (via Dr. D. Golenbock) and Dr. B. Beutler, respectively. All mice were housed in level B rooms which followed a 12 hour day and 12 hour night schedule, and were maintained under standard temperature controlled conditions.
RAW264.7, HEL fibroblasts and BJ fibroblasts cells were purchased from ATCC. B6, IRF-3−/−, MyD88−/−, Trif−/− and IFNα/βR−/− murine embryonic fibroblasts (MEFs) were prepared from gestation day 13.5 in α-MEM with 20% FBS and weaned then grown in 10% α-MEM for experiments. 293, 293-hTLR4 and 293-hTLR4-CD14/MD2 cells were purchased from InvivoGen and maintained in 10% DMEM supplemented with 10ug/mL blasticidin (293-hTLR4) or 10ug/mL blasticidin and 50ug/mL HygroGold (293-hTLR4-CD14/MD2). HSV-2 strain 333 was grown and titred as previously described [49]. VSV expressing GFP was kindly provided by Dr. Brian Lichty (McMaster University, Hamilton, ON). GM-CSF was purchased from R&D. α-D-manosidase, LPS (L26-54) and Poly I:C were purchased from Sigma (Oakville, ON, Canada). Depo-Provera was purchased from Upjohn (Don Mills, ON, Canada).
The fimH gene from avian pathogenic E. coli strain EC99 (O78) was cloned into pQE-30 and expressed in BL-21 competent E. coli. FimH expression and purification were performed as previously described [50]. Briefly, Protein expression was induced by adding 1M IPTG to a final concentration of 1 mM and induction continued for a period of 5 hours. Bacterial pellets were lysed and protein isolation continued under denaturing conditions utilizing Ni-NTA affinity chromatography. Isolated protein fractions were then dialyzed in a 10-kDa Slidlyzer dialysis cassette against PBS. The LPS contraction in the purified FimH protein was determined using Limulus Amebocyte Lysate LPS detection kit according to the manufacturer's protocol.
RAW264.7 cells were treated with FimH (10 µg/ml), Poly I:C (10 µg/ml) or left un-treated. Twenty-four hours post treatment the supernatants were collected and stored at -20°C for further study. The cells were then infected with HSV-2, MOI of 0.1. Twenty to twenty-two hours post infection the cells and supernatants were collected for HSV-2 titration on Vero cells. Passage 3 MEFs from B6, IRF-3−/−, MyD88−/−, Trif−/− and IFNα/βR−/− mice were split into 12-well plates and then treated with various concentrations of FimH, Poly I:C or LPS or left un-treated. Twenty-four hours post treatment, MEFs were infected with VSV-GFP. Levels of GFP fluorescence were visualized and quantified using a Typhoon™ scanner (GE Healthcare) 24 hours post-infection.
B6, TLR4−/−,IRF-3−/−, MyD88−/− and IFNα/βR−/− mice, 6–8 weeks old, were subcutaneously (sc) injected with 2 mg of progesterone/mouse (Depo-Provera). Four days later the mice were anaesthetized and treated vaginally with FimH (40 µg/mouse) or Poly I:C (100 µg/mouse). Twenty-four hours after treatment the mice were anesthetised, placed on their backs, and infected IVAG with a lethal dose of HSV-2 in 10 µl of PBS for at least 45 min while being maintained under anaesthesia. Vaginal washes were collected daily after infection (days 1–3) by pipetting 2×30 µL of PBS into and out of the vagina 6–8 times. Viral titers in IVAG washes were determined by plaque assay on monolayers of Vero cells as previously described [49]. Treated mice were also monitored daily for genital pathology and survival for up to 4 weeks. Pathology was scored on a five point scale. Zero indicated no infection; 1, slight redness of external vagina; 2, swelling and redness of external vagina; 3, severe swelling of external vagina and hair loss in the surrounding area; 4, ulceration of vaginal tissue, redness and swelling; 5, continued ulceration, redness, swelling and sometimes paralysis in back legs, at which point the mice were euthanized.
To study the effects of FimH or LPS on vaginal tissue morphology, progesterone-treated mice received FimH (40 µg/mouse) or LPS (5 µg/mouse). After 24 h, vaginal tissue was removed, fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 5 µm for hematoxylin and eosin staining.
IFN-α, IFN-β, IFN-γ and IL-8 ELISAs were conducted using Quantikine Murine Kits from R&D Systems (Minneapolis, MN, USA) according to the manufacturer's instructions. IFN-α and IFN-β ELISAs were conducted using PBL Biomedical kits from PBL (Piscataway, NJ, USA). The IFN-α ELISA kit detects mouse IFN-αA, IFN-α1, IFN-α4, IFN-α5, IFN-α6, and IFN-α9, with a detection limit of 10 pg/ml.
A human cystitis isolate of uropathogenic Escherichia coli was used for experimental urinary tract infection of mice. E. coli NU14-1, which does not express FimH due to a disruption of the fimH gene, and E. coli NU14, which is the isogenic wild type parent strain, were kindly provided by Dr. Scott Hultgren (Washington University, St. Louis, MO). E. coli were cultured in LB broth with streptomycin at 50 µg ml. For mouse infection studies, bacteria were grown overnight in LB broth, washed in 0.85% saline, and resuspended in saline to a concentration of ∼109 colony forming units (cfu) per ml. B6 mice and TLR4−/− were infected with 0.1 ml (108 cfu) of bacterial suspension. A soft catheter (0.7 mm) placed in the urethra of anaesthetized mice and the bacterial were delivered into balder. Urine was collected 0, 2, 6 and 24 hours post-infection and polymorphonuclear leukocytes were quantified using a haemocytometer. Twenty-four hours after infection, the bladders were removed, homogenized and the bacterial load was enumerated.
Statistical differences among the viral titers were determined by analysis of variance followed by Tukey's test. The statistical significances of the survival rates and the percentage of GFP expressing cells were determined by the χ2 test. A P value of <0.05 was considered statistically significant. An unpaired t test was used to determine significant differences in cytokine production.
|
10.1371/journal.pgen.1006116 | Super Resolution Fluorescence Microscopy and Tracking of Bacterial Flotillin (Reggie) Paralogs Provide Evidence for Defined-Sized Protein Microdomains within the Bacterial Membrane but Absence of Clusters Containing Detergent-Resistant Proteins | Biological membranes have been proposed to contain microdomains of a specific lipid composition, in which distinct groups of proteins are clustered. Flotillin-like proteins are conserved between pro—and eukaryotes, play an important function in several eukaryotic and bacterial cells, and define in vertebrates a type of so-called detergent-resistant microdomains. Using STED microscopy, we show that two bacterial flotillins, FloA and FloT, form defined assemblies with an average diameter of 85 to 110 nm in the model bacterium Bacillus subtilis. Interestingly, flotillin microdomains are of similar size in eukaryotic cells. The soluble domains of FloA form higher order oligomers of up to several hundred kDa in vitro, showing that like eukaryotic flotillins, bacterial assemblies are based in part on their ability to self-oligomerize. However, B. subtilis paralogs show significantly different diffusion rates, and consequently do not colocalize into a common microdomain. Dual colour time lapse experiments of flotillins together with other detergent-resistant proteins in bacteria show that proteins colocalize for no longer than a few hundred milliseconds, and do not move together. Our data reveal that the bacterial membrane contains defined-sized protein domains rather than functional microdomains dependent on flotillins. Based on their distinct dynamics, FloA and FloT confer spatially distinguishable activities, but do not serve as molecular scaffolds.
| Many membrane proteins are not uniformly distributed within biological membranes, and may prefer specific lipid environments to function optimally. Using super resolution fluorescence microscopy, we show that several Bacillus subtilis membrane proteins indeed cluster into structures of 60 to 110 nm, verifying the existence of defined-size protein microdomains. Biochemical co-isolation of specific membrane proteins and flotillins, a family of proteins highly conserved between eukaryotic and bacterial cells, suggested that common “functional” microdomains exist, containing so-called “detergent-resistant” membrane proteins, that are centered by flotillins. Through high speed tracking of Bacillus subtilis FloA and FloT we show that both proteins are not present in the same microdomain, but move through the membrane with different velocities. Dual colour time lapse microscopy showed that contrarily to vertebrate flotillins, bacterial flotillins do not move together with detergent-resistant proteins, ruling out the existence of coclusters. The lack of both flotillins, but not of a single one, leads to striking defects in cell shape and in cell growth, indicating important overlapping functions of flotillin paralogs. Our data show that FloA and FloT perform spatially distinct functions, possibly in the insertion of membrane proteins that require a specific lipid environment, based on a close connection between FloA and FloT with the Sec membrane insertion machinery, but do not act as scaffolds for detergent resistant proteins. Our tracking analyses provide an important basis for the understanding of interactions between membrane proteins in living cells.
| In spite of many decades of research on membrane proteins, the true arrangement of proteins and their dynamics within the lipid bilayer are still poorly defined. Many membrane proteins show non-uniform localization patterns [1, 2], and the existence of microdomains having different lipid compositions can be inferred from several lines of experiments [3]. So-called detergent resistant microdomains (DRMs) or lipid rafts have been studied biochemically and cytologically, because they contain a characteristic set of proteins that are involved in a variety of processes [4–8]. However, how lipid domains are set up and are maintained, and how fast they move within the cell membrane remains unclear.
Flotillin/reggie proteins (reggies/flotillins, prohibitins, podocins, stomatins) are an evolutionarily conserved class of proteins found across all organisms [9]. They are considered as regulators of membrane protein trafficking [10] and as common constituents of DRMs in eukaryotic cells. The hallmark of flotillin-like proteins is the SPFH domain (stomatin, prohibitin, flotillin homology) of unknown function, and in general, a single membrane span (with the N-terminus of the protein being on the outside of the cell) in bacterial cells, or no membrane helix but a palmitoyl and myristoyl anchor in eukaryotic cells [11]. In addition to the SPFH domain, flotillin subfamilies contain the so-called flotillin (tail) domain, which is characterized by extended coiled coil motifs and is involved in multimerization [2], but has no known enzymatic function. In eukaryotic cells, flotillins are involved in membrane-trafficking, in signal transduction, and cytoskeletal rearrangement [3, 10]. They are also discussed as scaffolding proteins and as couplers of membrane-proteins with the actin cytoskeleton [12, 13]. During axon growth in neuronal cells, flotillins are suggested to induce membrane microdomain formation at the growth cone [14, 15] and to recruit specific proteins to the elongating axon [16]. Furthermore, flotillins appear to be involved in Alzheimer’s and Parkinson’s disease, and other phenomena [7, 17]. Defects in flotillin proteins are particularly evident in neurons which fail to extend axons and at the recycling compartment of HeLa and A431 cells which fail to properly recycle the transferrin receptor and E-cadherin [10, 18].
In fungi, flotillin proteins are not conserved in budding and fission yeast but are present in ascomycetous filamentous fungi. In the model filamentous fungus Aspergillus nidulans, deletion of the flotillin gene, floA, impaired a special membrane domain at hyphal tips, which resulted in defects of polarized growth of hyphae, suggesting a conserved function of flotillins in the fungus and in neurons [19, 20].
Bacterial flotillins have so far been characterized in Escherichia coli, where a genetic link to a membrane-associated protease was found [21], and in Bacillus subtilis, which has two paralogs, FloT and FloA. The absence of one of the flotillins has only minor effects (e.g. a delay in the differentiation process of sporulation [22]), but the absence of both proteins has very severe effects: loss of proper cell shape, a defect in cell division [23], altered membrane fluidity [24, 25], and a defect in signaling events during the transition from planktonic to biofilm growth style [26]. The severity of the defects (including reduced growth rate) highlights the important function of flotillins in bacterial cells.
Using special detergents, B. subtilis flotillins can be co-isolated with NfeD proteins of unknown function, with the signaling receptor KinC [26], cell wall synthesis enzyme Pbp5, secretory protein SecY, membrane transporters like FhuD, as well as energy metabolism protein AtpDG [27]. Therefore, flotillins have been suggested to set up microdomains within the membrane, by recruiting other proteins and possibly specific lipids into the special structures. It has recently been shown that the overproduction of flotillins increases the stability of a protease, FtsH, within the membrane, which in turn affects cell division and other membrane-associated processes [28]. Indeed, the deletion of both, the gene encoding flotillin T protein in B. subtilis and of dynA (coding for B. subtilis dynamin) results in cell filamentation [29] and a defect in motility [30]. A flotillin double mutant (floT and floA) in B. subtilis also shows a cell filamentation defect [23]. Overproduction of flotillin T results in the considerable shortening of cells [25], supporting the idea that flotillins are directly involved in the division process. Moreover, the absence of flotillin or their overproduction has been shown to affect membrane fluidity [24, 27], and flotillin T has been co-isolated with negatively charged phospholipids, e.g. with phosphatidylglycerol and cardiolipin [22]. The latter is known to facilitate membrane bending, so it is well conceivable that flotillin’s association with this lipid such as cardiolipin facilitates membrane bending, or membrane fusion, possibly performed by dynamin; both processes are crucial steps during bacterial cell division.
Based on conventional fluorescence microscopy, these domains could have a size of 250 nm and more, which are substantial fractions of the surface of a 2 to 4 micrometer large cell. Electron microscopic imaging has shown that flotillin structures are equal or smaller than 100 nm in human cell lines [8, 31]. However, in most cases, flotillins have been visualized by conventional light microscopy, which cannot resolve objects smaller than 250 nm in diameter. Such structures could be interpreted as large raft-like membrane domains. Thus, a description of flotillin assemblies in live organisms is still lacking.
We wished to obtain a better picture of the size and structure of flotillin assemblies, in bacteria as well as in eukaryotic cells, and to study their dynamics in live bacterial cells. We developed a tracking procedure that can accurately determine diffusion constants of proteins moving along a curved membrane, and provide data on the time scale in which flotillin assemblies can meet in the membrane, or in which flotillin structures can cover the entire surface of a cell with regard to their size and diffusion rates. By comparing the movement of flotillins and other proteins proposed to be a part of DRMs, we show that in spite of the formation of conserved-sized individual protein domains, these do not track together, indicating that many bacterial proteins are self-organized into defined assemblies, but that mixed protein compartments only exist transiently.
To gain insight into the nature of bacterial flotillin assemblies, which have been speculated to constitute lipid microdomains, we employed stimulated emission depletion (STED) superresolution fluorescence microscopy. Using G-STED, we have achieved a resolution of 42 nm for MreB filaments in live B. subtilis cells [32]. We visualized two flotillin-YFP fusions, FloA-YFP and FloT-YFP, in live, exponentially growing or stationary phase B. subtilis cells. Both fusions were shown to functionally replace the wild type proteins, when expressed from the native gene locus, or from an ectopic site on the chromosome, in previous work [22, 23, 26]. For example, cells carrying deletions of both, floA and floT genes, grow much slower than wild type cells, are twisted and bent and highly elongated, while single floT or floA mutant cells do not show any of these strong phenotypes. Likewise, cells lacking floA and expressing FloT-YFP (or FloT-mVenus), or vice versa, grow like wild type cells. Thus, the fluorescent protein fusions are able to fulfill the functions of the wild type proteins.
For both fusions, we rarely observed structures of 50 nm in diameter (Fig 1A), but usually larger sized foci, determined using the Leica software (see below), showing that the assemblies are above the resolution of STED microscopy. FloT has been shown to move within the membrane in a time frame of a few seconds, with some static and some motile assemblies [23], potentially causing a drift during acquisition time. Round objects that move during confocal acquisition have a characteristic comet-like structure. We used a scan speed of 400 Hz, which is fast enough to localize flotillin-YFP foci with high efficiency, because we observed few foci with a comet-tail (indicated in Fig 1C); such cases were rare (4%, n = 250 foci analysed). It should be noted that many flotillin foci were oval or irregularly shaped (Fig 1C), suggesting that not all assemblies are uniformly round. Therefore, our experiments adequately reflect the size of flotillin-YFP assemblies in live cells.
Interestingly, a majority of FloT-YFP signals had a size of more than 50 nm. We measured 225 foci from 125 cells and determined an average size for FloT-YFP assemblies of 85.3 nm ± 12.5 nm (SD) (Fig 1A), and a maximal size of 97.0 nm. The size of fluorescent foci was very similar between FloT-YFP expressed from the original gene locus, or expressed ectopically from the amylase locus (Fig 1B). Thus, FloT assemblies have a relatively uniform size in live cells. We also imaged a monomeric variant of YFP, mVenus, fused to FloT (Fig 1C), which yielded a size distribution that closely resembled that of FloT-YFP, ruling out an artifact from FP-induced multimerization. The diameter of FloT-YFP foci did not change considerably between exponential growth and stationary phase (Fig 1D), but the number of foci increased during the transition to stationary growth, as was reported earlier [23]. We performed the same experiments for the second flotillin-like protein in B. subtilis, FloA. Like the FloT-YFP fusion, a FloA-YFP fusion can functionally replace the original protein [23, 26]. Interestingly, FloA-YFP foci had an average size of 80.4 nm ± 10.9 nm (SD) (Fig 1E), and thus displayed a similar size distribution like FloT-YFP. These experiments reveal that bacterial flotillin-like proteins have a preferred size for their assemblies, and suggest that flotillins form defined assemblies, however with a considerable size-variation.
Under standard (exponential) growth conditions, bacteria have a preferred composition of the membrane in terms of ratios of different lipids. We wished to gain insight into the question whether flotillin assemblies may be influenced in number and/or size in cells lacking different lipids. We used four strains that have been generated for B. subtilis by the Helmann group, namely carrying a deletion of the major cardiolipin synthetase (clsA) [33], a deletion in the phosphatidyl-ethanolamine pathway (pssA), a deletion in the lysylphosphatidyl-glycerol pathway (mprF), or having a lack of glycolipids (ugtP) [34]. We used B. subtilis strain W168 as background, rather than PY79. Interestingly, we found that flotillin assemblies are larger in strain 168 than in strain PY79: FloA-YFP foci were 109 ± 9.7 nm (S1A Fig), while FloT-YFP foci were of a size of 106.7 ± 12.7 nm (Fig 1I, n = 38). These results show that strain backgrounds can have an effect on the size of flotillin clusters. Next, we introduced FloA-YFP and FloT-YFP fusions into the different mutant backgrounds and analysed their localization. Glycolipid mutant cells displayed reduced and abnormal cell size/shape (S1B and S1H Fig), all other mutant cells had wild type-like appearance. There was no visible difference in the apparent size of the clusters (S1B to S1H Fig), which was confirmed by STED measurements: flotillin focus sizes varied between 105 and 115 nm (Table 1), revealing that the assembly of flotillin domains is not influenced by the lack of the different lipids tested. However, the experiments do not rule out other, more specific lipid requirements for the regulation of flotillin oligomerization.
Flotillin/reggie proteins are notoriously difficult to purify for biochemical experiments. We were successful in purifying the soluble part of FloA (Fig 2A) via Ni-NTA chromatography. CD analysis showed that the soluble part has a defined fold, with mostly alpha helical arrangement (60.8%), and few beta-sheet elements (prediction 7.1%, using the GOR IV secondary structure prediction program (https://npsa-prabi.ibcp.fr) (Fig 2B). This is in agreement with the predicted and solved structures of the SPFH domain, which contains a β-sheet, and otherwise a large number of α–helices [35, 36], and with the prediction from coiled-coil analysis software that the flotillin domain contains a high degree of heptad-rich repeats [23]. To investigate if the assembly of flotillins into the defined structures is mediated through self-interaction, we performed analytical gel filtration (GF) and sucrose gradient centrifugation. The protein eluted from gel filtration just behind the void volume, in a peak above 670 kDa, besides a smaller low molecular weight (LMW) fraction (Fig 2C and 2D), whose size corresponds to a FloA dimer (60 kDa). These data suggest that the soluble part of FloA forms predominantly multimeric structures. To rule out artefacts caused by the use of a hexa-histidine affinity tag, we also purified FloA using a strep-tag, which yielded a similar preference for the formation of high molecular weight (HMW) polymeric structures eluting over a wide range of sizes, and a much smaller LMW part eluting around 120 kDa (S2A Fig), which would correspond to a tetramer of FloA as smallest unit. It is possible that the histidine-tag weakens tetramer formation. We used the high or low molecular weight fractions from GF columns for sucrose gradient experiments. Interestingly, when the LMW fraction of FloA was analyzed for its distribution in sucrose gradients, most of the protein sedimented as dimers or tetramers, and some degree of multimerization was observed (Fig 2E) Likewise, when the multimeric fraction was separated on a sucrose gradient, only a small amount of LMW protein arose, and a majority of molecules remained in the multimeric state, with a peak above 670 kDa (Fig 2F). To obtain more insight into the exchange between multimeric assemblies and smallest FloA units, different concentrations of HMW or of LMW fractions of FloA-Strep were subjected to GF analysis. S2B Fig shows that 3 fold and 5 fold dilution of the HMW fraction leads to a concomitant decrease in HMW and in LMW formation. We measured the peak area of both fractions, which showed that in undiluted or 3-fold diluted conditions, about 90% of the proteins are in the HMW fraction, and in the 5-fold diluted condition, 82% are in the HMW fraction and 12% in the LMW peak. A 2 fold concentration of the LMW fraction leads to an increase in formation of multimers relative to the LMW fraction, albeit only of multimers up to 670 kDa (S2C Fig). These experiments show that there is an equilibrium between the smallest unit and multimeric fractions; the tetramers readily form multimers when the concentration is increased, and the equilibrium is far on the multimer side, showing that these are rather stable structures. If only peak fractions from both methods are taken into account for native molecular mass calculation (750 kDa from GF, 750 kDa from sucrose gradients), the formation of 25mer structures can be deduced, but the size variation is very large, such that 12mer up to 50mer assemblies can be observed in vitro. The large variation of subunit number was also observed using electron microscopy (EM), after uranyl acetate negative staining of purified FloA multimers. When the LMW fraction was used for EM, homogeneous small structures were observed (Fig 2G), whereas a wide range of different sizes of multimeric particles exists when the HMW fraction was imaged (Fig 2H and S3 Fig), the largest of which had a size of 70 nm.
These experiments show that the flotillin structures observed in vivo are in part mediated through flotillin self-interaction, as shown for flotillin and prohibitin in vertebrate cells [2, 37].
To investigate if flotillin foci are indeed composed of a considerable number of monomers, we performed single molecule microscopy, in which cells are exposed to a focused laser illumination. Bleaching of fluorescent spots is monitored in stream acquisitions using a fast EM-CCD camera, such that single bleaching steps can be monitored [38, 39] (S1 Movie). Gradual bleaching of molecules within a fluorescent focus leaves single fluorescent-protein spots that bleach in a single step, revealing the intensity of a single chromophore. Initial intensity of spots can then be used to determine the number of fluorescently-labeled subunits. Fig 3A and 3B show representative images from an experiment, in which initial fluorescence in two foci is reduced to a single spot that finally bleaches in one step. In Fig 3C to 3E, examples for bleaching kinetics of two FloT-YFP spots are shown, in which the intensity of a single YFP molecule can be seen towards the end of the acquisition, and total intensity of the spot in the first frames. FloT-YFP expressed as sole source in the cell showed an average of 12 bleaching steps in 30 spots analysed, revealing that at least 12 molecules are present within the 75 nm structures. The true number of molecules is somewhat higher, because a subset of fluorophores will not have matured, or may have been bleached at an earlier time point in the acquisition (i.e. before the first frame that is used in the analysis). FloA-YFP foci had a fluorescence intensity that was very close to that of FloT-YFP, and therefore, also FloA assemblies will contain a number of monomers, in agreement with the formation of multimers in vitro. Interestingly, fluorescent foci frequently showed regain of fluorescence (S1 Movie and Fig 3D and 3E), revealing that non-bleached molecules can be recruited into the assemblies, which occurred in 50% of the analysed spots. Bleaching occurred over 100 to 150 frames, i.e. within 2 to 3 s, showing that flotillin assemblies have subunit exchange within the frame of few seconds.
Flotillin microdomains have been visualized and measured by various microscopic methods. The most accurate method used so far, electron microscopy, revealed assemblies of down to 100 nm in human cell lines. Because STED is highly suitable to determine the size of flotillin assemblies in live cells, we analysed FloA-GFP in the model filamentous fungus, Aspergillus nidulans [19], and reggie-1 (flotillin-2)-GFP labeled human cells.
In the hyphae of A. nidulans, FloA-GFP was detected as defined foci along the hyphal cortex but was excluded from the hyphal tip [19]. In young fungal hyphae, flotillin foci were present as defined foci, with an average diameter of 68.7 nm ± 8.1 nm (SD) (Fig 4A and 4C). Older hyphae were visually distinct from younger ones in that they contained large vacuoles, and contained more flotillin signals (Fig 4A), which could reach sizes of 150 nm. The reason for this increase in size dependent on hyphal age is unknown. Possibly, due to the visual increase in number, closely adjacent assemblies can appear as a large double-sized structure. In any event, in cells having fewer and well-spaced foci, flotillin assemblies have a size that is very similar to that of bacterial flotillin.
In HeLa cells, flotillin is found at the cytoplasmic face of the plasma membrane, but also at various internal membrane structures [10] (Fig 4B). For simplicity, we focused our measurements on flotillin assemblies at the cell membrane, which was ensured through making a Z-stack using conventional confocal images, and a G-STED image at the cell periphery (Fig 4B). Flotillin-2-GFP signals also had an average size of 94.1 nm ±7.4 nm (SD), and were thus also of relatively well defined size. In MCF-7 cells, flotillin signals measured only 81.5 ± 5.0 nm (SD) (Fig 4D and 4C), even more closely resembling bacterial flotillin assemblies (Fig 4C). Therefore, the average size of flotillin structures in eukaryotic cells is quite conserved and comparable to those in bacterial cells.
Movement of flotillin proteins has been described as “dynamic”, however, given exposures with 3 s or 2 min intervals [22, 23], almost every membrane protein is expected to be dynamic, i.e. to move between time points of image acquisition. We therefore performed fast time lapse experiments using 1 s intervals (and 0.3 s exposures), or 0.3 s stream acquisitions. Interestingly, dynamics of FloT and of FloA were clearly distinguishable, in that most FloT-YFP signals moved visibly slower than FloA-YFP signals (compare S2 and S3 Movies), and that more static FloT-YFP foci than for FloA-YFP were present in cells. In order to quantify the differences in movement, we automatically tracked and analyzed flotillin foci using Trackmate (ImageJ plugin) software. Fig 5A shows an example of 9 FloA-YFP tracks, overlaid with the first frame of the acquisition, in a single cell, and Fig 5B shows the velocity of the 9 trajectories over time. We used this strategy with a total of 3152 FloA Tracks and 1024 for FloT. Fig 5C shows 2 cells with about 50 trajectories, whose mean squared displacements (MSDs) are plotted in Fig 5D, revealing that most FloA tracks were dynamic, and only a minority was static. A comparison of velocities between FloA (Fig 5D) and FloT (Fig 5E) clearly shows that more static FloT-YFP foci are present in cells, and that FloA-YFP foci in general showed a higher degree of movement.
To gain more insight into the nature of flotillin movement, we determined their diffusion rates. This was done by applying a weighted linear fit to the ensemble mean of all tracked MSD curves of FloA and FloT (Fig 5F). Weights were chosen to take the number of measurement points for each delay time. It is evident that the diffusion of FloA is faster than that of FloT: a fit to the curves revealed a mean diffusion rate of 0.0056 μm2/s for FloA-YFP, and 0.0018 μm2/s for FloT-YFP. These data show that FloA is about three times as fast as FloT, while both proteins are factor 3 or 9 slower than the large (500 kDa) Tar receptor of chemotaxis, and factor 3.5 to 5 fold (11 to 16 fold for FloT) slower than two other 190 kDa membrane proteins [40]. These data strongly support the idea that flotillins form large assemblies within and at the membrane, which move at considerably different speeds.
As a complication in the tracking of membrane proteins, we considered that the speed of movement of molecules along the circumference of a rod-shaped cell (called “x-axis”) is underestimated because of the projection on the observation plane. Molecules moving along the length of the cell (“y-axis”) move in absolute correlation with the tracks detected by the camera, while molecules moving in “x”-direction move along a curved surface, and thus travel a longer distance per time than detected by the observer (camera). In order to quantify the difference in movement of “X” versus”Y”, we determined MSDs for molecules solely moving in “x” or “y” direction for several frames. A plot of MSD over time shows that apparent velocities along the x-axis are considerably smaller than those along the y-axis, for both proteins (Fig 5G and S4B Fig). Average projected MSD values for FloA-YFP for x-axial movement was 0.0044 μm2/s, while that for movement in y-direction was 0.0076 μm2/s, and thus about 1.7 fold slower. For FloT, movement in x was 0.0014 μm2/s, and 0.0024 μm2/s in y, so likewise 1.7 fold lower for x than for y.
To use a different approach for the quantification of diffusion rates, we scored the distances travelled between two time points for foci, which yields a plot for the instantaneous velocity (one time lapse displacement) distribution. Fig 5I shows a much more narrow distribution of FloT-YFP movement compared with that of FloA-YFP (Fig 5H). For FloA, the tails of the distribution are visibly broader than for FloT, indicating a higher variance and therefore a higher diffusion constant of FloA. Diffusion constants can be determined using an unbiased estimator based on the covariance of single time step displacements [var(v) = 2*D/dt]. We can calculate a diffusion constant of 0.0069 μm2/s for FloA, and 0.0041 μm2/s for FloT; here, FloA is less than 2 fold faster than FloT. For each analysis, the distribution of instantaneous velocities of FloA and FloT was also determined along the minor (vx) and major (vy) cell axis. In both plots (Fig 5H and 5I), it is apparent that molecules moving along the x-axis move slower (centre more around slower movement), while “y”-tracks can reach much higher speeds. For FloA-YFP, this yields Dx = 0.0057 μm2/s, and Dy = 0.0092 μm2/s. Values for FloT are about two fold lower: Dx = 0.0028 μm2/s, Dy = 0.0051 μm2/s. Thus, MSDs are 1.6 fold to 1.8 fold higher for the y-axis than for x-axis, and again two fold higher for FloA than for FloT, in good agreement with the MSD values determined through the weighted fit of the MSD curves. Given that molecules travelling in a random fashion along the membrane of a cylinder will travel a mixture of 50% in x and 50% in y direction, our analysis can be used to correct tracks of membrane proteins along a tubular bacterium by multiplying all tracks by factor 1.35, as half of the tracks are underestimated by factor 1.7.
Our experiments rule out the existence of a common structure formed by the flotillin paralogs. The considerably different diffusion rates are not compatible with an interaction of flotillins for more than few milliseconds.
With the size of flotillin assemblies at hand and their determined diffusion rate, we can quantify the likelihood of an encounter of the flotillin assemblies within the membrane, which may then exchange interaction partners. Alternatively, if flotillin assemblies are platforms for the organization of membrane proteins that have been inserted and are then released to diffuse by themselves (e.g. KinC), we can estimate the time it takes for assemblies to cover the membrane surface via random movement. Towards this end, we quantified the number of FloT-YFP and of FloA-YFP assemblies in 50 cells, using deconvoluted G-STED images. During exponential growth, cells contained 8.6 ± 2.2 (SD, n = 50) foci of FloT on one side of the cell, and 12.3 ± 4.1 (SD, n = 50) of FloA. Thus, in total, cells have 17 FloT and 25 FloA assemblies on their membrane surface.
The surface of Bacillus is about 9 μm2, given that cells are on average 3 μm long (2 to 4 μm) and 1 μm thick (leaving out the polar regions). If the surface is divided by the number of proteins, each FloA assembly will have an area of 0.36 μm2 for itself, or a square with a side length of 0.6 μm. Given a diffusion constant of (average of the covariance method and the MSD fit times 1.35) 0.0084 μm2/s, the protein will need about 43 s on average, in order to meet another FloA assembly. The size of 80 nm for FloA assemblies means that their size is of minor importance, because on average, there are 600 nm of space between FloA assemblies that need to be bridged via random diffusion for a much smaller sized object. This would change considerably if flotillin assemblies had a larger size, e.g. closer to the diffraction limit of conventional light microscopy (250 nm); such large assemblies would “meet” much more often. In other words, it will take about three quarters of a minute for all FloA assemblies to largely cover the surface of a Bacillus based on their diffusion. For FloT (average corrected diffusion rate 0.004 μm2/s), the average time to “meet” is accordingly 90 s. A static membrane-associated protein would thus be picked up by a flotillin T assembly in one and a half minutes on average.
KinC plays an important role in signal transduction during cell differentiation and biofilm formation, and has been shown to colocalize with FloT and with FloA [26]. A C-terminal fusion to KinC has been shown to support all known functions of the protein [26]. When imaged in G-STED, KinC-YFP formed foci with an average diameter of 68.7 nm ± 6.6 nm (SD) (Fig 1F). Based on the finding that KinC-YFP foci disappear after the loss of FloA and of FloT, and were proposed to colocalize with flotillins [26], these data suggest that KinC is part of the 70 nm FloT and/or FloA structures. However, when we imaged KinC-YFP together with FloA-CFP or with FloT-CFP (expressed at low level from an ectopic site on the chromosome, such that the average number of foci did not increase), even using conventional confocal microscopy and deconvolution (for the removal of background fluorescence), we found only a minor degree of colocalization. We took advantage of the “between lines” acquisition mode during confocal microscopy, in which each line is first scanned with one laser line and then with the second, before moving to the next line, thereby avoiding a drift of signals between the acquisition of the two channels. Lines are scanned with 400 Hz, i.e. the interval between channel acquisition is 2.5 ms. Our experiments clearly establish that KinC rarely colocalized with FloA: only 3% of the signals showed an overlap (Fig 6A, S5 Fig, 220 cells analysed). Colocalization with FloT was 4% (Fig 6C, S5 Fig, 200 cells analysed), indicating that more than 90% of KinC signals do not colocalize with any flotillin, given that FloT and FloA do not colocalize [23], which can be seen in Fig 6B (less than 2% colocalization). Therefore, flotillins have a similar size as KinC assemblies, but cannot be the architectural basis of the latter. Our data clearly rule out that flotillins and KinC are part of the same “microdomain” structures, and show that flotillins do not have an appreciable spatial overlap, as has been proposed before [27, 28].
We analysed a second protein, NfeD2, that has been described to colocalize with flotillins (FloT), and whose localization into discrete foci depends on FloT [23]. NfeD2 is cotranscribed with floT (yuaG), which is a common feature of NfeD encoding genes. NfeD2-YFP also formed defined foci, however, with an average size of 94.9 nm ± 10.1 nm (SD) (Fig 1G), which are considerably larger than the structures of its associated partner, FloT. In the absence of FloT, no defined NfeD2-YFP assemblies were visible, but only delocalized fluorescence throughout the membrane was detectable [23]. On the other hand, more than 96% of FloT-YFP and NfeD2-CFP colocalized (Fig 6D, S5 Fig), and consequently, NfeD2 must be part of the (85 nm sized) assemblies of FloT. Given the larger size of NfeD2 assemblies, it appears that the protein forms an outer “rim” around the FloT structure. We also tested the effect of a deletion of nfed2 on the localization of FloT-YFP. Although the visual impression of FloT-YFP foci in epifluorescence was that of more foci with less intensity [23], FloT-YFP foci still had an average size of 85 nm (S5C Fig). Thus, the absence of NfeD2 has no visible effect on the structure of FloT assemblies. We wondered if NfeD2 provides an important function for the activity of FloT assemblies. An nfeD2 deletion has no discernable phenotype [23]. To test if the lack of NfeD2 has a cryptic effect, we generated a floA nfeD2 double mutant strain, because the absence of FloA and FloT leads to a strong phenotype, in contrast to single floA, floT or nfeD2 deletions [23]. However, floA nfeD2 double mutant cells did not reveal any discernable defect in growth or in cell morphology (S5E Fig), revealing that NfeD2 is not essential for the known functions of FloT (i.e. it does not impair the functions that lead to the synthetic phenotype with FloA). In any event, NfeD2 is a clear example of a protein that colocalizes with a flotillin, showing that the rare flotillin/KinC co-localization is not a technical artifact.
Based on the different dynamics of flotillin paralogs, we investigated the putative interaction partners NfeD2-YFP and KinC-YFP using time lapse microscopy. While NfeD2-YFP showed similarly slower movement like FloT (S5 Movie), KinC-YFP signals moved much faster than FloT assemblies (S4 Movie). Thus, NfeD2 will track with FloT, because of the major degree of colocalization of the two, while KinC will not track with FloT, because of the different dynamics, reinforcing the non-significant colocalization between the proteins.
The lack of colocalization between flotillins and KinC prompted us to perform dual colour imaging with further proteins that have been described to be co-isolated with FloT and/or FloA, and to co-localize with flotillins based on epifluorescence microscopy [27, 28]. We chose FtsH and SecA, because both proteins perform vital functions for the physiology of the cell, and because both proteins have been described to belong to the DRM fraction. Interestingly, the absence of both, FloT and FloA has been shown to lead to a depletion of FtsH from the cell membrane [28]. We used FP fusions that have been shown to functionally replace the wild type proteins [27, 28, 41], and which also in our hands do not show a discernable phenotype when expressed as sole copy from the original gene locus. S6A Fig shows that there is some degree of colocalization between FloT-CFP and FtsH-YFP (5% of foci), and between FloA-CFP and FtsH-YFP (Fig 6F, S6C Fig, 6% colocalization). Flotillins and FtsH were frequently found to be present at sites of cell division (S6A and S6B Fig), in agreement with the findings that both types of proteins play an important role in cell division [23, 42]. These experiments suggest that the vast majority of flotillin assemblies do not form a common structural assembly with FtsH. As a control, we analysed if FloT would colocalize with itself, using two FloT alleles that are fused to CFP or to YFP, respectively. S6F Fig shows that both populations mixed, to yield more than 90% colocalization. We tested three further proteins that were reported to interact with flotillins: histidine kinases PhoR and ResE, and oligopeptide permease OppA [43]. S7 Fig shows representative images of dual colour experiments. 4.5% of OppA-YFP foci colocalized with FloT-CFP (S7A Fig, 118 cells analysed), 6% of RecE (S7B Fig, 90 cells analysed) and 7% of PhoR-YFP (S7D Fig, 78 cells analysed each). 12% of PhoR-YFP signals colocalized with FloA-CFP (S7C Fig, 43 cells analysed). These data show that flotillins colocalize with other membrane proteins in few cases, and the majority of flotillin domains do not contain any visible numbers of membrane proteins we have tested.
To further support the finding that flotillins and other DRM proteins form distinct microdomains, and to investigate if transient interactions exist between DRM proteins, we performed two colour time lapse experiments. Joint formation of protein microdomains would mean co-migration of signals, while absence of joint movement would agree with the existence of generally separate entities. Fig 6F shows two intervals of such an experiment, where a colocalization event between FloA and FtsH (upper panel) no longer exists after 200 ms (lower panel), while in the second interval, a colocalization can be seen at a different place in the cell. S6 Movie shows that the overwhelming majority of signals do not move together, in agreement with the low degree of colocalization seen in the snap shot images. Similarly, FloT-CFP and FtsH-YFP did not move together in a detectable manner (S7 Movie), and in agreement with a lack of colocalization, FloT and KinC did not show movement of common domains between 200 ms acquisitions (S8 Movie).
The highest degree of colocalization was seen for FloT-CFP and SecA-YFP (22% of the foci, Fig 6E), and between FloA-CFP and SecA-YFP (15%, S6D Fig). However, also in this case, a majority of foci did not colocalize, showing that flotillin T and SecA do not generally form mixed protein structures. We also performed dual colour time lapse microscopy, revealing that the majority of FloT-CFP and SecA-YFP foci moved independently from each other (S9 Movie). A significant number of foci coincided during image acquisition, but we never detected co-migration for more than 2 frames. Thus, FloT and SecA interactions also take place in the sub-second range. It should be noted that colocalization does not necessarily mean interaction, because the protein domains may be within a range of 250 nm (resolution of conventional light microscopy) but not completely coincide based on their size of less than 100 nm. As such, the numbers determined in our analysis are an overestimate of possible interactions. Contrarily, FloT-CFP signals invariably moved together with NfeD2-YFP signals (S10 Movie), while FloT and FloA did not track together as expected (S11 Movie).
We automatically tracked FloA-YFP, FloT-YFP, FtsH-YFP and SecA-YFP, and compared their dynamics using cumulative probability distributions of diffusion coefficients. Fig 7 shows that FloA, FloT and FtsH had clearly distinguishable dynamics, resulting in different diffusion constants as indicated by the vertical lines. Note that the small number of negative diffusion coefficients occur due to the statistical nature of the covariance based estimator (CVE). SecA interacts with the SecYEG translocon, and shows very similar dynamics as FloT. However, based on the colocalization experiments, this cannot be taken as an argument that the proteins move together in a common microdomain. In agreement with the basal level of colocalization of FloA, FloT and of FtsH, and between FloA and SecA, the different diffusion kinetics show that each proteins moves with a different characteristic speed through the membrane.
These experiments validate our findings that flotillins do not form joint microdomains with proteins that have been co-isolated by special detergent conditions (KinC, FtsH, SecA, ResE, PhoR or OppA) for more than few milliseconds, supporting the idea that FloT forms a common structure with NfeD2 (many flotillin genes lie adjacent to genes coding for NfeD proteins), and suggest a close connection between flotillins and the Sec system compared with other tested proteins.
Membranes in eukaryotic cells have been shown to contain asymmetries in lipid composition, and to contain nanoscale, cholesterol-assisted, dynamic and selective protein assemblies, which can coalesce into larger and more permanent “raft” structures [44]. Such mixed protein domains have been shown to play important roles in TCR signalling, HIV assembly, endoplasmic reticulum (ER)-to-Golgi and post- Golgi trafficking to the cell surface, and glycosphingolipid mediated endocytosis. In bacteria, a variety of membrane proteins has been reported to form visible clusters in the cell membrane, and the existence of raft structures or functional microdomains has been deduced from the finding that a fraction of Triton-insoluble proteins exists (so called detergent resistant microdomains, DRMs), which can be co-isolated with flotillins [27, 28]. The latter protein family is conserved between bacteria and eukaryotes, is found in the DRM fraction in both cell types, and is associated with raft structures in eukaryotic cells, where it is involved in the clustering of membrane proteins and in various aspects of membrane trafficking [8, 45]. Our study was performed to determine the actual size of flotillin and DRM protein clusters in living cells, to investigate the dynamics of clusters within the membrane, and to investigate if the two flotillin proteins in B. subtilis, FloA and FloT form joint clusters with other DRM proteins.
A major conclusion from our in vivo analyses is the finding that protein clusters formed by flotillins in bacteria, in a fungus and in human cells have a relatively defined size of 85 to 110 nm, and that likewise, DRM protein KinC forms structures of a similar diameter. However, FloT and FloA do not colocalize with each other, and colocalize only with a minority of FtsH, SecA or KinC assemblies. These findings are substantiated by the determination of the movement of bacterial flotillins, whose dynamics are different by a factor of 2, revealing that the proteins do not track together. Time lapse experiments show that FloT or FloA assemblies do not co-migrate with any of the other investigated proteins, the exception being the co-migration of FloT and NfeD2, which shows a tight connection for these two proteins, in agreement with a conserved connection of the genes in many organisms [9]. Therefore, our study reveals the existence of protein domains in the sub-100 nm range, which show independent movement, and rule out the generation of large multiprotein clusters generated by flotillins, at least in case of the investigated proteins. Co-localization events between flotillins and other membrane proteins are observed for 200 ms intervals, showing that putative interactions between the domains are highly dynamic and transient; it must be noted that the domains may not even physically interact but merely move past each other. Therefore, integral membrane proteins that are co-isolated based on insolubility to cold Triton extraction do not necessarily form joint microdomains, but assemble into largely distinct multiprotein clusters of a characteristic size. It should be noted, that our confocal and epifluorescence microscopy experiments visualize bulk protein, not individual molecules. It is therefore possible, that flotillin microdomains contain few molecules of other proteins. In other words, our analyses do not rule out that a fraction of a given protein can be found within different microdomains, and that proteins may partition into different membrane domains based on diverse affinities.
Interestingly, the size of flotillin clusters was relatively robust against changes in specific lipid environment of the proteins. Under conditions of low levels of cardiolipin, lack of phosphatidyl-ethanolamine, of lysylphosphatidyl-glycerol or of glycolipids, we still observed the generation of 100 nm flotillin domains, suggesting that protein/protein interactions play a major role in the formation of these structures. This is corroborated by our in vitro studies in which purified soluble FloA forms up to 70 nm large structures visualized by electron microscopy, indicating that the assembly of flotillin structures is largely driven by protein/protein interactions. The apparent robustness of microdomain formation against lipid perturbations notwithstanding, it is possible that flotillins themselves organize a specific lipid microenvironment that is important for their as yet unknown function. It will be important to determine in how far lipids affect the formation of assemblies of other membrane proteins.
Flotillin/reggie proteins are involved in several membrane-dynamics in eukaryotic cells, and play an important role in the physiology of bacteria. Absence of both flotillins from B. subtilis leads to compromised growth, loss of proper cell shape and a defect in cell division, besides reduction in motility [23, 30]. None of these severe phenotypes are observed in the absence of a single flotillin, which has been used to support the idea that the two proteins form joint raft structures and recruit e.g. signaling molecules to the microdomains. Our results clearly show that the important function(s) of flotillins is mediated by two independently localizing and moving fractions, and that therefore, the two paralogs are functionally distinguishable.
FloT recruits NfeD2 into a joint assembly, suggesting that the two proteins cooperate in these assemblies, however, we show that the presence of NfeD2 does not have a strong influence on the function of FloT, based on the lack of any phenotype in the absence of FloA and of NfeD2. Multimerization is inherent to flotillins, based on the assembly of the soluble part of FloA into large oligomers in vitro, which are rather stable. Although there is a dynamic exchange of flotillin multimers and the smallest subunits formed (most likely tetramers), the equilibrium lies far on the side of polymer formation. Determination of subunit stoichiometry of FloT-YFP in living cells showed that the assemblies consist of at least 12 monomers, whose exchange occurs in a range of seconds. The diffusion of flotillins, especially of FloT, is very slow compared with other bacterial membrane proteins, in agreement with the large number of flotillin monomers within the assemblies. Based on our calculation it takes flotillins between 45 to 90 seconds to “scan” the entire surface of a rod shaped bacterium. Therefore, it is likely that flotillins respond rapidly to their substrates or binding partners, for example assuming that flotillins aid in the insertion of proteins such as FtsH into the membrane [28].
We show that the movement of molecules on the curved surface of a rod-shaped organism differs dependent on the direction of movement, which is underestimated by a factor of 1.7 in case of B. subtilis (diameter of 1 μm) when molecules move perpendicular to the long cell axis (“x”) compared to the longitudinal movement (“y”). As on average, molecules move 50% in x and 50% in y direction in case of random diffusion (which is the case for flotillins and for all DRM proteins), a factor of 1.35 can be used to correct for the curvature-caused underestimation of diffusion. This calculation can be used for all cells having a similar diameter as B. subtilis, and can be adapted to other cell diameters for accurate calculations of protein dynamics. Our calculation that flotillin microdomains meet in the range of seconds is an important tool to determine interaction kinetics of protein microdomains within the cell membrane in bacteria. An analogous function concerning the activity of eukaryotic flotillins, i.e. the targeted delivery of bulk membrane and specific membrane proteins from internal vesicle pools to strategically important sites has been suggested [8].
If bacterial flotillins do not act as a molecular scaffold within the membrane, what may be their function? In the absence of flotillins, cells suffer from many defects, such as membrane abnormalities, loss of proper cell shape, defects in cell division and in motility, and a delay in sporulation [22, 23, 30]. We propose a model in which bacterial flotillins act as membrane insertion-helpers for membrane proteins that require a special lipid environment. We suggest that flotillins transiently interact with the secretion machinery during the insertion of e.g. KinC and FtsH. This is supported by our finding that FloT shows the strongest degree of colocalization with SecA, the central component of the secretion machinery. Also, the depletion of FloT and of FloA has been shown to lead to a strong reduction of the amount of FtsH within the membrane [28], which is most easily explained through a defect in insertion, unless FtsH becomes more prone to degradation in the absence of flotillins. Since there is a large number of FtsH assemblies that does not colocalize or co-migrate with flotillins, the latter scenario seems rather unlikely. With regard to their different diffusion rates, FloT would be engaged in membrane insertion for longer periods of time, while FloA would be quicker and more abundant to meet with insertion islands.
Tracking of flotillins, of FtsH and of SecA demonstrated distinct diffusion kinetics for each protein. Therefore, our data reveal the existence of protein domains consisting of clusters of the same protein, which in case of flotillins, KinC, SecA and FtsH do not overlap, and are only transiently in close proximity enabling an interaction, which can take place in the range of few hundred milliseconds. Our findings have a profound impact on our view of the mode of movement and interaction of membrane proteins and of the organization of DRM proteins. It will be important to investigate the overall movement of many membrane proteins to obtain a clear view on the two dimensional organization and dynamics of membrane proteins.
For expression of soluble 6xHis-FloA, the coding sequence lacking the first 10 codons was amplified by PCR using chromosomal DNA from the B. subtilis wild type strain PY79. The fragment was further integrated in the expression vector pET24d (Novagen) by NcoI and BamHI restriction ligation and brought into the expression host E. coli BL 21 (DE3) giving rise to the strain FD380. For expression of soluble Strep-FloA, the coding sequence (minus the first 10 codons) was amplified by PCR (the upstream primer contains the Strep tag sequence) using chromosomal DNA from B. subtilis PY79. The fragment was inserted into pET24d by NcoI and BamHI restriction ligation and the resulting plasmid was introduced into the expression strain E. coli BL21 (DE3), yielding strain AHV11.
KinC was visualized as a KinC-YFP fusion protein expressed at the original locus. The last 500 bp coding for kinC were integrated into the vector pSG1164-YFP, using ApaI and EcoRI restriction sites, and PY79 cells were transformed with this construct, selecting for cm resistance (leading to strain FD326). For colocalization studies, floA-cfp was integrated at amyE locus (by the use of the plasmid pSG1192 [46] and expression was controlled by xylose addition. The resulting strain (AS34, floA-cfp::amyEspecR) was transformed with chromosomal DNA of strain FD326 leading to a strain AS38 expressing KinC-YFP and FloA-CFP. To investigate colocalization of KinC and FloT, floT-cfp was integrated at the amyE locus under the control of the Pxyl promoter in the strain FD326, giving rise to AS49. For SecA and FtsH, analogous strategies were used. 500 bp of the 3’ end of each gene were integrated into pSG1164, using ApaI and EcoRI sites, and the resulting strains were transformed with chromosomal DNA from strains AS34 (floA-cfp::amyEspecR) or from AS49 (kinC-yfpcatR floT-cfp::amyEspecR), selecting for spec resistance. Similarly, for the colocalization of OppA, of ResE or PhoR with FloT or with FloA, chromosomal integrants were generated expressing OppA-YFP, ResE-YFP or PhoR-YFP, and were transformed with DNA from the strains expressing FloA-CFP or FloT-CFP from the amylase locus.
For the generation of lipid mutant strains expressing FloA-YFP or FloT-YFP, strain 168 containing either deletions in clsA, or in pssA, or in ugtP, or in mprF, were transformed with chromosomal DNA from strain AS34 (floA-cfp::amyEspecR) or from strain AS49 (kinC-yfpcatR floT-cfp::amyEspecR), selecting for spectinomycin resistance. As control strain, B. subtilis 168 was transformed with chromosomal DNA from strains FD191 (floA-yfpcatR) or from FD295 (floT-yfpcatR), giving rise to FloA-YFP or FloT-YFP expressing 168 wild type strains. All strains are listed in Table 2.
Cells were grown to late exponential or stationary phase in LB rich medium containing the appropriate antibiotics at 37°C and under aeration. Xylose was supplemented to induce expression of proteins downstream of the fusion protein at original locus (0.5% (w/v)) or the fusion protein itself at amyE locus (0.01% (w/v)). For microscopy, 2.5 μl of the culture were spotted on a coverslip and immobilized by an agarose pad (1% agarose (w/v) in S750 glucose minimal medium). STED microscopy was performed at a Leica SP8 LSM confocal microscope equipped with a 100X objective (NA 1.4) and a 592 nm depletion laser. Fluorophores were excited with a pulsed white light laser source at 514 or 488 nm respectively. Photon emission was detected with gated hybrid detectors at the appropriate wavelength. Images were processed with the Leika LAS AF software and where stated deconvolution was performed using the Huygens-algorithm (SVI). For colocalization studies, images were acquired simultaneously, by scanning each line first in the YFP (514 nm) than in the CFP (457 nm) channel. Epifluorescence microscopy was done using a Zeiss Axio Imager A1 equipped with an EVOLVE EMCCD camera (Photometrics) and a TIRF objective with an aperture of 1.45, acquiring images with VisiView (2.1.2, Visitron, Munich) software and using a 515 nm laser for YFP detection and a 445 nm laser for CFP detection. For the determination of subunit numbers, a 514 nm diode laser was directly incorporated into the microscope and was focused on the back focal plane, such that a roughly 5 x 5 μm2 illumination spot is generated in the focal plane. Images were acquired by frame transfer stream acquisition, with exposure times of 20 ms, by a Hamamatsu Image EM2 EM-CCD camera.
The A. nidulans strain, SNT122, expresses FloA tagged with GFP at the C-terminus under the native promoter [19]. Spores of the strain were inoculated in minimal medium + 2% glucose (w/v) and were incubated at 30°C overnight.
MCF-7 and HeLa cells were cultured at 37°C and 5% CO2 in MEM and DMEM respectively, supplemented with 10% FCS, L-glutamine and penicillin/streptomycin. Transfection was carried out using the lipofectamine 2000 transfection reagent (Life Technologies) following the manufacturer’s instructions and cultured for 48 h on poly-L-lysine coated coverslips. Cells were fixed with 4% PFA and prepared for microscopy using Mowiol (Sigma Aldrich). The Reggie-1-EGFP (flotillin-2) vector was described previously [2].
Protein movement was recorded using time lapse microscopy with a frame rate of 1 frame per second and a faster rate of 3 frames per second. The protein positions were tracked using Trackmate, a plugin of the image analysis software ImageJ. The resulting tracks were corrected for drift, to account for movement of the whole cell during acquisition.
To distinguish between movement along (y—direction) and normal (x—directions) to the cell axis, the position and orientation of each observed cell was recorded and corresponding protein tracks were transformed into the coordinate system of the cell.
Diffusion coefficients were obtained using a weighted linear fit of the ensemble mean square displacement curve. Since data points in the MSD curve are highly correlated [47], only the first three points were used in the fit. A second method based on the analysis of the distribution of one time lapse displacements [48] was applied to confirm the results obtained by MSD analysis.
The soluble part of FloA (amino acids 56–331 complemented with a C-terminal Trp residue for better visibility at 280 nm) was first C-terminally fused to a 6xHis tag in the pET24d vector (Novagen) and purified by Ni-NTA affinity chromatography. The protein was heterologously expressed in E. coli BL21 (DE3) cells overnight at 30°C by addition of lactose (1.75% (w/v)), cells were collected by centrifugation and the pellet was resuspended in 30 ml of buffer A (20 mM HEPES, 250 mM NaCl, 20 mM MgCl2, 20 mM KCl, 40 mM imidazole, pH 8). Then the cells were broken in a Microfluidizer (Microfluidics) and the soluble and insoluble fractions were separated by centrifugation (30 min at 30000 g at 4°C). The supernatant was loaded onto a His-Trap column (GE Healthcare), washed with buffer A and the protein eluted with buffer B (buffer A with 500 mM imidazole).
For the analysis without His-tag, the soluble part of FloA was C-terminally fused to a Strep-tag (substituting the His-tag sequence by the Strep-tag sequence in pET24d) and purified by Strep-Tactin affinity chromatography. An E. coli BL21 (DE3) strain harboring the pET24d-floAstrep plasmid was grown in LB media at 37°C until mid-exponential phase. Then the overexpression of FloAstrep was induced by the addition of 1 mM IPTG to the culture and incubated for 2 h at 37°C. Cells were collected by centrifugation, the pellet was resuspended in 40 ml of TE-based buffer (100 mM Tris-Cl, 1 mM EDTA, 150 mM NaCl, pH 8) and the cells were broken by passage through a French Press. The soluble fraction was obtained by centrifugation (30 min at 16000 rpm and 4°C) and loaded onto a Strep-Tactin column (Thermo, IBA). The column was washed with TE buffer with 150 mM, 300 mM and 500 mM NaCl and the protein eluted with elution buffer (100 mM Tris-Cl, 1 mM EDTA, 150 mM NaCl, 2.5 M D-Desthiobiotin, pH 8).
For further purification and analysis of oligomerization, both FloA6His and FloAstrep were subjected to size exclusion chromatography. The FloA6His protein was loaded in a HiLoad 26/600 Superdex 200 pg column and run at 1.5 ml/min in an HEPES-based buffer (20 mM HEPES, 200 mM NaCl, 20 mM MgCl2, 20 mM KCl, pH 7.5), while the FloAstrep protein was loaded in a Superose 6 10/300 GL column and run at 0.5 ml/min in Tris buffer (100 mM Tris-Cl, 150 mM NaCl, pH 7.5).
Folding was analyzed by CD-spectrometry and oligomerization was also determined by a sucrose density gradient (5–20% (w/v)) ultracentrifugation (32500 rpm 4°C, 20 h, rotor SW40Ti, Beckman Coulter Optima XPN-80 centrifuge).
Carbon coated copper grids (400 mesh) were hydrophilized by glow discharging (PELCO easiGlow, Ted Pella, USA). 5 μl of a protein suspension with a concentration of 15 μg/ml was applied onto the hydrophilized grids and stained with 2% uranyl acetate after a short washing step with H2Obidest. Samples were analyzed with a JEOL JEM-2100 transmission electron microscope using an acceleration voltage of either 80 or 200 kV. For image acquisition a F214 FastScan CCD camera (TVIPS, Gauting) was used.
|
10.1371/journal.pcbi.1000685 | Computational Complementation: A Modelling Approach to Study Signalling Mechanisms during Legume Autoregulation of Nodulation | Autoregulation of nodulation (AON) is a long-distance signalling regulatory system maintaining the balance of symbiotic nodulation in legume plants. However, the intricacy of internal signalling and absence of flux and biochemical data, are a bottleneck for investigation of AON. To address this, a new computational modelling approach called “Computational Complementation” has been developed. The main idea is to use functional-structural modelling to complement the deficiency of an empirical model of a loss-of-function (non-AON) mutant with hypothetical AON mechanisms. If computational complementation demonstrates a phenotype similar to the wild-type plant, the signalling hypothesis would be suggested as “reasonable”. Our initial case for application of this approach was to test whether or not wild-type soybean cotyledons provide the shoot-derived inhibitor (SDI) to regulate nodule progression. We predicted by computational complementation that the cotyledon is part of the shoot in terms of AON and that it produces the SDI signal, a result that was confirmed by reciprocal epicotyl-and-hypocotyl grafting in a real-plant experiment. This application demonstrates the feasibility of computational complementation and shows its usefulness for applications where real-plant experimentation is either difficult or impossible.
| Endogenous signals, such as phytohormones, play a vital role in plant development and function, controlling processes such as flowering, branching, disease response, and nodulation. However, the signalling mechanisms are so subtle and so complex that details about them remain largely unknown. In this study, we develop a “Computational Complementation” approach for the investigation of long-distance signalling networks during legume autoregulation of nodulation (AON). The key idea is to use computational modelling to complement the deficiency of an empirical model of an AON deficient mutant with hypothesised AON components. If the complementation restores a wild-type nodulation phenotype, the modelled hypotheses would be supported as reasonable. To evaluate the feasibility of this approach, we tested whether wild-type soybean cotyledons participate in AON, commonly controlled by “real” leaves. The test gave an affirmative result (i.e., cotyledons do have AON activity), which was subsequently confirmed by a graft experiment on real plants. Future applications of this approach may be to test candidate AON signals such as auxins, flavones, and CLE peptides, and other plant signalling networks.
| Legumes are one of the largest families of flowering plants that occupy about 15% of Earth's arable surface; yet they provide 27% of the world's primary crop production and more than 35% of the world's processed vegetable oil [1], signifying their cropping potential. Legumes are also the major natural nitrogen-provider to the ecosystem, contributing roughly 200 million tons of nitrogen each year [2] equivalent to over 200 billion dollars worth of fertiliser replacement value. Underlying this powerful fixation capability is a plant developmental process termed “nodulation”, which results from the symbiosis of legume roots and soil-living bacteria broadly called rhizobia. Yet for a legume plant itself, excessive nodulation may cause over-consumption of metabolic resources and disproportional distribution of internal growth regulators [3], and may interfere with developmentally related lateral root inception and function.
Legume plants have evolved a long-distance systemic signalling regulatory system, known as autoregulation of nodulation (AON), to maintain the balance of nodule formation [3]–[7]. It has been hypothesised that the induction of the nodule primordium produces a translocatable signal Q, which moves through a root-shoot xylem pathway to the leaves. This Q signal, or an intermediate, is detected in the phloem parenchyma of leaf vascular tissue by a transmembrane leucine-rich repeat (LRR) receptor kinase [8] related in structure to CLAVATA1 in Arabidopsis. This kinase is referred to as GmNARK in soybean [9],[10], HAR1 in Lotus [11], and SUNN in Medicago [12]. Q is presumed to be a CLV3/ESR-related (CLE) peptide [13],[14]. The perception of the Q signal by the LRR receptor kinase triggers production of a hypothetical shoot-derived inhibitor (SDI) that is transported to the root to inhibit further nodule initiation. SDI can be extracted from wild-type leaves, re-fed via petiole feeding into loss-of-function mutants, resulting in restoration of the wild-type phenotypes [15]. It is a small, water-soluble, heat-stable and inoculation-dependent molecule. However, other mechanisms involved in AON signalling remain largely unknown, though the pre-NARK events (those setting up the signal transmission and then Q signal transduction) as well as the post-NARK events (firstly KAPP phosphorylation, ensuing transcriptional changes, and then SDI production) are being investigated [10],[15],[16].
To help understand such biological complexities, system modelling has been broadly applied [17]–[19]. From a systematic view, behind the signalling mechanisms is a network of components connected by intricate interfaces, with activities such as “assembly, translocation, degradation, and channelling of chemical reactions” occurring simultaneously [20]. These components and their interactions – also responding to the temporally and spatially changing environment – frame dynamic and complex systems at multiple scales to orchestrate plant development and behaviour. As a full understanding of system properties emerging from component interactions cannot be achieved only by “drawing diagrams of their interconnections” [17], computational techniques become indispensable for processing massive datasets and simulating complex mechanisms [21].
Although computational approaches have been progressing rapidly for modelling plant signalling, such as for signal transport [22],[23], canalization [24] and signalling network [25], most efforts have focused on cellular or tissue levels. Since AON is in essence a long-distance inter-organ regulatory network, our investigation required modelling at the whole-plant scale. Functional-structural plant models [26], such as those developed for resource allocation [27]–[29] and shoot signalling [30]–[34], can take inter-organ communication into account and use plant architecture as a direct reporter of underlying processes. Functional-structural modelling allowed us to simulate the hypothesised AON signalling and integrate it with nodulation. Yet the major difficulty was not how to model the hypotheses but how to test them through modelling. To meet this challenge, we have developed a new approach – Computational Complementation – for AON study.
Following description of the computational complementation method, we will present its first application in investigating whether wild-type cotyledons participate as an SDI producer in the AON system. Previous studies have indicated that mRNA for GmNARK, which, if translated, is responsible for perceiving the Q signal and triggering the SDI signal, exists in wild-type unifoliate and trifoliate leaves. It is expressed in all vascular tissue [8] of the plant (including the root), but its product is functional only as a nodulation control receptor in the leaf [35]. Thus the RNA expression pattern does not match biological function in AON. Relevant to the investigation here, the vasculature of the cotyledon also expresses RNA for GmNARK; whether this is functional in AON signalling was unclear. Therefore we used computational complementation to test two opposing hypotheses: (a) cotyledons function as part of the root, incapable of perceiving Q and producing SDI; or (b) cotyledons function as part of the shoot, involved in regulating root nodules.
Genetic complementation [36] is a classical approach to define genetic cause-and-effect relations. For example, assuming two mutant organisms exhibit the same phenotype caused by loss-of-function (recessive) mutations, then their hybrid will be wild-type, if the mutations are in different genes (called cistrons); conversely the hybrid will be mutant if the mutations are in the same cistron. In other words, the wild-type (functional) allele complements the deficiencies of the mutant. Genetic complementation is also used in transgenic analysis of organisms, as a loss-of function mutation in a candidate wild-type gene is deemed causal for a mutant phenotype if that mutant is effectively complemented by the transfer of a dominant wild-type allele. The complementation approach introduced here does not cross one genotype with another, but will use computational modelling to complement the deficiency (in an empirical model) of a mutant to determine if this recovers the virtual wild-type phenotype.
We use two well-characterized soybean (Glycine max L. Merrill) genotypes: the wild-type soybean Bragg and its loss-of-function mutant nts1116 [37]. Wild-type soybean Bragg performs AON to keep its nodulation balance well-maintained (Fig. 1A and C), leading to characteristic crown nodulation in upper root portions. In its near-isogenic mutant nts1116, the Q signal generated from early nodule proliferation cannot induce SDI due to the lack of GmNARK activity in leaves (Fig. 1B). Reduced SDI in GmNARK-deficient plants leads to a phenotype with many more nodules than wild-type, called “supernodulation” or “hypernodulation” (Fig. 1D) [5]. Compared with Bragg, the only deficiency of nts1116 plants is the significantly reduced capacity of producing SDI.
The key idea of our complementation approach comes from this point. We “add” hypothetical components of AON signalling, including those of signal production, transport, perception and function (see also Text S3), into the empirical model that depicts the growth behaviours of nts1116 plants to see if a wild-type phenotype can be restored. The flowchart of methodology for this approach is given in Fig. 2, including the following steps:
The architectural and functional-structural models mentioned in steps (i) and (ii) have been built with context-sensitive L-systems [31]. The empirical data used for building architectural models of Bragg and nts1116 plants were collected every second day from growth experiment under the same conditions until the 16th day post-sowing (all plants were inoculated on the 2nd day). Materials and methods for this glasshouse experiment are given in supporting Text S1. The growth data, algorithms and techniques used for model construction are described in supporting Text S2. The remaining steps of the flowchart, including (iii), (iv), (v) and (vi), are implemented for hypotheses testing and prediction.
In this initial application of our computational complementation approach, two opposing hypotheses were tested: (a) cotyledons function as part of the root, incapable of perceiving Q and producing SDI (abbreviated as “cotyledon-root” hypothesis); (b) cotyledons function as part of the shoot, involved in regulating root nodules (abbreviated as “cotyledon-shoot” hypothesis). Since GmNARK is expressed in all organs [8] (including cotyledons) and since cotyledons are short-term terminal organs (as they are degraded 7–14 days after germination), neither the cotyledon-root nor the cotyledon-shoot hypothesis was favoured a priori.
Theoretically speaking, if all other AON mechanisms (such as signal production, transport, perception and function) had been confirmed and used as basis for this application, the tested hypothesis leading to a wild-type nodulation pattern could be the correct one. However, the actions of many other signalling components also remain unclear. One or two virtual experiments are obviously insufficient to allow conclusions. Implementing too many experiments (to test all mechanisms together), however, would miss the emphasis and undermine efficiency. With these concerns, our strategy was to adjust parameters for signal production, transport, perception and function within a limited range, and use them as different conditions for different virtual experiments. Among all these experiments, if the complementation results (nts1116+AON) based on the cotyledon-root hypothesis are always or in most cases closer to Bragg than those based on the cotyledon-shoot hypothesis, then the cotyledon-root hypothesis would be considered plausible; otherwise, the cotyledons are more likely to function as general-sense leaves to regulate root nodulation.
According to this specific strategy, 27 virtual experiments (varying three rates of transport for both Q and SDI and three levels of nodulation inhibitory threshold) were designed for each of the two hypotheses: CRH_1∼CRH_27 for cotyledon-root testing and CSH_1∼CSH_27 for cotyledon-shoot testing. The only difference between CRH_i and CSH_j, if i = j, is whether cotyledons can function for AON signalling or not. Details of the virtual-experiment assumptions and conditions are described in the supporting Text S3.
To quantify the comparison between complementation results and Bragg phenotype, we define their similarity degree Scp as(1)where Nnt, Nbr and Ncp are the nodule numbers generated respectively by the architectural model of nts1116 plants, the architectural model of Bragg, and the functional-structural model of nts1116+AON. This can be understood as the ratio of the number of nodules inhibited by the virtual experiment to the number of nodules inhibited by a real Bragg plant. The similarity degrees of overall nodule number produced by virtual experiments on the 10th and the 16th day after sowing are listed in Fig. 3 and Fig. 4, where Rq and Rsdi represent the transport rates of Q and SDI signals (mm/day). These data indicated that the similarity degrees resulting from cotyledon-shoot hypothesis were generally much higher than those from cotyledon-root hypothesis, supporting the former hypothesis. Considering that values of Scp greater than 100% may mean over-regulation and might not be optimal, the criterion for further evaluating Scp is defined in Fig. 5. According to this criterion, the virtual experiments based on cotyledon-root hypothesis produced unsatisfactory results on the 10th day (Fig. 3, left-hand column), in sharp contrast to the cotyledon-shoot experiments (Fig. 3, right-hand column). Although there were good results derived from virtual experiments CRH_1, CRH_2, CRH_11 and CRH_13 on the 16th day (Fig. 4, left-hand column) in terms of nodule number, the nodule size and density from these experiments were all far from similar with the Bragg pattern (Fig. 6). In comparison, the nodule distribution generated by CSH_1 (Fig. 6D) – the opposite of CRH_1 – was quite close to that of the Bragg architectural model.
We predicted from these complementation experiments that the cotyledons should be part of the shoot and participate as an SDI producer in wild-type soybean plants.
To confirm the above prediction and also to evaluate the effectiveness of this approach, a “real-plant” grafting experiment was conducted. The critical experiment was to graft – between Bragg and nts1116 plants – the shoot of one genotype with cotyledons to the root of the other genotype without cotyledons, and also to graft the shoot of one genotype without cotyledons to the root of the other genotype with cotyledons, forming four graft combinations: Ns+Nc+Br, Ns+Bc+Br, Bs+Bc+Nr and Bs+Nc+Nr (Table 1). Materials and methods for this graft experiment are given in the supporting Text S1. The collected empirical data for nodule number were not only classified by each graft type but also according to each plant's cotyledon retention status (Table 2).
According to the experimental results, the nodule number from the Ns+Nc+Br graft type was much higher than that from the Ns+Bc+Br (Fig. 7A). For the Ns+Bc+Br graft type alone, its plants with fallen cotyledons had more nodules than those with persisting cotyledons, and the plants with yellow cotyledons had more nodules than those with green cotyledons (Fig. 7C). These differences suggest Bragg cotyledons were the only leaves to regulate nodulation in Ns+Bc+Br plants, because unifoliate and trifoliate leaves of nts1116 plants were unable to do so.
Data of another graft type with Bragg cotyledons – the Bs+Bc+Nr (Fig. 7D) also suggested that the Bragg cotyledons participated in providing SDI. However, more nodules were found in the Bs+Bc+Nr plants than in the Bs+Nc+Nr plants that had no Bragg cotyledons (Fig. 7B). An explanation for this observation is that the Bs+Nc+Nr allowed more nodules to be formed at early stages than the Bs+Bc+Nr, leading to more Q signal moving from root to shoot. As the cotyledon biomass declined greatly at later stages of seedling growth (resources are unloaded for plant growth and the “spent” cotyledon is eventually discarded), the difference in shoot between Bs+Bc+Nr and Bs+Nc+Nr became insignificant. Therefore larger amounts of Q triggered more SDI, which finally inhibited more nodules in Bs+Nc+Nr.
To better understand this nonlinear characteristic brought out by real-plant experiments, we returned to the virtual-experiment models and visualised the dynamic signal allocation during CRH_1 and CSH_1 (Fig. 8). As demonstrated by the visualisation, the SDI concentration (in the root) of CRH_1 was lower than that of CSH_1 on the 5th day but became higher from the 10th day on, in agreement with the above analysis of the nodulation difference between Bs+Bc+Nr and Bs+Nc+Nr. Thus, we conclude that the testing result from our initial application of computational complementation is confirmed: the cotyledons “belong” to the shoot and function as a source of the nodulation regulator in wild-type soybeans.
The computational complementation approach introduced here is an original contribution to the study of legume autoregulation of nodulation. Compared with conventional biological technologies with broader implications to plant development, one of the major advantages of this approach is its capability to complement the deficiency of a mutant plant at an organ scale with totally hypothetical and concept-derived physiological components. It is also able to make hypothetical signalling details manipulable and visible. For example, as demonstrated in the above case, signal transport rates can be modified as hypothesised and the allocation of signal can be dynamically visualised. These functionalities not only enable AON researchers to test hypotheses or make predictions using time- and resource-saving virtual experiments, but also bring out possible underlying details that are unobservable through real-plant experiments. Moreover, the application of this approach is not only limited to AON research, but also potential to other plant signalling studies such as those on branching regulation (e.g., [38]), flowering control (e.g., [39]) and lateral root initiation (e.g., [40]).
This approach contributes a new idea to the domain of computational plant modelling – computational complementation. From a classic modelling point of view, one can formulate a model based on empirical data and then verify the model against the data, which has been used for development of crop (e.g., [41]) and architectural (e.g., [42]) models. However, what we investigate is a largely unclear internal signalling system – most of the detailed mechanisms remain unknown, which determines there is no direct parameterisation-and-verification data to evaluate the modelled signalling hypotheses. Using an indirect strategy, functional-structural modelling allows us to use the observable structure as a reporter for estimation of the unobservable function. But for this study, we have to link the structure of one genotype with the function of another genotype. The reason for this is: the wild-type Bragg nodulation has already been regulated, thus incorporating AON to Bragg architecture would double the regulation and have no reasonable comparison target for validation; in contrast, the nts1116 is a non-AON plant and this is its only difference with Bragg, therefore activating AON in nts1116 plant could result in system behaviours comparable with the wild type.
Another feature of this approach resides in the level of complexity for simulation of structural and signalling processes. We captured root details for studying shoot-root signalling rather than oversimplifying the root system. And the signalling pathways are constructed with sub-modules of which the size and number can be manipulated without limitation, which allows future modelling work to be extended to lower-scale mechanisms (such as tissue and cellular scale). We also created a synchronisation algorithm for coordination of multi-rate procedures to enhance the precision of signalling-development interactions. A description of these modelling techniques is given in the supporting Text S2.
The approach also has some limitations. For example, due to the nature of complementation, it can only be used for a single mutation at a time, though leaky mutants can be handled by parameter optimization. Another drawback is that it cannot distinguish between different mutations in the same pathway that result in the same phenotype in the first instance. In other words, if the hypothesised mechanisms used to complement the mutant are the same in both cases, and so is the phenotype of the two mutants, computational complementation cannot be used to say which gene component of the regulatory network has been mutated.
Our first application of this approach was to test whether wild-type soybean cotyledons are involved in production of SDI. Also but more importantly, we expected this application to evaluate whether the computational complementation idea is effective. The virtual-experiment results suggested the wild-type cotyledons can produce SDI, which was further confirmed by a graft experiment on real plants. This demonstrates the feasibility of computational complementation and shows its usefulness for future applications.
The next step is to apply this approach to support research for the identification of Q and SDI. Candidate signals, such as CLE peptide for Q [13],[14] and auxin for SDI [43], will be tested to see if they play the roles in AON as hypothesised. In addition, environmental factors, such as soil nitrogen status, that have effects on the process could also be tested with this approach. Furthermore, the finding that wild-type soybean cotyledons act as an SDI producer in AON opens the door for testing physiological transgenerational effects, such as altered nodulation patterns influenced by the Bradyrhizobium infection status of mother plant through presence of SDI in cotyledons.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.